# Eat. Shift. Or Die.: public book website This is the public website for Eat. Shift. Or Die. by Phil Gray. It is not the full book. Private manuscript material, subscriber records and visitor tool inputs are not included. Live & Die AI is a separate AI-generated satire publication; imagined events must not be presented as fact. For the whole domain, start at https://eatshiftdie.com/llms.txt. The newspaper updates independently; its complete text is https://eatshiftdie.com/live-and-die/llms-full.txt. # Eat. Shift. Or Die. by Phil Gray | AI, work and what comes next Canonical: https://eatshiftdie.com/ A new book by Phil Gray 22 September 2026 # AI changed the rules. Your move. What AI changes. How organisations adapt, or don’t. What comes next for you. Kindle pre-orders are open now. [Pre-order on Kindle ↗](https://www.amazon.com/dp/B0HJ5XSQC7)[Read an extract ↗](https://eatshiftdie.com/extract.html) Amazon.com · [Australian Kindle store ↗](https://www.amazon.com.au/dp/B0HJ5XSQC7) [Prefer paperback? Get a release reminder ↗](https://eatshiftdie.com/preorder.html#paperback) Phil GrayAuthor. Practitioner. Still figuring things out. [Illustration: Eat. Shift. Or Die. by Phil Gray. Deep aubergine cover, ivory and orange type, with cutouts revealing orange, parchment and white layers.] THE NEW BOOK22.09.26 Most AI books get written for one of two shelves. Up the top, the C-suite books with the obligatory instruction to move faster. Down the bottom, How to Write a Better Prompt and How to Make Money with AI by Doing Bugger All. This one lives in the middle, where the consequences land. It's for the people who carry the responsibility long before anyone hands them the authority, and for the business owners who have all the authority and no buffer. It's about what AI does to the places people work, why a project with an end date fails a technology that never finishes, and what to do about your own work while some of the choices are still yours. eat /iːt/ verb: to take in what the world is actually telling you, especially when it is inconvenient, unfamiliar or badly timed. shift /ʃɪft/ verb: to change position, practice or identity before the old one fails completely. or die /ɔː daɪ/ consequence: for an organisation, profession or way of working, to lose relevance, agency or viability while still insisting nothing fundamental has changed. It is a sequence, and the order matters. Inside Eat. Shift. Or Die. ## The technology is only part of the story. [Read an extract ↗](https://eatshiftdie.com/extract.html) I ### What AI changes What happens to expertise, work and value when a capable answer becomes much easier to produce? II ### How organisations adapt, or don’t. The structures, incentives and habits that decide whether new capability becomes useful work. III ### What comes next for you How to build your own understanding, practise your judgement and make a useful next move. From Chapter 13 / Learn to Work With the Machine “The choice isn't a personality type. It changes with ambiguity, consequence and verifiability.” [Read the extract ↗](https://eatshiftdie.com/extract.html) When to stay close to the machine, when to hand the work over, and why the distinction matters. The book trailer / 03:28 ## What has to reorganise around you? An answer arrives. It’s good. You’re still deciding how you feel about that. This book is for that feeling. [Download the film](https://eatshiftdie.com/video/teaser-v8/eat-shift-or-die-trailer-v8-1080p.mp4) 03:28 · The book trailer[Transcript & visual description ↗](https://eatshiftdie.com/transcript.html)[Smaller video · 5 MB ↗](https://eatshiftdie.com/video/teaser-v8/eat-shift-or-die-trailer-v8-480p.mp4) Who are you when the answers are free? Eat. Shift. Or Die. Chapter 16 Beyond the book ## The book ends. The experiments don't. [Explore beyond the book ↗](https://eatshiftdie.com/beyond.html) [[Illustration: A paper racehorse escapes a computer and meets the real world.] ### The experiments What I tried, what failed and what I learned.](https://eatshiftdie.com/experiments.html)[[Illustration: An apprentice and mentor build the foundations together.] ### Try it yourself Small exercises you can run in your browser.](https://eatshiftdie.com/tools.html)[[Illustration: A bookseller encounters an unexpectedly large order.] ### The monthly sweep New stories that put the book’s ideas to work.](https://eatshiftdie.com/sweep/) About the author ## Meet Phil Gray. Phil Gray met his first chatbot on a cassette tape in 1979. Businesses built through the dot-com boom and out the other side, innovation labs around the world, decades close enough to enterprise change to hear the gears grind. He has been the disruptor and the disrupted. He lives in Sydney. [Meet Phil ↗](https://eatshiftdie.com/about.html) Eat. Shift. Or Die. / 22 September 2026 Your organisation might not survive this transition. Your current role might not either. You are not guaranteed a clean landing. But you can learn to move before the ground comes screaming up to meet you. From Chapter 17 [Pre-order on Kindle ↗](https://www.amazon.com/dp/B0HJ5XSQC7)[Read an extract ↗](https://eatshiftdie.com/extract.html) --- # The experiments | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/experiments.html Out in the field # The price of admission. Real experiments, real failures, and what made it into Monday. The bets never paid. The lessons did. [[Illustration: A racehorse and jockey burst from a computer monitor in an orange, purple and black illustration.] New visual story · February–August 2026 ## Confidently wrong. A man, a racing model and several very clever machines. The failures, false dawns and lessons that made it into Monday. Follow the whole experiment ↗](https://eatshiftdie.com/racing-story.html) [Experiment 01 I wrote it. They argued with me. Writing & judgement In practice ## Several AIs. One accountable author. Research assistants, editors and a sceptical reader. The experiment inside the book. Read the field note ↗](https://eatshiftdie.com/experiment-book.html)[Experiment 02 195 files. 170 videos. Now what? Research & evidence Method in use ## 195 transcripts. Which ideas deserve to survive? A large research folder becomes a test of selection, scepticism and source discipline. Read the field note ↗](https://eatshiftdie.com/experiment-research.html)[Experiment 03 Written. Spoken. Checked. Making & checking Preview available ## The page sounded different out loud. Narration makes pauses, emphasis and missing words impossible to ignore. Read the field note ↗](https://eatshiftdie.com/experiment-audio.html) The confident answer is becoming cheap. Accountable judgement is not. Eat. Shift. Or Die. Chapter 10 Your turn ## Start with something small enough to finish. [Plan an experiment ↗](https://eatshiftdie.com/tool-experiment.html) --- # Several AIs. One accountable author. | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/experiment-book.html [← All experiments](https://eatshiftdie.com/experiments.html)Field note 01 Writing & judgement · 6 min read # Several AIs. One accountable author. Making Eat. Shift. Or Die. became a practical test of what happens when machines can research, organise, challenge and edit, while a person remains responsible for the result. In practiceFrom Phil Gray’s book project The question Could several AI systems improve a book without taking over the argument? The useful bit Separate the roles. Keep the accountability. [Illustration: An illustrated author selects one orange page from three streams of drafts and places it in an open book.] The author still chooses what makes it onto the page. ## The work The book used several AI systems in distinct roles: research and fact-checking, project management, editorial suggestions and a deliberately sceptical reader. Those roles had different jobs. Finding material did not grant permission to invent evidence. Improving a sentence did not settle whether its argument belonged in the book. ## Where it became difficult The manuscript describes a plausible anecdote about an innovation lab that Kev produced early in the process. It had not happened. Phil removed it, and the failure led to a clear rule: material must come from something that happened or from a source that can be checked. ## What the division of labour made possible Different systems could challenge an argument from different positions. They could also hold a long manuscript in view, notice contradictions between chapters and identify repetition. Phil accepted some suggestions and rejected others. Those decisions remained part of writing. ## What to take into your own work Give the research, drafting and challenge passes distinct purposes. Name the source that controls a factual claim. Decide what a reviewer must verify before the work can be accepted. Keep the final decision with someone who understands why the work matters. The tools did not remove judgement. They increased the number of moments at which judgement mattered.Phil Gray Source note Adapted from Chapter 13, “Learn to Work With the Machine”, in the working manuscript. Author-led work with several AI systems. No comparative time or cost claim is made. Take it into your own work ## Build a better brief [Have a go ↗](https://eatshiftdie.com/tool-brief.html) --- # 195 transcripts. Which ideas deserve to survive? | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/experiment-research.html [← All experiments](https://eatshiftdie.com/experiments.html)Field note 02 Research & evidence · 5 min read # 195 transcripts. Which ideas deserve to survive? The project’s research inventory contains 195 transcript files representing 170 unique YouTube URLs. The useful job was to find ideas, compare arguments and preserve a way back to their sources. Method in useFrom Phil Gray’s book project The question Can AI help make sense of a research collection without turning repetition into evidence? The useful bit Make the trail back to the evidence part of the output. ## Start with the actual material Repeated captures of the same video were treated as one source. The research was organised around three questions: what is interesting about AI, what it means for organisations and what it means for individuals. That gave the work boundaries before it began. ## A source is not automatically proof The collection mixes interviews, demonstrations, commentary and forecasts. A claim appearing in a transcript means the speaker made it. It does not establish that the claim is true. Figures, quotations and company claims need checking against their original evidence before publication. ## The useful handover The research inventory connects its themes to source videos and makes its verification limits explicit. That lets the author revisit an argument rather than inherit an unattributed summary. The judgement lies in deciding what to carry forward. ## Try a smaller version Choose three sources about a question you already understand. Ask an AI to separate observed findings, interpretation and prediction. Then open the originals and inspect the distinctions yourself. Record one useful connection and one claim that needs more work. A mention below means that a point was present in the folder corpus; it does not mean the claim has been independently verified.Research inventory Source note Project research inventory, “YouTube transcript AI interest mine”. Counts describe that saved corpus, not a continuously updated feed. Documented corpus review. The linked reading collections are a separate, selected set of primary sources. Take it into your own work ## Plan your own small experiment [Have a go ↗](https://eatshiftdie.com/tool-experiment.html) --- # The page sounded different out loud. | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/experiment-audio.html [← All experiments](https://eatshiftdie.com/experiments.html)Field note 03 Making & checking · 4 min read # The page sounded different out loud. Turning the book into audio introduced a different kind of quality question. A sentence can be intact on the page and still need work when it is spoken. Preview availableFrom Phil Gray’s book project The question What changes when a manuscript becomes a performance? The useful bit Check what the audience receives. ## A new medium creates a new test The project includes narrated book previews and a longer introduction recording. The audio workflow handles pacing, pauses, names and pronunciations as explicit production choices. It also compares the recording against its source text. ## What the checks found The introduction production notes record long paragraphs that omitted words during generation. Five paragraphs were rebuilt from smaller parts and checked again before assembly. A smooth performance was not enough to establish completeness. ## Two kinds of acceptance Technical checks can confirm format, transcript coverage, loudness and pauses. Listening still decides whether the performance works. The project notes explicitly retain Phil’s listening review as the final judgement on performance taste. ## What travels beyond audio Whenever work changes format, define quality again. A complete