The rounds

The book in release mode (decision 008): a release candidate, then a round that takes one pile of material and ends in the next candidate, until the rounds shrink from parts to sentences. This page is the librarian's count of each round by size, the ideas that came in, and the diff Dinis reads.

The candidates

Round 1: v2.0.0-rc1 to v2.0.0-rc2

The book rewritten under the Gen ZX frame Dinis chose (decision 014): two generations, the founder as the reader, the objective that by the end the reader has started a company, a new part Start the company at the centre, Building, safely rebuilt as Build it without a server, a new first chapter and four more hands-on chapters, the measure in the last chapter; the round-1 batch of newsroom articles processed into ideas and written in. Where the newsroom has no article yet (the UK company, Lithuania, Portugal, the EU, the first customers, the community), the chapter stands as its title, its ideas and what is missing.

2parts added1parts removed11chapters added0chapters removed5chapters moved5chapters retitled198blocks changed486blocks added139blocks removed548blocks kept

The sizes are the round's shape (decision 008: parts, chapters, sections, paragraphs): this round added 2 parts (Start the company, Build it without a server), removed Building, safely, added 11 chapters, moved 5 and retitled 5; at the paragraph level 198 blocks changed, 486 were added and 139 removed against 548 kept. The diff, block by block, is what Dinis reads (decision 011).

The ideas that came in

88 ideas in the batch rc2: 15 advice, 33 claim, 9 evidence, 7 frame, 16 method, 4 process, 4 story. Each is on the register with its sources; none has been through a hearing: Dinis brought the pile in himself (decision 014).