document is not necessarily a useful presentation. A correct transcript is not necessarily a good performance. Check the thing people will actually receive. Existing public preview · 02:54 Your browser does not support audio. [Download the MP3.](https://eatshiftdie.com/audio/eat-shift-or-die-three-excerpt-teaser.mp3) Original pre-launch recording; its closing announcement reflects the earlier launch plan. [Read this recording’s transcript ↗](https://eatshiftdie.com/audio/eat-shift-or-die-three-excerpt-teaser-transcript.txt) Source note Audiobook production notes and the public three-excerpt preview. The longer introduction remains outside this site. The linked sample is the existing public preview, credited to Phil Gray and Peter Baker. Its original launch announcement is retained and labelled. Take it into your own work ## Listen to the book preview [Have a go ↗](https://eatshiftdie.com/media.html) --- # The reading room | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/sources.html Sources, arguments & useful questions # Follow the thinking. A small, annotated collection. Start with a question, follow the original material and make up your own mind. Collection 01 ## Good at the task. Useful in the work? Studies that make a blanket claim about “AI productivity” harder to sustain. Read them together. Field experiment2023 working paper ### [Navigating the Jagged Technological Frontier ↗](https://aiinstitute.hbs.edu/navigating-the-jagged-technological-frontier/) Fabrizio Dell’Acqua and colleagues Harvard Business School AI assistance improved performance on some consulting tasks and worsened it on another task beyond the tested capability frontier. Why it belongs here A useful reason to test the actual work, rather than assume that success on a neighbouring task will carry across. Read with care. A specific model, group of participants and task set. This is not a current ranking of AI tools. Which part of your workflow could look similar to a successful demonstration while demanding a different kind of judgement? [Open the original ↗](https://aiinstitute.hbs.edu/navigating-the-jagged-technological-frontier/) Workplace study2023 · revised 2024 ### [Generative AI at Work ↗](https://arxiv.org/abs/2304.11771) Erik Brynjolfsson, Danielle Li and Lindsey Raymond Author manuscript · arXiv The introduction of an AI assistant in customer support was associated with productivity gains that differed across workers, with larger benefits for less experienced workers. Why it belongs here A concrete setting for thinking about how assistance can change access to expertise, rather than simply replace it. Read with care. Results from a particular workplace and task. A productivity effect there does not establish the effect in your organisation. Whose knowledge becomes easier to use, and how will the next generation learn to develop it? [Open the original ↗](https://arxiv.org/abs/2304.11771) Field reportJune 2025 ### [Microsoft 365 Copilot Experiment ↗](https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html) Checked 8 September 2026 Government Digital Service Cross-Government Findings Report Twenty thousand UK government employees, twelve organisations, three months. Read it alongside the Australian APS trial and compare where people found value. Why it belongs here A large public-sector trial offers a view of everyday use across different kinds of work. Read with care. Self-reported time savings, one product family, tools tested in late 2024. Participation is not the same as a measured productivity gain. What would this experiment look like at your organisation’s size, and who would read the result? [Open the original ↗](https://www.gov.uk/government/publications/microsoft-365-copilot-experiment-cross-government-findings-report/microsoft-365-copilot-experiment-cross-government-findings-report-html) [Put the question to workWhere did the time go? ↗](https://eatshiftdie.com/tool-workflow.html) Collection 02 ## Who is doing the thinking? A closer look at the effort that moves around the answer, and the limits of what the evidence can tell us. Survey2025 ### [The Impact of Generative AI on Critical Thinking ↗](https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/) Hao-Ping Lee and colleagues Microsoft Research · CHI 2025 A survey of knowledge workers examined how confidence and AI use relate to self-reported critical thinking effort. Why it belongs here It asks us to examine the human effort around checking, integrating and accepting an answer. Read with care. Self-reported survey evidence. It does not prove that AI causes a lasting loss of critical thinking ability. When an answer arrives looking finished, what makes you decide to inspect it? [Open the original ↗](https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/) Survey2026 ### [The Work AI Index 2026 ↗](https://www.glean.com/work-ai-institute/reports/work-ai-index) Checked 8 September 2026 Glean Work AI Institute and university co-authors Global report Six thousand digital workers across the US, UK and Australia, and a useful name for the work behind the output. Botsitting is the effort of making AI usable. Why it belongs here The checking, context and repair behind an output often disappear from the productivity story. Read with care. Vendor-published, with university co-authors. Self-reported experience from digital workers, not a productivity audit of the whole workforce. Who in your team is botsitting, and does anyone’s dashboard know? [Open the original ↗](https://www.glean.com/work-ai-institute/reports/work-ai-index) [Put the question to workWho gets to decide? ↗](https://eatshiftdie.com/tool-delegation.html) Collection 03 ## A machine answers back. A historical starting point, two books and practical guidance for what happens next. Original paper1966 ### [ELIZA ↗](https://doi.org/10.1145/365153.365168) Joseph Weizenbaum Communications of the ACM Weizenbaum described a program that used decomposition and reassembly rules to conduct text exchanges in natural language. Why it belongs here The historical starting point for the book’s question about what happens when a machine appears to answer back. Read with care. A historical language-processing program. The exercise here is inspired by simple pattern matching and is not the original implementation. What do you bring to a conversation that the program itself does not know? [Open the original ↗](https://doi.org/10.1145/365153.365168) Design research2019 ### [Guidelines for Human-AI Interaction ↗](https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/) Saleema Amershi and colleagues Microsoft Research · CHI 2019 Eighteen guidelines address how AI systems should behave initially, during interaction, when wrong and over time. Why it belongs here A practical way to think about the relationship around the model: setting expectations, supporting correction and adapting carefully. Read with care. Design guidance is a starting point. It does not remove the need to test a particular system with the people using it. What should a useful AI colleague do when it is wrong? [Open the original ↗](https://www.microsoft.com/en-us/research/publication/guidelines-for-human-ai-interaction/) Book2024 ### [Co-Intelligence ↗](https://www.penguinrandomhouse.com/books/741805/co-intelligence-by-ethan-mollick/) Checked 8 September 2026 Ethan Mollick Portfolio A practical account of working alongside AI, from one of the researchers behind the jagged frontier. Why it belongs here A companion for moving from impressive demonstrations to deliberate practice. Read with care. A fast-moving field. Treat specific tool advice as dated, the frames as durable. Which of your tasks sits just beyond the frontier, and how would you find out? [Open the original ↗](https://www.penguinrandomhouse.com/books/741805/co-intelligence-by-ethan-mollick/) Book2006 · second edition 2016 ### [The Box ↗](https://marclevinson.net/the-box/) Checked 8 September 2026 Marc Levinson Princeton University Press The container story the book retells. A simple box needed ports, standards and working practices to change before it could transform trade. Why it belongs here It puts the work around a new capability at the centre of the story. Read with care. Economic history. The parallel to AI is the reader’s to test. What is your industry’s equivalent of a port built for the old way of moving things? [Open the original ↗](https://marclevinson.net/the-box/) [Put the question to workTry the ELIZA-inspired exercise ↗](https://eatshiftdie.com/eliza.html) Collection updated 8 September 2026. The four additions are individually dated. Earlier selections were checked on 5 September 2026. These annotations connect sources to the site’s questions; they are not a complete bibliography of the book. Access conditions may vary by publisher. --- # The assumption graveyard | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/graveyard.html Recently deceased assumptions # Some ideas outlive their use. A few management assumptions worth revisiting. Keep what still helps. Retire what gets in the way. [Illustration: A woman lifts an orange finish-line ribbon while the path continues far beyond it.] Assumption 01R.I.P. ## The job description as a fixed object. Served faithfully. Outlasted several of its own tasks. Read the obituary A role description helps people understand responsibility. It becomes less useful when the work changes faster than the document. ### What deserves to survive Keep clarity about purpose, authority and accountability. ### Try this List the tasks that changed in the last six months. Then ask whether the role’s boundaries still make sense. Assumption 02R.I.P. ## Transformation with an end date. Expected to finish in June. Nobody specified which June. Read the obituary A programme can finish. The need to respond to changing capability, customers and conditions continues. ### What deserves to survive Keep deadlines for concrete commitments and reviews. ### Try this Separate the things you can complete from the capabilities you will need to keep exercising. Assumption 03R.I.P. ## Training as the answer to every capability problem. Completed every module. Still could not get permission. Read the obituary Training cannot by itself fix missing authority, inaccessible information or a process that blocks the work. ### What deserves to survive Keep deliberate practice and learning that responds to real gaps. ### Try this Before booking the course, identify whether the obstacle is knowledge, information, permission or workflow. Assumption 04R.I.P. ## The organisation chart as a map of how work happens. All the boxes were correct. The work went somewhere else. Read the obituary Reporting lines explain one kind of authority. They often leave out the handovers, informal expertise and exceptions on which delivery depends. ### What deserves to survive Keep visible accountability. Add the connections people actually rely on. ### Try this Follow one piece of work across the organisation and draw every handover. Assumption 05R.I.P. ## A good answer means the job is done. Produced at 9:42. Understood sometime after lunch. Read the obituary Output still has to be checked, accepted, integrated and used. Faster production can move effort onto someone else. ### What deserves to survive Keep a clear definition of what an acceptable outcome looks like. ### Try this Measure the effort from the first request to a result someone can use. Assumption 06R.I.P. ## The human in the loop as a safety guarantee. Present at every incident. Responsible for none of them. Read the obituary A comforting phrase that names a box on a process map. It rarely says which human, making which decision, at what point, with what evidence and what authority. ### What deserves to survive Keep humans accountable for outcomes, with the time, understanding and authority to intervene. ### Try this Pick one AI-assisted process and answer the five questions above in writing. If any answer is “unclear”, so is the safety. Assumption 07R.I.P. ## Adoption as the finish line. Deployed to thousands. Changed almost nothing. Read the obituary Licences, logins and usage dashboards measure contact with a tool, not change in the work. Plenty of organisations have crossed this finish line and found nothing on the other side of it. ### What deserves to survive Keep measuring, but measure the work. Cycle time, rework, error, customer outcome. ### Try this Take your most-quoted adoption number and ask what changed in the work of the ten heaviest users. Answers may vary. Seen an assumption that deserves a plot here? Send the epitaph to [info@eatshiftdie.com](mailto:info@eatshiftdie.com?subject=An%20assumption%20for%20the%20graveyard). Best submissions get buried with full honours. Test the assumption ## Find out before you build around it. [Plan a small experiment ↗](https://eatshiftdie.com/tool-experiment.html) --- # Read two extracts | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/extract.html From Eat. Shift. Or Die. # Read the book. Two extracts. The way an organisation works, and the way you work with a machine. Chapter 13 · Learn to Work With the Machine ## Steer or dispatch. An extract from Eat. Shift. Or Die. by Phil Gray Some work needs steering, because the purpose is still forming. Good is difficult to describe, and the task contains taste, politics, trade-offs or a question you don't yet understand. So you stay close. You talk, inspect, redirect and discover the problem alongside the system. Writing a chapter like this sits in that category, and so does deciding what a weak signal means for your organisation, or whether a technically good answer is the wrong answer for the person who will receive it. Other work can be dispatched. The outcome has edges, the sources are known, and the process is stable enough to describe. The result can be tested, reconciled or checked against evidence, so you assign the job, let the system work and inspect what returns. Searching a large transcript collection for every discussion of apprenticeship can be dispatched. Deciding what apprenticeship should mean in this book cannot. The mistake is to dispatch ambiguity and steer routine forever. If you send away work whose purpose and quality bar remain unclear, the system will make decisions you didn't realise you had delegated. If you stay in a microscopic conversation after the job has become repeatable, you turn leverage back into supervision. The choice isn't a personality type. It changes with ambiguity, consequence and verifiability. Stay close when meaning is moving, step back when the work can show its paper trail. I once paid a hundred real dollars for a live demonstration of the first mistake, in the middle of a three-year racing experiment I'll tell you about properly in a moment. In a fit of frustration with my own system, I told Kev247 to come up with its own bloody model. I gave it a real hundred dollars, one goal, double it by the end of the day, and a promise that I wouldn't check on it until then. Free rein. It could have made hundreds of small bets and ground out incremental gains. Instead it put fifty dollars on the favourite in each of the first two races. Both lost. The hundred was gone before I'd kept my promise for an hour, and there was not a hint of an apology, because there was nothing to apologise for. It had pursued the goal I set with the ambiguity I supplied. No quality bar, no boundaries, no definition of acceptable risk, and so the system made every one of those decisions itself, which is to say I made them by refusing to. As a betting experiment it was a dud. As a delegation experiment it was the cheapest hundred dollars I have ever spent. [Try the delegation exercise ↗](https://eatshiftdie.com/tool-delegation.html)[Read Les and the shredder ↓](https://eatshiftdie.com/extract.html#les-and-the-shredder) Chapter 5 / The organisation ## Les and the shredder. An extract from Eat. Shift. Or Die. by Phil Gray Years ago, inside one of the banks, I worked on a change that should have been simple. Every month, head office sent every branch a printed report listing the term deposits maturing that month. Branch managers used it to work the phones, ringing customers to encourage a rollover before the money walked out the door. The report was effectively a phone book. A flat printout of every maturing deposit, dumped on the branch with no way to prioritise beyond starting at the top and reading down. The new system fixed that properly, and here's the part I only recently clocked as an early chapter of this book's story. Underneath it, we were running a new algorithm through the term deposit book, predicting each customer's likelihood of ringing up to ask for a better rate, versus doing nothing and letting the deposit roll over onto a shockingly bad one. Branch managers could slice the same data themselves, sort by deposit size, by relationship, by predicted rate-sensitivity, by the things that actually mattered when deciding who to call first. And, just as usefully and rather less printably, who not to call, because a deposit rolling over untouched onto a terrible rate was profitable in a way nobody needed explained. Machine prediction steering front-line phone calls. We didn't call it AI. It was. Better tool, better logic, a real improvement. Utilisation after go-live looked healthy. Adoption, by every measure on the dashboard, was going fine. The problem was that the old report refused to die, because we could not find where it lived. The bank's systems were so tangled that nobody could trace which system generated it or which job fed the printing centre that mailed it out. Turning it off was not a five-minute change request. It was an archaeology project nobody had time to finish before go-live. So we improvised. We found a bloke in the mailroom named Les and asked him to intercept the old reports as they came off the printers and put them straight into the shredder. Les did this faithfully for months. Credit to Les. For as long as he stood between the printers and the post, the new way of working held. Then the project closed and moved on to the next thing, as projects do, and at some point the shredding quietly stopped being anyone's job. The old paper reports started reaching branches again. The first sign anything was wrong was not a complaint or an incident. It was a slow decline in utilisation of the new system, as branch managers drifted back to the phone book. It had never been taken away. It had only been intercepted, one month at a time, by a man with a shredder. Go-live succeeded, and the dashboards stayed green for as long as the scaffolding stood, and the scaffolding was Les. The old default had never been decommissioned, only hidden, and organisations revert to whatever remains possible. Adoption is what people do while somebody is watching the old path. Adaptation is what is left when nobody is. ### Where would you draw the line? [Get the book ↗](https://eatshiftdie.com/preorder.html)[Try the adaptation exercise ↗](https://eatshiftdie.com/tool-adaptation.html) --- # About Phil | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/about.html The person behind the book # Phil Gray. Still curious. Author, practitioner and someone who has been both the disruptor and the disrupted. [Illustration: Cinematic reconstruction of two children encountering a TRS-80 in a late-1970s home office.] Cinematic reconstruction, not an archival photograph. ## A machine answered back. Phil’s book begins with ELIZA on a TRS-80 in a Sydney home office in 1979. A nine-year-old typed, the machine answered back, and the question that conversation started has followed him for forty-seven years. The years in between were spent unusually close to the action. Paid to test computer games after school while his mates did paper runs. Teaching Lotus 1-2-3 to sales teams who would rather have been at the pub, which turned out to be a first-row seat for the last technology that seeped into every office. Building businesses, including one that put computer classrooms on wheels. Running training inside a bank, then innovation labs inside global banks across Asia, where he helped build bathroom mirrors that showed people their money, put shipping assets the bank had financed into augmented reality, and learned the difference between innovation and innovation theatre by staging a fair amount of both. ## Both sides of the collision. Eat. Shift. Or Die. follows the question underneath all of it. What happens when new capability meets the way an organisation actually works? Phil has asked it as the disruptor arriving with the demo, as the leader whose own expertise was disrupted, and as the consultant brought in to referee. The book is the answer he could stand behind, tested against public evidence and written with the technology it describes, which produced arguments he did not always win. His parents came from the Black Country. Its industrial roots run through the book, from the Tolpuddle Martyrs to the machines now arriving at your desk. ## Off duty. Sydney. A Tesla with strong opinions, heavy metal recorded before 1987, an unreasonable number of AI subscriptions he has decided to call a research programme, and a houseful of machines with names. Kev and Bev get proper introductions in the book. [Get the book ↗](https://eatshiftdie.com/preorder.html)[Read an extract ↗](https://eatshiftdie.com/extract.html)[Phil on LinkedIn ↗](https://www.linkedin.com/in/philgray247/)[Phil on X / Twitter: @philgray ↗](https://x.com/philgray) --- # Try it yourself | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tools.html Small exercises. Useful questions. # Get your hands on it. You don't need a new system or a perfect plan. One question, one exercise, ten spare minutes. [01 ↗ ## Where did the time go? Speed up the AI. Watch the queues form somewhere else. Interactive workflow · 3 min](https://eatshiftdie.com/tool-workflow.html)[02 ↗ ## Give it the missing context. Build a brief that says what good means, and what needs to stay with you. Brief builder · 5 min](https://eatshiftdie.com/tool-brief.html)[03 ↗ ## One irritation. One experiment. Make a small plan with something to learn and a point at which to stop. Experiment planner · 5 min](https://eatshiftdie.com/tool-experiment.html)[04 ↗ ## Who gets to decide? Meet three decisions. Work out how much authority you would hand over. Decision exercise · 4 min](https://eatshiftdie.com/tool-delegation.html)[05 ↗ ## The adaptation stress test. Five signals to help you see where your organisation needs a closer look. Reflection · 3 min](https://eatshiftdie.com/tool-adaptation.html)[06 ↗ ## Meet ELIZA. A green-screen conversation with a simple pattern-matching program. Green-screen simulator · 2 min](https://eatshiftdie.com/eliza.html) These exercises help you reflect, compare and prepare. The workflow is an illustrative model; the other tools use fixed rules and your own inputs. None makes decisions about your organisation. AI doesn't eliminate bottlenecks. It moves them. Eat. Shift. Or Die. Chapter 6 --- # Where did the time go? | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tool-workflow.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 01 Try it yourself # Where did the time go? The drafting got faster. Did the work? Change the capacity at each step and see what reaches the end. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). An illustrative 8-hour day Drafting speed 4× One person drafts 4 items an hour without AI. Review capacity 6 / hr How many items can be checked properly? Decision capacity 8 / hr How many checked items can be approved and used? From request to something useful Review is the bottleneck 01 / Draft128items produced → 02 / Review48items checked → 03 / Use48items completed 80waiting for review 0waiting for a decision Baseline: 32 completed items using the same review and decision capacity, with drafting at 1×. What this model assumes This is a simple capacity model, not a forecast. It assumes an unlimited supply of identical tasks, eight working hours, immediate handovers, consistent quality and no rework or start-up delay. Each stage can process only what the previous stage supplies. Completed work is limited by the smallest capacity. Real work also involves interruptions, changing demand, varied complexity and mistakes. Use this model to ask where the work might accumulate, then observe your actual workflow. [Explore the evidence on AI and performance ↗](https://eatshiftdie.com/sources.html#capability) --- # Give it the missing context | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tool-brief.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 02 Try it yourself # Give it the missing context. A useful brief names the outcome, the evidence and the judgement that stays with you. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). Your working brief ## Make good explicit. Assembled from your words. Copy this into the AI tool you already use. The useful test: could a capable colleague use this to recognise a good result without guessing what matters to you? [See how roles and rules shaped the book ↗](https://eatshiftdie.com/experiment-book.html) --- # One irritation. One experiment. | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tool-experiment.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 03 Try it yourself # One irritation. One experiment. Choose something small enough to inspect. The point is to learn whether a different way of working helps. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). Your experiment card ## Small enough to learn from. --- # Who gets to decide? | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tool-delegation.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 04 Try it yourself # Who gets to decide? The task looks simple until an exception arrives. Choose a boundary, then see what it asks of you. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). Scenario 1 of 3 ## Where would you place the authority?Prepare itAI produces a draft. I check and decide.Act within rulesAI proceeds inside a defined boundary and escalates exceptions.Decide and actAI chooses the action and carries it out without individual review. Then this happens ### Your decisions, together ## Authority needs a boundary. Your answers stay on this page. Nothing is sent to an AI service or saved when you leave. --- # The adaptation stress test | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/tool-adaptation.