• The reader this book is for is the founder, and its objective is that by the end you have started a company frame · gen-zx
• Two generations, Z and X, together Gen ZX: back to the ZX Spectrum frame · gen-zx
• Gen ZX are developers: coding with agents the explorer coding way, which is what vibe coding should be called claim · explorer-coding-and-the-handover
• Ship now, and ship your startup serverless: no servers of yours, no database, client side, with privacy by design and control of your agents advice · building-blocks
• The book is measured by how many companies are created from it, and builds a community of the people who tell us process · how-this-book-was-made
• How to actually do it: a limited company in the UK, Lithuania, Portugal, the European Union, the cloud providers and the programmes advice · the-company-itself
• The book is 200 printed pages or more; short and punchy stays the rule for chapters, not for the book process · how-this-book-was-made
• A change decided by precedent goes on a list Dinis can overturn, and Dinis reads the diffs of the book process · how-this-book-was-made
• The hands-on part is inside the book, dated, in the release candidate process · how-this-book-was-made
• Ship it, give it away briefly, take it away, and ask whether they miss it method · ship-it-and-see-if-they-miss-it
• The miss-it test applies to the builders: if you can change it and still do not use it, something is missing claim · ship-it-and-see-if-they-miss-it
• The speed of the loop from a comment to a feature is the moat, not the technology claim · ship-it-and-see-if-they-miss-it
• The four moves after 'do they miss it': define it, price it, deliver it at a flat marginal cost, offer it free to the people who know you method · price-it-then-give-it-away
• Free is never free for the other side: measure the cost you place on them, then cut it advice · price-it-then-give-it-away
• Do not make them apply: 'we have twenty, would you like one, or shall I give it to someone else' advice · your-first-price-and-first-ten-customers
• Count what matters: created, customised, used, and the minutes it cost the other side; set the stop rule first method · your-first-price-and-first-ten-customers
• Pay after the value: credit first, a meter, a balance that can go below zero, and no subscription by default advice · price-it-then-give-it-away
• Fund it, or do not: the worst time to raise is before you are profitable, and the best is when they call you advice · fund-it-or-do-not
• We study survivors: a risky practice can look like the cause of success when only winners are counted frame · fund-it-or-do-not
• Investors select for the founder they can steer; control is written into the contract, and founders get replaced claim · fund-it-or-do-not
• Sell, don't issue: the investor buys the founder's shares, the founder lends the company what it needs, and nobody buys the steering wheel method · fund-it-or-do-not
• What breaks: EIS is lost on bought shares, a big early sale reads as doubt, the founder-lender is a creditor, and a broad non-compete closes the fork claim · fund-it-or-do-not
• The expert founders, in numbers: 45 is the mean age, experience doubles the odds, a third of US startups have one founder evidence · be-a-founder
• Europe's slower, sceptical money suits a founder who can build first claim · fund-it-or-do-not
• SG/Send: first commit on 8 February, live on 13 February, Stripe on 1 March, ten paying customers on 2 March, a feature request shipped in 90 minutes story · ship-it-and-see-if-they-miss-it
• The technology worked and the use case did not: 'users buy credits, test once, and don't come back' claim · ship-it-and-see-if-they-miss-it
• The agents wrote the code, Dinis wrote the briefs and merged: 38% of commits in February, over 85% from September evidence · how-to-work-with-agents
• The SaaS apocalypse will be decided by inertia, not by AI, because AI is available to both sides claim · be-a-founder
• Users were never happy: 91% frustrated, 80% of features rarely or never used, $21 million a year in idle licences evidence · be-a-founder
• What SaaS refused to build to protect its moat is exactly what a customer's agent needs claim · be-a-founder
• A company sits in the two moves on the map: a guess becoming a product, and a product becoming a commodity frame · be-a-founder
• A budget per step, declared before anything runs, is the objective way to say good enough; the five whys move it method · direction-and-judgement
• Send an agent, not a spreadsheet: due diligence derived from the code, not composed by the vendor claim · sell
• A startup should double down on understandability: less code, the same tools, and it can be checked and pass advice · sell
• The agent behaviour policy is the due diligence document: a buyer reads the blast radius off it, and its absence says the most claim · controlling-your-agent
• Selling a service: each step small enough to say yes to, and each ends with something they keep method · sell
• The market is crowded at the top and thin on evidence: sell the record the client keeps claim · sell
• A service only scales with people: a curated marketplace of consultants, joined by proof of work method · sell
• Three honest tiers for a model on a site with no server: a local matcher, your own key, or a vault bridge method · ship-a-serverless-product-in-a-weekend
• Scope the feature to one question, and quote rather than paraphrase advice · ship-a-serverless-product-in-a-weekend
• Billing by not building billing: the commercial model is a property of the key method · ship-a-serverless-product-in-a-weekend
• Twice now the product was possible and the licence was the blocker: read the terms, not only the docs claim · ship-a-serverless-product-in-a-weekend
• VoiceDebrief: a product with no server that shipped in three weeks and stopped at the one thing it could not sell story · ship-a-serverless-product-in-a-weekend
• Git for things you cannot put on GitHub: what an encrypted vault is frame · privacy-by-design