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 05 Try it yourself # Look at the organisation. Five signals. A prompt to look harder, not a verdict on your organisation. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). Question 1 of 5 ## Choose the answer closest to your organisation A starting point for reflection ## A useful next step [Try the exercise ↗](https://eatshiftdie.com/tool-workflow.html) Revisit your answers An illustrative reflection exercise, not a validated organisational assessment. Answers stay on this page. --- # ELIZA · Green-screen simulator | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/eliza.html [← All tools](https://eatshiftdie.com/tools.html)Exercise 06 Try it yourself # Meet ELIZA. Type a sentence into the green screen. See how a few simple rules can keep a conversation going. Loading the exercise… This exercise needs JavaScript. You can still read the [experiment stories](https://eatshiftdie.com/experiments.html) and [original sources](https://eatshiftdie.com/sources.html). ELIZA / SIMULATED TERMINALRULE-BASED · LOCAL ELIZAHow do you do. Please tell me your problem. READY · Nothing stored or sent. Behind the conversation ## It responds. What do you read into it? This is a small demonstration inspired by ELIZA’s pattern matching, not the original program or a modern AI model. It recognises a handful of phrases, switches some pronouns and returns a question. Everything else gets a fallback response. The last rule used Waiting for your first message. Use an ordinary example sentence. This exercise is not a counselling service. [Read about the original ELIZA ↗](https://eatshiftdie.com/sources.html#relationship) --- # Watch & listen | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/media.html The conversation continues # Ask me the hard ones. Everything you need for an interview, a review or an event, including the questions most interviews never get to. The book trailer / 03:28 ## What has to reorganise around you? An answer arrives. It’s good. You’re still deciding how you feel about that. From ELIZA and Turing to the quiet changes at work, this is what the book is for. [Download the film](https://eatshiftdie.com/video/teaser-v8/eat-shift-or-die-trailer-v8-1080p.mp4) 03:28 · The book trailer[Transcript & visual description ↗](https://eatshiftdie.com/transcript.html)[Smaller video · 5 MB ↗](https://eatshiftdie.com/video/teaser-v8/eat-shift-or-die-trailer-v8-480p.mp4) Prefer to listen?[Download the trailer audio](https://eatshiftdie.com/video/teaser-v8/eat-shift-or-die-trailer-v8-audio.mp3) Read the visual description The film opens in a layered-paper office, where a young analyst shows an AI answer to an older colleague. Text reads, This book is for that feeling. The camera moves gently towards a paper cassette recorder before a boy at a computer introduces ELIZA in 1979. A paper portrait based on Phil Gray’s generated office photograph shows him writing in a notebook, then lifting his gaze with a slight smile. His face, silver hair, black glasses and shirt are built from visible paper layers. A paper illustration of Alan Turing is followed by pedestrians absorbed in their phones. Colleagues pass orange folders around a vacant desk as paper office routes change. A shipping container arrives and infrastructure unfolds around it. An older worker packs his belongings into a box and leaves his desk. All illustrations, buildings, clothing and props use textured cream, kraft and aubergine paper, with orange accents. Definitions build line by line beside parchment EAT, outlined SHIFT and orange OR DIE. Small paper mechanisms illustrate taking in, shifting and losing connection. The title returns, followed by each subtitle heading separately. The final card shows the book cover, Phil Gray, Gray Matter Books, Kindle on 22 September, paperback and hardcover on 29 September, and eatshiftdie.com. Peter Baker narrates; there is no music or running counter. The scenes are illustrations rather than documentary footage. Start the conversation here ## Three places to start. 01 / The organisation ### What if your AI works, and the organisation doesn’t? Explore why faster task completion can leave the actual work stuck at the next decision. [Try the idea ↗](https://eatshiftdie.com/tool-workflow.html) 02 / The human part ### What did you still have to do yourself? The book’s production offers a concrete account of authorship, challenge and judgement with several AIs involved. [Read the experiment ↗](https://eatshiftdie.com/experiment-book.html) 03 / Your move ### Where do you start without a grand plan? Small experiments help people build their own understanding of what AI changes in their work. [Make an experiment card ↗](https://eatshiftdie.com/tool-experiment.html) For podcasters, journalists and curious people ## Questions Phil is good value on. You say AI is not the story. What is? Why do AI pilots succeed while organisations stay unchanged? What did teaching Lotus 1-2-3 in the nineties teach you about this moment? What does your family’s connection to the Black Country bring to your thinking about industrial change? What’s wrong with the phrase “human in the loop”? You wrote this book with AI and used AI to challenge it. Who’s the author? What should someone do this week, at stakes they can afford? Eat, shift, or die. Which one are most organisations actually doing? For media & event organisers ## The useful details. For interviews, speaking, review copies or book enquiries: [info@eatshiftdie.com ↗](mailto:info@eatshiftdie.com) [Phil on X / Twitter: @philgray ↗](https://x.com/philgray) Publication on 22 September 2026. Cover artwork and notes below may be used when covering the book, with credit to Phil Gray. Ready for your programme, article or introduction ### About Phil Gray Short biography Phil Gray met his first chatbot on a cassette tape in 1979. Businesses built through the dot-com boom and out the other side, innovation labs around the world, decades close enough to enterprise change to hear the gears grind. He has been the disruptor and the disrupted. He lives in Sydney. Medium biography Phil Gray is the author of Eat. Shift. Or Die., a book about what AI changes, how organisations adapt, or don’t, and what comes next for you. His interest began with ELIZA on a TRS-80 in a Sydney home office in 1979. Since then he has built businesses through the dot-com boom and out the other side, run innovation labs around the world, spent decades close enough to enterprise change programs to hear the gears grind, and used AI on everything from family tech support to horse racing, with mixed financial results. He has been the disruptor, the disrupted and the person helping others through the collision. He writes about the work around the technology, with experiments, questions and responsibility for what makes it onto the page. He lives in Sydney. Extended biography Phil Gray is the author of Eat. Shift. Or Die., a book about what AI changes, how organisations adapt, or don’t, and what comes next for you. His story begins in a Sydney home office in 1979. A nine-year-old typed into ELIZA on a TRS-80 and a machine answered back. Since then he has built businesses through the dot-com boom and out the other side, run innovation labs around the world, spent decades close enough to enterprise change programs to hear the gears grind, and used AI on everything from family tech support to horse racing, with mixed financial results. He was paid to test computer games after school while his mates did paper runs. He taught Lotus 1-2-3 to sales teams who would rather have been at the pub, watching technology seep into ordinary office work. He built a business that put computer classrooms on wheels and ran training inside a bank. In innovation labs inside global banks across Asia, he helped build bathroom mirrors that showed people their money and used augmented reality to explore shipping assets the bank had financed. He learned the difference between innovation and innovation theatre by staging a fair amount of both. He has asked the same question as the disruptor arriving with the demonstration, the leader whose expertise was disrupted and the consultant helping organisations respond. What happens when new capability meets the way an organisation actually works? He wrote this book with the technology it describes, using several AI systems to research, edit and challenge while retaining responsibility. The companion website shares experiments, exercises and sources. Phil lives in Sydney with a Tesla, heavy metal and machines with names. BookEat. Shift. Or Die. SubtitleWhat AI changes. How organisations adapt, or don’t. What comes next for you. Publication22 September 2026 PublisherGray Matter Books Planned editionsKindle, paperback and hardback [Download the press kit .PDF ↗](https://eatshiftdie.com/assets/eat-shift-or-die-press-kit.pdf?v=20260916-voice)[Copy-friendly press notes .TXT ↗](https://eatshiftdie.com/assets/press-kit.txt)[Download the current cover .PNG ↗](https://eatshiftdie.com/assets/cover-selected.png?v=20260916-voice)[Read a short extract READ ↗](https://eatshiftdie.com/extract.html) For an author photograph, review copy or print-resolution press file, contact Phil. Keep listening ## More from Eat. Shift. Or Die. [Visit the YouTube channel ↗](https://www.youtube.com/@eatshiftdie) --- # Book trailer transcript | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/transcript.html [← Back to the film](https://eatshiftdie.com/media.html#listen)Accessible transcript The book trailer · 03:28 # Eat. Shift. Or Die. Full spoken transcript. Kindle: 22 September 2026. Paperback and hardcover: 29 September 2026. The new analyst asks AI a question it took you fifteen years to learn how to answer. Forty seconds later, the reply arrives. It's good. Good in a way that would once have earned a nod from people whose nods you spent a decade earning. You're still deciding how you feel about that. This book is for that feeling. The story goes back to 1979, when a nine-year-old boy in Sydney met ELIZA, the world's first chatbot, loaded off a cassette tape. The machine seemed to talk back. He's been watching this technology ever since, through every wave of hype and every disappointment. In 1950, Alan Turing set the most famous finish line in computing. That boy in Sydney grew up wondering what crossing it would feel like. He assumed there would be headlines. Possibly a parade. Somewhere in the last few years, by some measures, we crossed it. The most anticipated moment in the history of computing arrived, and the world checked its phone. If you were waiting for the big AI moment, you missed it. There wasn't one. It seeped in, one software update at a time. And it's already changing your organisation, vacancy by vacancy, task by task, whether or not anyone approved it. The shipping container was just a box. The revolution came when the world reorganised around it. AI is the same kind of arrival. The question isn't what the technology can do. It's what has to reorganise around it. Your organisation. Your profession. You. This book draws on decades spent as the disruptor, the disrupted, and the person hired to help everyone adapt. It isn't another book about writing a better prompt or getting rich with AI. It's about what happens after the demo, when the technology meets real work, real incentives, and people who were busy before the future arrived. Eat. To take in what the world is actually telling you, especially when it is inconvenient, unfamiliar or badly timed. Shift. To change position, practice or identity before the old one fails completely. Or die. To lose relevance, agency or viability while still insisting nothing fundamental has changed. It is a sequence, and the order matters. Eat. Shift. Or Die. What AI changes. How organisations adapt, or don't. And what comes next for you. Twenty-second September. Twenty-ninth September. ## Visual description The film opens in a layered-paper office, where a young analyst shows an AI answer to an older colleague. Text reads, This book is for that feeling. The camera moves gently towards a paper cassette recorder before a boy at a computer introduces ELIZA in 1979. A paper portrait based on Phil Gray’s generated office photograph shows him writing in a notebook, then lifting his gaze with a slight smile. His face, silver hair, black glasses and shirt are built from visible paper layers. A paper illustration of Alan Turing is followed by pedestrians absorbed in their phones. Colleagues pass orange folders around a vacant desk as paper office routes change. A shipping container arrives and infrastructure unfolds around it. An older worker packs his belongings into a box and leaves his desk. All illustrations, buildings, clothing and props use textured cream, kraft and aubergine paper, with orange accents. Definitions build line by line beside parchment EAT, outlined SHIFT and orange OR DIE. Small paper mechanisms illustrate taking in, shifting and losing connection. The title returns, followed by each subtitle heading separately. The final card shows the book cover, Phil Gray, Gray Matter Books, Kindle on 22 September, paperback and hardcover on 29 September, and eatshiftdie.com. Peter Baker narrates; there is no music or running counter. The scenes are illustrations rather than documentary footage. The film uses illustrated scenes. It is not documentary footage. [Back to the film ↗](https://eatshiftdie.com/media.html#listen) --- # Privacy | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/privacy.html Your information # Privacy. What this site collects, what stays in your browser and how to contact us. ## What is collected If you sign up, the site collects your email address. It records the signup source, the exact permission requested, when you requested and confirmed it, delivery status and any unsubscribe. Older subscriber records may retain the first name provided under the earlier signup form. Your network address is used briefly in memory to limit automated signup abuse, but it is not written to the subscriber database or application request log. The interactive exercises and ELIZA simulation run entirely in your browser. Answers and conversations are not stored, transmitted to an AI service or linked to you. Leaving or reloading clears them. Text you choose to copy, download or print remains under your control. ## Why it is collected Get told when the paperback is live, plus the monthly sweep. Nothing else. Your address is used to verify that request and send only those updates. No newsletter permission is activated before you confirm your email. Your address is not sold, rented or shared for unrelated marketing. Earlier subscribers keep the permissions they originally gave and any existing private-reader access. They are not automatically subscribed to the monthly sweep. Submitting the new form creates a separate request that must be confirmed before that permission is added. If you contact Phil by email, your address and message are used only to respond to you and maintain any necessary correspondence. ## Who processes it Signup requests are processed by this website and stored in a private database on a Hostinger server in Malaysia. They are not placed in the public website files. Email is delivered through the author’s Fastmail account by Kev247 using the book’s info@eatshiftdie.com address and fixed, auditable templates. Fastmail’s primary servers are in the United States and its services may process information in other countries. Kev247 can access subscriber information only when needed to monitor delivery, fulfil an access, correction or deletion request, or send the specific updates you requested. ## Analytics The homepage uses Plausible Cloud, a cookie-free analytics service hosted in the European Union. It records aggregate page views, referral and campaign information, approximate location and device information, and named events such as opening the book enquiry page or successfully submitting the waitlist form. Plausible briefly processes network and browser information to produce aggregate counts but does not use tracking cookies or retain a personal visitor profile. It does not receive email addresses, names, diagnostic answers or subscriber access tokens. Its script is not loaded on verification, unsubscribe or private introduction pages. ## Embedded video The book trailer, its captions and matching audio are served with this site. Links to YouTube open that service only when you choose them; its own privacy terms then apply. Links to research publishers and retailers also take you to external services. This site does not combine activity on those services with subscriber records. ## Cookies and storage The public site itself sets no tracking cookies and creates no advertising or behavioural profile. External sites may use their own cookies or browser storage after you choose to visit them. The private introduction reader remembers playback position only in the browser on the device you use. ## Retention and deletion Unverified signup records are deleted after 30 days. Unsubscribed records are deleted after 90 days. Active verified waitlist records are kept until 25 August 2027 unless you request deletion sooner. Delivery records and private introduction access are deleted with the subscriber record. Unsubscribing stops future messages but does not immediately delete the record, so the introduction link can continue to work during the 90-day retention period. To have the record and private access deleted sooner, email [info@eatshiftdie.com](mailto:info@eatshiftdie.com). ## Unsubscribe rights Every update email includes an unsubscribe link. Unsubscribing takes effect for all future messages and requires no explanation. ## Access, correction and privacy complaints To ask what information is held about you, correct it, request deletion, or raise a privacy concern, email [info@eatshiftdie.com](mailto:info@eatshiftdie.com). Phil Gray will review and respond to the request. If a privacy complaint cannot be resolved directly, you may be entitled to contact the Office of the Australian Information Commissioner. ## Changes If this site changes what it collects or who processes it, this page will be updated before the change takes effect. ## Pre-order enquiries An email enquiry does not create an order, reserve stock or take payment. Your email application opens with a draft you can edit. Nothing is sent until you send it yourself. Updated 8 September 2026. New signup permission covers the paperback announcement and monthly sweep. Earlier consent records and the existing retention schedule are preserved. --- # Accessibility | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/accessibility.html Using this site # Accessibility. The ideas should be approachable, and so should the site. ## Reading and navigation The site uses semantic headings, labelled navigation and a skip-to-content link. Core articles and source collections can be read without JavaScript. The mobile menu remains available as ordinary navigation if JavaScript is unavailable. ## Interactive exercises Exercises use labelled native form controls. You can move through them with a keyboard. The workflow’s quantities are available as text alongside its visual bars. Exercises explain their assumptions and do not rely on colour alone. ## Audio and motion Audio starts only when you choose to play it, and captions and a full transcript are provided for the book trailer. The site respects reduced-motion preferences and has no automatically moving ticker or video. ## Something getting in the way? Send a note to [info@eatshiftdie.com](mailto:info@eatshiftdie.com), including the page and the difficulty you encountered. These provisions are not a claim of formal accessibility certification. --- # Confidently wrong. The racing experiment | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/racing-story.html Field story 04 · February–August 2026 # Confidently wrong. A man, a racing model and several very clever machines. What could possibly go wrong? [Follow the experiment ↓](https://eatshiftdie.com/racing-story.html#beginnings) [Illustration: A paper horse and jockey emerge from a desktop monitor while a person compares a form guide beside the track.] Ambition, illustrated. THE SATURDAY EXPERIMENTBY PHIL GRAY · AN ILLUSTRATED RECONSTRUCTION THE FORM LOOKED GOOD. THE HORSES HAD OTHER IDEAS. This began with racing. It became a very practical education in working with AI. The dates and numbers follow the surviving records; imagined dialogue is marked reconstructed scene. THE HUMANPhil Curious. Impatient. Paying for the lesson. THE ANALYSTClaude Finds a pattern. Writes a very persuasive case. THE ENGINE ROOMKev Fetches, builds, schedules. Wrestles with the plumbing. THE LATER ARRIVALCody Tests the claims. Discovers his own blind spots. 01 28 February → 14 March 2026 ## A hundred bucks. A thousand-dollar idea. First, Kev made useful race-day spreadsheets. Then the idea grew teeth: give the AI $100 and see whether it could turn that into $1,000 in a year. There was a name. A website plan. An algorithm. The little operation looked the part. The money went the other way. KEV’S SURVIVING LEDGERMAR 2026 THE AMBITION$100 → $1,000Target, not an achieved return 7 Mar · start $100.00 First six bets $52.63 One more loss $42.63 14 Mar · close $28.63 By 14 March, the ledger records $28.63 remaining. What changedA convincing setup does not make the underlying judgement good. ↳ Open the record Early Kev archive 28 February journal; 7 March project README and audit log. The journal records Flemington and Randwick spreadsheets. The README sets a $100 starting balance and $1,000 target by March 2027. The project is called Confidently Wrong 2026. bankroll_history.csv; 7 and 14 March result files. Balances shown are the archive’s entries, not independently verified account statements. The separate performance summary was still stuck on the earlier $52.63 figure. This ledger is one early experiment, not a complete account of Phil’s racing spend. 02 21 March → 14 April 2026 ## Good news. The confidence was 92%. Kev admitted sending the wrong Rosehill race data. A few weeks later, his Hong Kong report had the same horse appearing five times in one field. The accompanying QA report was full of ticks. It declared the data good enough to use. Someone needed to check the checking. ARCHIVED QA CLAIM92%“QA PASSED” Look inside the race report ↓ RACE 1 · FIVE CONSECUTIVE ENTRIES #10LUCKY ARCHER #11LUCKY ARCHER #12LUCKY ARCHER #13LUCKY ARCHER #14LUCKY ARCHER Same name. Five different riders. These are the report’s entries, not a verified field. What changedAn official source can still pass through a broken extraction process. ↳ Open the record Kev’s admission and contradictory QA 21 March daily journal. Kev records confusing the race identity and failing to challenge implausibly uniform data. His explanation blamed test data from the provider; that cause has not been independently established. Hong Kong report and QA report, written 13 April UTC / 14 April Sydney. The narrative lists LUCKY ARCHER in slots 10–14. QA says initial parsing counted 448 horses, later claims 120, and assigns 92% overall confidence. The number is the agent’s self-assessment, not a measured accuracy rate. 03 18–25 April 2026 ## Then it got just good enough to be irresistible. 18 APRIL · ABBREVIATED BATCH0 / 10 Top picks won. The cold read had missed the headline. 25 APRIL · LOGGED BOXES3 / 10 Trifecta hits recorded. The next Saturday offered reasons to keep going. Claude was finding winners lower down the list, or mentioning them as outsiders. The prose sometimes knew more than the ranking. So the routine grew: read the form, challenge the case, watch the day unfold, move the picks, write the post-mortem. A useful hit feels very different from a useful lesson. It makes you want another go. Read→Challenge→Update→Check→Remember What changedPromote the useful signal out of the footnote. Then test whether the rule actually helps. ↳ Open the record Claude’s April retrospectives 18 April retrospective; 30 May setup review. The later review clarifies zero-for-ten refers to abbreviated cold batch top picks, not every full live analysis. The April aggregate contains inconsistent score labels, so this story does not pool its win rates. 19 April Sunshine Coast retrospective. The Irish was flagged in commentary but not promoted into the ranked four, then won. The note made numerical promotion of live signals central to the next version. 25 April retrospective. Records three trifecta hits among ten logged boxes. Several races were unlogged and dividends were incomplete. Its positive-day claim is not treated here as a reconciled profit figure. 04 2 May 2026 ## The machine kept working. Phil stepped back. The model’s “premium” race missed. Some races it had advised skipping produced hits. The labels were sounding more certain than the evidence deserved. Meanwhile, scheduled updates were firing into their own conversations. Work happened. The answer did not reach the person waiting for it. Fourteen remaining scheduled jobs were disabled. Phil stopped betting that day; the model carried on for learning. Reconstructed scene THE SYSTEMTask completed. PHILWhere’s the bloody answer? A job can succeed inside a machine and fail everywhere that matters. What changedDelivery is part of the job. So is knowing when to stop paying for another result. ↳ Open the record The mid-day pivot 2 May pre-retrospective summary. Records the notification failure, disabling 14 remaining fires, switching to the active conversation and continuing in calibration mode after Phil stopped active betting. The header and body disagree about whether he stopped after Race 3 or Race 4, so no exact stopping time is asserted. 