• One string is address, authorisation and key; the read key derived from it is safe to publish claim · privacy-by-design
• An app in a vault runs against the vault without ever being handed a key claim · privacy-by-design
• Encryption on the client moves the problem from the data to the keys, and the sovereignty line to whoever holds them frame · privacy-by-design
• Where the keys live: a password manager for the keys a person holds, a registry vault for the keys agents make method · privacy-by-design
• A vault per conversation and per job: a leaked key exposes one small thing, not an inbox method · controlling-your-agent
• Append lanes: one POST from anywhere, readable only by the vault's owner method · privacy-by-design
• Nobody buys it for the encryption: the features came from the constraint claim · privacy-by-design
• Client-side encryption reduces the exposure under GDPR; it does not take you out of it claim · privacy-by-design
• Open source as a constraint: all of it, or it is an evaluation version with a better name claim · open-source
• What serverless costs when nobody is using it, product by product, 18 October 2026 evidence · building-blocks
• The name shows the glue: a first version is often a villager's work dressed as a town planner's claim · building-blocks
• Move the cloud, not the code: a twin the application cannot tell from the real service frame · building-blocks
• The reason to move is never sovereignty or cost: sell value that lands the day they switch claim · sell
• A villager move: no new features while you move, or nobody can tell which change broke what advice · explorer-coding-and-the-handover
• Every piece of software should run, unmodified, against a twin of every service it depends on advice · shipped-means-the-user-has-it
• The startup credit programmes, as their pages read on 7 October 2026 evidence · the-credits-and-programmes-2026
• Apply only where you meet the criteria, with the page that says what you would do, and publish the results advice · the-credits-and-programmes-2026
• The desktop is the barrier, not the model: prompt injection is not solved, so the box has to hold claim · controlling-your-agent
• A read-only vault needs no code; a read-write vault server is 83 lines on a free plan evidence · ship-a-serverless-product-in-a-weekend
• Reach is a computed value, and new capability lands as excess, never as new authority method · controlling-your-agent
• Who can pull the plug: an agent that cannot be stopped is riskier than one that can claim · controlling-your-agent
• Open source everything, and sell the customised, tested, maintained build advice · open-source
• Answer the hard questions in full, publish what is live and what is design, and check your own site against the answers method · sell
• A third input: a behaviour policy beside the system prompt and the prompt, which the vendor acknowledges, enforces where it can and answers for claim · controlling-your-agent
• Models do not separate code from data, so a rule in the prompt can be argued with by an email claim · controlling-your-agent
• Aviation and payments got safe when the rule, the liability and the business model pointed the same way evidence · controlling-your-agent
• Zoom into an agent's behaviour policy and you find the business logic, which only the business can write claim · controlling-your-agent
• Software was the law: the rule the screen enforced by omission is no longer enforced once an agent calls the API claim · controlling-your-agent
• Every rule lands in one of three states: backed by a control, governed by an accepted risk, or resting on hope method · controlling-your-agent
• Open source is not free: somebody pays, and the people who get the value have no easy way to claim · open-source
• Below about five pounds the card fee is the price, so pence payments only work aggregated evidence · price-it-then-give-it-away
• 'Nobody knows how it is built' is fixable: make the agents leave a record, a mandate, a memory and a graph advice · explorer-coding-and-the-handover
• The bonus test: if nobody will put their name or their bonus on it in production, it is not ready to be in line claim · explorer-coding-and-the-handover
• What the newsroom says about Lithuania and the Portuguese company, as of 12 October 2026: two lines evidence · the-company-itself
• The thinking is never wasted: pause in the open and pick up when the moment comes story · connect-the-dots-backwards
• Your own newsroom is the public example of your product, and underneath it a CRM and a publishing workflow claim · your-own-newsroom
• An explorer-coded app is a brief, not a product: not ready to be in line, and the further towards commodity the fewer agents in line claim · explorer-coding-and-the-handover
• Every product needs an explorer version, with paid conversions to a villager product and a town planner one advice · explorer-coding-and-the-handover
• Write down what your agent can reach in twenty minutes: list its tools, say what you wanted it for, write the gap method · controlling-your-agent
• Do not sell the need; answer a question somebody already has to answer, with a deadline advice · sell
• The opportunity for a small country is the plumbing, built once by the state like a road, not the race for GPUs claim · civic-responsibility
• No sovereignty without open source and open schemas; what the state pays for belongs to the commons; buy the alternative before the day you need it claim · civic-responsibility
• Smaller models, more of them, with the least agency: a public model is a public road, and constraints are the input claim · civic-responsibility
• What spending millions on Upwork taught: build the infrastructure first, and the jobs can go back to where people came from story · the-company-itself

Chapter by chapter

Chapter maps: the current one is admin/content/graph/chapters.json, the rc1 one is kept as chapters-rc1.json. The block diff uses the newsroom's page_versions.diff, the same differ as the versions page.