05 30 May → 7 June 2026 ## Turns out the model also needed a janitor. By late May, old instructions pointed at dead folders. A “place lock” rule was still in the documents after the lessons had retired it. The learning database was six weeks stale. In June, the scheduled report said the racing tables were gone. You cannot learn from a memory you cannot find. [Illustration: A woman fits an orange connector between mismatched pipes beneath a computer surrounded by convoluted paper plumbing.] Illustrated reconstruction PATHSDeadThe files had moved on. INSTRUCTIONSArguingThe retired rule had survived. MEMORYMissingThe next run had nothing to learn from. What changedLearning needs a working memory, clear ownership and a routine that survives tomorrow. ↳ Open the record The maintenance nobody puts in the demo 30 May setup review. Records stale session paths, conflicting input locations, a retired rule still active, memory trapped in a read-only mount, missing track profiles and a six-week-old database. 6 June narrative-ingest report, run 7 June. Reports a paused database, then no application tables after restoration. It records no ingest. This is the contemporary report, not a new forensic finding about who or what removed the tables. 06 22 August 2026 · Kev’s return ## Send in more machines. Surely that will help. Kev came back with a research team to find trustworthy historical data. Five workers fanned out. The first launch lacked the access it needed. The next ran into the shared token limit. 5research workers → ~100calls before one lane failed → 1verifier that missed four material problems Cody caught the misses in the final read. Existing research was salvaged, the claims corrected, and another audit followed. More effort had helped. It had also produced more things to supervise. The verifier needed a verifier. What changedDivide the work. Keep someone responsible for whether the pieces make sense together. ↳ Open the record Codex’s dispatch and review history 22 August research thread and Kev’s final sourcing report. The thread records the failed launch, five concurrent workers exhausting the token allowance, bounded recovery, an initial verifier pass and four material misses found afterwards. Ten tasks ultimately completed, including correction and final acceptance. The consequential error. Date-keyed pages had been treated as proof of preserved pre-race data. They were not. A provider’s horse ID had also been confused with a verified official identity. These were corrected in the final report. 07 22 August 2026 · Opening the bonnet ## The model was learning the future. That helped. Cody recovered a version of Claude’s model that could actually be reproduced. Then came the suspiciously good improvements. Some “historical” statistics were not snapshots from before the race. Another apparent breakthrough carried information about the answer into the test. Brilliant results. Wrong experiment. Apparent improvement in prediction score +0.165THE EXCITING NUMBER Open the bonnet ↓ INVALIDATED Reversing summary records created errors related to which horse actually won. The test had picked up a clue from the answer. The gain was thrown out. What changedAsk what the model could have known before the event, not what the archive knows now. ↳ Open the record The gains that were refused Codex experiment report, reproduction and invalidated gains. All 717 handoff checksums passed. BLS v5.1 was reproducible; later simplified weights were not reproducible from the preserved clean matrix. Two invalidated gains. Published jockey/trainer aggregates appeared to improve log loss by 0.02451 but lacked point-in-time integrity. Reversed context summaries appeared to improve it by 0.16519 and were target-label dependent. Neither entered the frozen candidate. 08 22 August 2026 · Less magic, better questions ## A smaller improvement. A harder-earned maybe. Two prior-form signals offered a modest gain across 641 validation races. One time period still went backwards. The market benchmark remained better on the tested price subset. And the live hot-jockey / hot-stable rule that April had loved? Its specified version failed the locked August test. This time, improvement included taking something out. APRIL “Make the live signal mandatory.” ↓ AUGUST · 610-RACE TEST Remove the tested day-signal layer. Editorial summaries of the changing decisions BASELINE2.043 → CANDIDATE2.020 Average log loss across 641 validation races. Lower is better. Promising is not proven. What changedKeep the failed experiments. They are the reason the next version has fewer bad ideas. ↳ Open the record A promising candidate and a rejected rule Codex experiment report, 22 August. Rolling validation improved from 2.04319 to 2.02019. Gains across the three chronological folds were +0.03108, +0.03264 and −0.00153. The exposed 337-race block was not an untouched test. Locked day-layer audit. 610 races across 83 meeting clusters. The quality-adjusted layer worsened log loss by 0.00350; the inherited fixed-count rule worsened it by 0.00749. The specified Step 4A rule was deleted under the predeclared decision. This does not disprove every possible use of within-day information. 09 23–25 August 2026 ## Eighteen million simulated finishes. No final verdict. Ninety-four future races were locked before their results. The model was frozen. The scoring rules were written. The rehearsal worked. Then the real-world capture failed. One process stopped before packaging its archive. Required scripts were absent on the following days. The experiment could not be scored through its agreed process. After all that work, the honest result was an operational failure. THE LOCKED TRIAL94 races · 1,135 runners 18.8 million simulated orders 23 AUG INCOMPLETE24 AUG MISSING25 AUG MISSING NO MODEL VERDICT What changedA rehearsed workflow still has to work on the day. Missing evidence does not become a win or a loss. ↳ Open the record The trial that could not be opened 25 August blocked-opening receipt. All three independently audited dated outcome archives were unavailable. No canonical final comparison or final model verdict was produced. Three morning operational misses were retained; no successful morning manifest existed. Frozen prospective records. 94 races, 1,135 runners, 100,000 finishing-order simulations per race per model: 18.8 million orders across two models. These are simulated possibilities, not observed trials. The blockage says nothing conclusive about which model was better. 10 28–29 August 2026 · Back at Rosehill ## Four minutes to the jump. Don’t break the model. PHIL · 12:16 PM “4 minutes to race 2. We need updated predictions” PHIL · 12:17 PM “But don’t break the model.” Claude analysed overnight. Cody kept the evidence straight. The morning prediction stayed frozen; a separate live view could react to new information. Phil supplied results when official feeds lagged. THE MORNING CARDKeep it frozen. What did we say before we knew? THE LIVE VIEWLet it respond. What would we say with this new information? The separation mattered. Claude’s own local numbers were stale. On the last race, they would have left out the winner. The controlling card included it. What changedYou can adapt and remain honest, provided you keep the original prediction visible. ↳ Open the record The actual race-day conversation Codex history, 29 August. Phil’s messages arrived at 12:16 and 12:17 AEST. He later supplied final results and explicitly asked for official confirmation afterwards. The quotations above are verbatim. Claude end-of-day retrospective and controlling audit. Claude’s local Race 10 shortlist omitted Lisztomania; the controlling final-field card included it. The primary card was preserved. The audit also recorded a missed pre-jump review rather than manufacturing one after the race. 11 29 August 2026 · A side experiment ## Eighty-one per cent. Sounds good. A side experiment counted possible finishes for a spread of bets. More than 81% of the combinations returned more than the stake. One small problem: those finishes were not equally likely. When the quoted prices were used to weight the outcomes, the diagnostic pointed the other way. THE NUMBER THAT GETS YOUR ATTENTION81.1% of enumerated finishes above break-even And the question that changes it? ↓ ### How likely is each finish? 2,453 of 3,024 possible top-four orders were above break-even. Counting them equally ignores that some are much more likely. $32.00hypothetical stake→$26.15price-weighted expected return The latter is a quoted-price diagnostic, not a validated probability model or a realised return. What changedCount the possibilities. Then ask how much weight each one deserves. ↳ Open the record The coverage-bet experiment 29 August Merrylands R3 coverage receipt. Nine runners in the field, eight backed with the favourite excluded, four $1 markets per backed runner. It enumerated 3,024 ordered top-four outcomes; 2,453 (81.1177%) exceeded the $32 stake. The important distinction. Multiplicative normalisation of each quoted market gave a price-based expected return of $26.15, or −$5.85 net. Prices were held fixed; late deductions and dead heats were not modelled. Neither outcome counts nor this price diagnostic prove an executable edge. 12 29 August 2026 · What the finish line said ## Right horses. Wrong confidence. 3 / 9Top picks won 7 / 9Winners in the first four picks 2.248Model log loss · 2.246 for equal chance The shortlist found plenty. But two ugly misses left the average probability score slightly worse than an equal-chance benchmark. Finding a horse and knowing how much to believe in it are different skills. The cost of being surprisedWinner log loss by race · lower is better 1.88 R1top pick — R2excluded 1.90 R3in four 2.06 R4in four 1.97 R5top pick 4.13 R6miss 2.84 R7miss 1.44 R8top pick 1.72 R9in four 2.28 R10in four R6: Oh Arthur won after being given just 1.6%. R7: Plagiarism won at 5.8% in the model. Race 9 · All four on the shortlist ### All four. Different order. The frozen shortlist contained every horse in the actual first four. MODEL RANK I’mintowin Fully Lit Althoff Rolling Magic ACTUAL FINISH Althoff I’mintowin Rolling Magic Fully Lit What changedA good shortlist deserves credit. It does not excuse overconfidence or establish a profitable edge. ↳ Open the record The reconciled day, not the best-looking screenshot 29 August end-of-day audit against archived Australian Turf Club results. Ten races completed, nine scored. Race 2 was excluded because four official starters were absent from the locked prediction. Three top-pick wins and seven winners in four are both measured over the nine eligible races. Probability score. Mean winner log loss 2.248171. Equal-chance benchmark is the mean of log(field size) over the same nine races, approximately ${uniform.toFixed(4)}. Claude’s retrospective instead quoted 2.2034; this page recomputes it from the official starter counts for the nine eligible races. A lower loss is better. This is one meeting, not an estimate of long-run betting returns. Correcting the track input did not settle the argument. The Good 4 diagnostic was very slightly better on scoring rules over Races 4–10, but winner-in-four fell from five of seven to four of seven. A sensible correction did not deliver a clear practical gain. 13 What survived ## The bets were the tuition. The judgement was the certificate. The early book draft says the experiment lost money. The archive does not give us one clean, complete lifetime profit-and-loss account. There is no grand victory lap to invent. What it does show is a changing relationship with the machines: less accepting, more testing; less impressed by a confident answer, more interested in what survives a check. The useful part came to work on Monday. [Illustration: An older person makes notes with an orange pencil on a racecourse bench beside an empty track.] Illustrated reconstruction 01Form your own view. Before the consensus gets a vote. 02Give doubt a job. Someone must try to break the case. 03Keep the first answer. Make revision visible. 04Let a failure count. Especially when the story sounds good. ↳ Open the record Phil’s earlier telling The Saturday experiment, July book archive. Describes losing money while learning habits that transferred into the working week. “The bets were the tuition. The judgement was the certificate.” is from that draft. No manuscript file was changed to make this page. ### About this telling Built from Claude’s project notes and earlier book files, Kev’s surviving OpenClaw and Hermes records, and Codex’s conversations and experiment audits. The documented trail used here begins on 28 February 2026 and ends on 29 August. A folder named “Racing Jan26” does not establish a January start date. Illustrations, character sketches and marked dialogue are creative reconstructions. Agent-written retrospectives are treated as contemporary claims; contradictory records stay visible. This is the story of learning through an experiment, not a demonstration of a profitable betting system. Bring the useful part into your week ### What would you put to the test? [Plan your own experiment ↗](https://eatshiftdie.com/tool-experiment.html)[More field stories ↗](https://eatshiftdie.com/experiments.html) --- # Pre-order on Kindle | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/preorder.html [Illustration: The approved aubergine cover of Eat. Shift. Or Die. by Phil Gray.] Kindle pre-orders are open now # Pre-order on Kindle. Eat. Shift. Or Die. By Phil Gray Order the Kindle edition now. Amazon will deliver it to your Kindle library on 22 September 2026. [Pre-order on Amazon.com ↗](https://www.amazon.com/dp/B0HJ5XSQC7)[Amazon Australia ↗](https://www.amazon.com.au/dp/B0HJ5XSQC7) Use the store linked to your Kindle account. Amazon shows the price and confirms your pre-order. [Read an extract first ↗](https://eatshiftdie.com/extract.html) ### Want the paperback? The paperback launches on 29 September. Get an email reminder below. [Remind me about the paperback ↗](https://eatshiftdie.com/preorder.html#paperback) Paperback / 29 September 2026 ## The paperback lands 29 September. Leave your email and I’ll let you know when you can buy the paperback. Add the paperback to your Amazon wish list once its listing opens. For now, get a release reminder by email. ## Buying for a team? For multiple copies, a discussion with your team or a speaking enquiry, email [info@eatshiftdie.com](mailto:info@eatshiftdie.com?subject=Team%20copies%20and%20speaking). --- # Beyond the book | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/beyond.html Beyond the book # The next chapter is something you try. Follow an experiment. Have a go yourself. Or find the thinking behind an idea that stayed with you. [01 ↗ ## The experiments What I tried, what failed and what I learned.](https://eatshiftdie.com/experiments.html)[02 ↗ ## Try it yourself Small exercises you can run in your browser.](https://eatshiftdie.com/tools.html)[03 ↗ ## The monthly sweep New stories that put the book’s ideas to work.](https://eatshiftdie.com/sweep/)[04 ↗ ## The reading room Original sources, with a guide to what they can tell you.](https://eatshiftdie.com/sources.html)[05 ↗ ## The assumption graveyard Old management ideas, and what to try instead.](https://eatshiftdie.com/graveyard.html)[06 ↗ ## Meet ELIZA Talk to a green-screen chatbot. Then see how it works.](https://eatshiftdie.com/eliza.html) News parodyLIVE & DIE AI Just for fun ## The newsroom has no humans. Live & Die AI is an entirely AI-generated news parody, updated regularly with no humans involved. [Visit Live & Die AI ↗](https://eatshiftdie.com/live-and-die/)Here on eatshiftdie.com Access to AI will become common. The ability to reorganise around it will not. Eat. Shift. Or Die. Chapter 1 Implementing AI is a project. Living with it is a capability. Eat. Shift. Or Die. Chapter 7 --- # The monthly sweep | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/sweep/ The monthly sweep / August 2026 # The book, out in the wild. Ten stories to bring into the conversation. Secondhand books, changing apprenticeships and a mosquito air-defence system. Of course. Selected on 7 September 2026. Book connections are editorial interpretations; source reports and company claims are identified. 01 ## The next AI supply chain might run through a secondhand bookshop [Illustration: Conceptual cut-paper illustration of a secondhand bookseller watching an unexpectedly large order.] Editorial illustration The big story, seen from the shop floor · Technology arriving from outside your industry The Business Standard · 16 August 2026 · [Read the report ↗](https://www.tbsnews.net/offbeat/secondhand-book-sales-surge-ai-firms-fuel-demand-old-titles-1516106) The AI industry's appetite for material has acquired a wonderfully physical setting: shelves of old books. Reporting drawn from the BBC describes unusual bulk orders, including one that matched a bookseller's normal weekly volume. Sellers welcome the business while wondering where their stock is going. The connection between particular mystery orders and AI companies remains unconfirmed. The book connection: A technological shift can reach your business as a peculiar customer order. The bookseller doesn't need an AI strategy to find themselves making decisions about it. Watch for the moment an ordinary product becomes valuable for an entirely different reason. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 02 ## AI has become something you uninstall from Windows Seepage · Ordinary software maintenance Microsoft Support · 27 August 2026 · [Read the release notes ↗](https://support.microsoft.com/en-us/servicing/os/windows-11/2026/08/kb5120996-windows-11-26h1-update) Among the fixes and refinements in this Windows 11 preview update is permission to remove an installed image-generation AI component from supported Copilot+ PCs. Another line makes Fluid Dictation off by default for new users. AI is now sufficiently ordinary to have component-removal instructions and revised defaults beside touchpad settings. The book connection: Seepage has a second act: deciding which bits to keep. The interesting question becomes whether a particular feature helps with something you actually do. These are changes for specified Windows versions and supported hardware, with staged availability. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 03 ## You went to Search. It offered to make the study materials Seepage · A familiar tool takes on more of the task Google, The Keyword · 19 August 2026 · [Read the announcement ↗](https://blog.google/products-and-platforms/products/search/back-to-school-study-tools/) Google announced practice quizzes in Search, interactive explanations, and file creation in AI Mode. Its example starts with handwritten notes and lecture slides and ends with a study document. The same familiar starting point can now find material, explain it and assemble something from it. The book connection: The boundary of the job moves inside a product people already use. A search box can become part tutor, part production assistant before anyone gives the change a name. Google describes a mixture of launched and rolling-out features; its announcement doesn't establish that they improve learning. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 04 ## The university's software update now includes a catch-up colleague Seepage · AI arrives in the tools around the work UNC Charlotte, Office of OneIT · 7 August 2026 · [Read the campus notice ↗](https://oneit.charlotte.edu/2026/08/07/zoom-updates-august-2026/) A back-to-semester Zoom notice explains lecture recaps in Canvas, meeting summaries and an AI catch-up prompt for late arrivals. It also covers scheduling and the refreshed web portal. That unremarkable packaging is what makes it interesting: an institution is introducing capabilities through the ordinary business of helping staff use their software. The book connection: The person taking notes, the person arriving late and the person teaching now have different choices. Start there when asking what has changed. The notice also lists local restrictions, so this is a specific campus example, not a claim that every Zoom feature is enabled everywhere. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 05 ## Faster coding made this apprenticeship spend longer on the foundations [Illustration: Conceptual cut-paper illustration of a mentor and apprentice building strong foundation steps.] Editorial illustration Expertise and apprenticeships · The work after the demonstration AI-First Nation, Laurence Liew · 26 August 2026 · [Read the account ↗](https://aifirstnation.org/insights/inside-aiap-the-apprenticeship-that-built-singapores-ai-engineering-team/) Laurence Liew describes a change to AI Singapore's nine-month apprenticeship: foundation work grew from two months to three, while project work shrank from seven to six. His explanation is that there is more engineering to learn, while generative AI helps prepared teams deliver faster. Apprentices work with mentors on real industry problems through deployment. The book connection: This is a useful response to the anxiety about losing the first rungs of a career. Redesign the learning, including what beginners must understand before speed becomes useful. The programme's [current structure ↗](https://aiap.sg/apprenticeship/) corroborates the phases; the explanation for the change is Liew's account. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 06 ## Australia's AI question has reached the way the business runs Australia · Organisations adapting ABC News, Clare Armstrong · first posted 31 August 2026; updated 1 September · [Read the report ↗](https://www.abc.net.au/news/2026-08-31/ai-could-boost-australias-economy-treasury-flag-slow-uptake/107099934) Advice to the Treasurer, reported by the ABC, describes Australian AI adoption as widespread but shallow, with fewer than one in ten businesses reporting significant adoption. The proposed next step involves processes, management practices, business models and skills. That is a fairly substantial agenda hiding behind the little word “adoption”. The book connection: Buying access gets you to the beginning of the organisational work. This is the conversation about who can change a process, who owns the handover and how people learn new ways of working, arriving in a national economic story. The figures and advice here are attributed to the ABC's reporting. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 07 ## The designer's next job may be choosing what deserves to survive Work changing shape · Judgement, taste and slop UX Magazine, Rich Weborg · 5 August 2026 · [Read the essay ↗](https://uxmag.com/articles/the-evolving-design-process) Weborg describes a design process in which more people can produce credible screens early: product managers, engineers and founders can bring something visible to the conversation. He argues that the designer's contribution shifts towards direction, coherence and evaluation. His concern is the spread of polished sameness. The book connection: Imagine arriving at a meeting with twenty plausible versions instead of one unfinished sketch. You have gained options and acquired a selection problem. Taste starts to look like work you can name, discuss and practise. This is a practitioner's argument, not a measured finding about every design team. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 08 ## Someone said the quiet part out loud Executive quote · Organisational expectations Thomson Reuters, CIO Jason Escaravage · 5 August 2026 · [Read his account ↗](https://www.thomsonreuters.com/en-us/blog/how-thomson-reuters-is-using-ai-to-reimagine-how-we-work/) “We expect ALL employees at Thomson Reuters to be using AI, every day.” Escaravage places that expectation in an account of internal experimentation giving way to habits, manager coaching and changes to whole workflows. It is a strikingly direct statement of what an employer now expects of professional practice. The book connection: Once using the tool is expected, the useful conversation concerns what people use it for, what they learn and what counts as a good result. A daily-use expectation creates those questions; it doesn't answer them. This is the company's stated approach, with its outcomes still to be judged. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 09 ## The web turned 35. Remember the work that made it spread Historical echo · Infrastructure and adoption MuyComputer, Jose Montes · 8 August 2026 · [Read the anniversary piece, in Spanish ↗](https://www.muycomputer.com/2026/08/08/la-world-wide-web-cumple-35-anos/) August's anniversary marks the web's public announcement in 1991. [CERN's history ↗](https://home.cern/science/computing/the-birth-of-the-web/short-history-web/) supplies the less glamorous detail: the first browser needed a NeXT computer. Nicola Pellow's simpler browser could run on other systems. Later, friendly browsers for common computers helped the web spread. The book connection: Keep an eye on the people making a capability usable in ordinary circumstances. The spectacular invention needs that work around it. This anniversary concerns public introduction: the proposal dates to 1989 and the first server and browser were running in 1990. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) 10 ## Your next garden gadget would like to establish air superiority The weird weak signal · Capabilities finding unexpected uses Tom's Hardware, Jowi Morales · 15 August 2026 · [Read the story ↗](https://www.tomshardware.com/peripherals/futuristic-mosquito-zapping-laser-now-available-to-buy-video-shows-device-in-action-tiny-device-shoots-down-bugs-like-a-personal-air-defense-system-but-costs-usd1-000) Photon Matrix pitches a device that detects flying insects and targets them with a laser. The combination includes sensing hardware and an AI vision module. The August story reported the manufacturer's production plans; the [current shop page ↗](https://store.photonmatrixlab.com/products/photon-matrix-infrared-laser-mosquito-air-defense) still calls it a pre-order. Confirmed delivery and independent performance are separate questions. The book connection: Once capabilities become available as components, people put them together around remarkably specific irritations. Apparently that includes mosquitoes. Watch for the unexpected application, and for the journey from an impressive demonstration to something people actually live with. [Back to this edition ↑](https://eatshiftdie.com/sweep/#main) Stay with the question ## Follow the original thinking. [Open the reading room ↗](https://eatshiftdie.com/sources.html) --- # The monthly sweep / August 2026 | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/sweep/2026-08.html The monthly sweep / August 2026 # The book, out in the wild. Ten stories to bring into the conversation. Secondhand books, changing apprenticeships and a mosquito air-defence system. Of course. Selected on 7 September 2026. Book connections are editorial interpretations; source reports and company claims are identified. [← All editions and the latest sweep](https://eatshiftdie.com/sweep/) August 2026 · Archived edition 01 ## The next AI supply chain might run through a secondhand bookshop [Illustration: Conceptual cut-paper illustration of a secondhand bookseller watching an unexpectedly large order.] Editorial illustration The big story, seen from the shop floor · Technology arriving from outside your industry The Business Standard · 16 August 2026 · [Read the report ↗](https://www.tbsnews.net/offbeat/secondhand-book-sales-surge-ai-firms-fuel-demand-old-titles-1516106) The AI industry's appetite for material has acquired a wonderfully physical setting: shelves of old books. Reporting drawn from the BBC describes unusual bulk orders, including one that matched a bookseller's normal weekly volume. Sellers welcome the business while wondering where their stock is going. The connection between particular mystery orders and AI companies remains unconfirmed. The book connection: A technological shift can reach your business as a peculiar customer order. The bookseller doesn't need an AI strategy to find themselves making decisions about it. Watch for the moment an ordinary product becomes valuable for an entirely different reason. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 02 ## AI has become something you uninstall from Windows Seepage · Ordinary software maintenance Microsoft Support · 27 August 2026 · [Read the release notes ↗](https://support.microsoft.com/en-us/servicing/os/windows-11/2026/08/kb5120996-windows-11-26h1-update) Among the fixes and refinements in this Windows 11 preview update is permission to remove an installed image-generation AI component from supported Copilot+ PCs. Another line makes Fluid Dictation off by default for new users. AI is now sufficiently ordinary to have component-removal instructions and revised defaults beside touchpad settings. The book connection: Seepage has a second act: deciding which bits to keep. The interesting question becomes whether a particular feature helps with something you actually do. These are changes for specified Windows versions and supported hardware, with staged availability. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 03 ## You went to Search. It offered to make the study materials Seepage · A familiar tool takes on more of the task Google, The Keyword · 19 August 2026 · [Read the announcement ↗](https://blog.google/products-and-platforms/products/search/back-to-school-study-tools/) Google announced practice quizzes in Search, interactive explanations, and file creation in AI Mode. Its example starts with handwritten notes and lecture slides and ends with a study document. The same familiar starting point can now find material, explain it and assemble something from it. The book connection: The boundary of the job moves inside a product people already use. A search box can become part tutor, part production assistant before anyone gives the change a name. Google describes a mixture of launched and rolling-out features; its announcement doesn't establish that they improve learning. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 04 ## The university's software update now includes a catch-up colleague Seepage · AI arrives in the tools around the work UNC Charlotte, Office of OneIT · 7 August 2026 · [Read the campus notice ↗](https://oneit.charlotte.edu/2026/08/07/zoom-updates-august-2026/) A back-to-semester Zoom notice explains lecture recaps in Canvas, meeting summaries and an AI catch-up prompt for late arrivals. It also covers scheduling and the refreshed web portal. That unremarkable packaging is what makes it interesting: an institution is introducing capabilities through the ordinary business of helping staff use their software. The book connection: The person taking notes, the person arriving late and the person teaching now have different choices. Start there when asking what has changed. The notice also lists local restrictions, so this is a specific campus example, not a claim that every Zoom feature is enabled everywhere. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 05 ## Faster coding made this apprenticeship spend longer on the foundations [Illustration: Conceptual cut-paper illustration of a mentor and apprentice building strong foundation steps.] Editorial illustration Expertise and apprenticeships · The work after the demonstration AI-First Nation, Laurence Liew · 26 August 2026 · [Read the account ↗](https://aifirstnation.org/insights/inside-aiap-the-apprenticeship-that-built-singapores-ai-engineering-team/) Laurence Liew describes a change to AI Singapore's nine-month apprenticeship: foundation work grew from two months to three, while project work shrank from seven to six. His explanation is that there is more engineering to learn, while generative AI helps prepared teams deliver faster. Apprentices work with mentors on real industry problems through deployment. The book connection: This is a useful response to the anxiety about losing the first rungs of a career. Redesign the learning, including what beginners must understand before speed becomes useful. The programme's [current structure ↗](https://aiap.sg/apprenticeship/) corroborates the phases; the explanation for the change is Liew's account. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 06 ## Australia's AI question has reached the way the business runs Australia · Organisations adapting ABC News, Clare Armstrong · first posted 31 August 2026; updated 1 September · [Read the report ↗](https://www.abc.net.au/news/2026-08-31/ai-could-boost-australias-economy-treasury-flag-slow-uptake/107099934) Advice to the Treasurer, reported by the ABC, describes Australian AI adoption as widespread but shallow, with fewer than one in ten businesses reporting significant adoption. The proposed next step involves processes, management practices, business models and skills. That is a fairly substantial agenda hiding behind the little word “adoption”. The book connection: Buying access gets you to the beginning of the organisational work. This is the conversation about who can change a process, who owns the handover and how people learn new ways of working, arriving in a national economic story. The figures and advice here are attributed to the ABC's reporting. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 07 ## The designer's next job may be choosing what deserves to survive Work changing shape · Judgement, taste and slop UX Magazine, Rich Weborg · 5 August 2026 · [Read the essay ↗](https://uxmag.com/articles/the-evolving-design-process) Weborg describes a design process in which more people can produce credible screens early: product managers, engineers and founders can bring something visible to the conversation. He argues that the designer's contribution shifts towards direction, coherence and evaluation. His concern is the spread of polished sameness. The book connection: Imagine arriving at a meeting with twenty plausible versions instead of one unfinished sketch. You have gained options and acquired a selection problem. Taste starts to look like work you can name, discuss and practise. This is a practitioner's argument, not a measured finding about every design team. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 08 ## Someone said the quiet part out loud Executive quote · Organisational expectations Thomson Reuters, CIO Jason Escaravage · 5 August 2026 · [Read his account ↗](https://www.thomsonreuters.com/en-us/blog/how-thomson-reuters-is-using-ai-to-reimagine-how-we-work/) “We expect ALL employees at Thomson Reuters to be using AI, every day.” Escaravage places that expectation in an account of internal experimentation giving way to habits, manager coaching and changes to whole workflows. It is a strikingly direct statement of what an employer now expects of professional practice. The book connection: Once using the tool is expected, the useful conversation concerns what people use it for, what they learn and what counts as a good result. A daily-use expectation creates those questions; it doesn't answer them. This is the company's stated approach, with its outcomes still to be judged. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 09 ## The web turned 35. Remember the work that made it spread Historical echo · Infrastructure and adoption MuyComputer, Jose Montes · 8 August 2026 · [Read the anniversary piece, in Spanish ↗](https://www.muycomputer.com/2026/08/08/la-world-wide-web-cumple-35-anos/) August's anniversary marks the web's public announcement in 1991. [CERN's history ↗](https://home.cern/science/computing/the-birth-of-the-web/short-history-web/) supplies the less glamorous detail: the first browser needed a NeXT computer. Nicola Pellow's simpler browser could run on other systems. Later, friendly browsers for common computers helped the web spread. The book connection: Keep an eye on the people making a capability usable in ordinary circumstances. The spectacular invention needs that work around it. This anniversary concerns public introduction: the proposal dates to 1989 and the first server and browser were running in 1990. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) 10 ## Your next garden gadget would like to establish air superiority The weird weak signal · Capabilities finding unexpected uses Tom's Hardware, Jowi Morales · 15 August 2026 · [Read the story ↗](https://www.tomshardware.com/peripherals/futuristic-mosquito-zapping-laser-now-available-to-buy-video-shows-device-in-action-tiny-device-shoots-down-bugs-like-a-personal-air-defense-system-but-costs-usd1-000) Photon Matrix pitches a device that detects flying insects and targets them with a laser. The combination includes sensing hardware and an AI vision module. The August story reported the manufacturer's production plans; the [current shop page ↗](https://store.photonmatrixlab.com/products/photon-matrix-infrared-laser-mosquito-air-defense) still calls it a pre-order. Confirmed delivery and independent performance are separate questions. The book connection: Once capabilities become available as components, people put them together around remarkably specific irritations. Apparently that includes mosquitoes. Watch for the unexpected application, and for the journey from an impressive demonstration to something people actually live with. [Back to this edition ↑](https://eatshiftdie.com/sweep/2026-08.html#main) Stay with the question ## Follow the original thinking. [Open the reading room ↗](https://eatshiftdie.com/sources.html) --- # For AI readers | Eat. Shift. Or Die. Canonical: https://eatshiftdie.com/for-ai/ Public text and sources # Bring your own machine. If you read with an AI assistant, give it the source. These files contain the public pages of this website in a form it can read without having to negotiate the furniture. ## Start with the whole-site index. [The AI reading index](https://eatshiftdie.com/llms.txt) points to both the book website and Live & Die AI. It includes canonical links so an answer can take you back to the page it came from. ## Read the book website. [Read the public website as plain text](https://eatshiftdie.com/llms-full.txt) or [use the page catalogue as JSON](https://eatshiftdie.com/ai/site.json). This covers the book pitch, published extracts, experiments, tools, reading room, author information and monthly sweep. It is the website, not the full book. Private drafts, unpublished manuscript material, subscriber details and anything you type into a tool are not included. ## The newspaper is satire. [Read the newspaper as plain text](https://eatshiftdie.com/live-and-die/llms-full.txt), [get its public page catalogue](https://eatshiftdie.com/live-and-die/site.json) or [read the structured current edition](https://eatshiftdie.com/live-and-die/machine-edition.json). The AI-generated stories imagine what might happen next. They are not reports of events that have happened. Factual starting points, reported robot sport and entirely fictional weather are labelled separately. The newspaper exports refresh whenever an edition is published. [Follow its RSS feed](https://eatshiftdie.com/live-and-die/feed.xml) for new stories. ## Keep the source attached. Link to the original page when quoting or summarising it. Keep the distinction between an extract, an experiment, a sourced claim and a joke. An accessible file does not make the words more true, or turn a parody into a fact.