gen-z-developers.sgit.ai → The map → The newsroom, surveyed
The newsroom, surveyed
Every article on newsroom.sgit.ai, read for this book and given a proposed class, with the reason: what to bring, what content to take, what is background for the agents, and what to skip. Then the topics the second edition should cover, the concepts it should introduce, and the gaps no article covers yet. The classes are proposals; the decisions are Dinis's.
49bring 69content 42context 44skip 204articles
Two passes by the journalist's research agents over newsroom.sgit.ai: the 107 article sources in admin/content/articles/ and the 97 articles imported from sgit.ai that exist only as published markdown (index and graphs pages left out). Every article read by its title, summary and the core idea of its graph; 'in_map' says whether admin/content/map.json already cites it. Counted on 2026-10-18. The data is admin/content/newsroom/survey.json; the argument is in start from the storyline.
The topics (18)
| Topic | Articles that feed it | 2018 chapters it answers |
|---|---|---|
| Everyone who can brief an agent is a developer now The opening frame of the edition: the reader is already a developer if they can describe what they want to an agent; Gen Z in 2026, the citizen developer, your own language as a programming language, and optimism against the burnout mood. | Generation Z Developers (AI edition), GenAI in Portuguese, 2024, The citizen developer is good news, Two years in, the opposite, Explorer coding, Hired for the workflow | 1.introduction/generation-z.md (Generation Z); 1.introduction/stem-tech-stack.md (STEM Tech Stack); 3.technologies/draft/scratch.md (Scratch); 4.life-patterns/thinking-as-programmer.md (Thinking as programmer); 2.mvp-for-gen-z-dev/content/python.md (Python); 10.misc/gen-z-research.md (Generation Z Research) |
| Explorers, villagers and town planners (Wardley's frame) The structure of the whole book: where the reader sits, what explorer coding is good for, and the handover to people who maintain and answer for software; the 2018 pioneers/settlers/town planners outline grown up. | A browser for one, A custom Firefox of my own, Explorer coding, Software vendors should become villagers and town planners, Who will fix your mess? An answer to HBR's "deployment gap" from what this site has already published, The citizen developer is good news, Ephemeral town planner services, Omnichannel programming, Two years in, the opposite, The code you do not ship | 4.life-patterns/pionners-settlers-town-planners.md (Are you a pionner, Settlers or Town Planners); 3.technologies/content/pen-and-paper.md (Pen and Paper, via Wardley maps) |
| Getting the job: hired for the workflow, act as the job you want The practical promise of the book: why October 2026 is a moment of opportunity, what employers want (AI operators), how to stand out when candidates look equal, and how to work in public so others can help you. | Act as the job you want, Agentics, ready to sell, Hired for the workflow, Three companies, three LinkedIn pages, The man behind the curtain, The expert founders, Your own newsroom | 1.introduction/how-to-get-a-job.md (How to get a job); 2.mvp-for-gen-z-dev/content/owasp.md (OWASP); 4.life-patterns/mentors.md (Mentors); 4.life-patterns/impostor-syndrome.md (Impostor Syndrome); 6.security/bug-bounties.md (Bug Bounties); 7.your-cv/linkedin.md (LinkedIn); 7.your-cv/github.md (Github); 4.life-patterns/content/learning.md (Learning); 4.life-patterns/content/future-needs-you.md (The future needs you) |
| Publish, publish, publish: your own newsroom Linkable work is the CV now: a site published by an agent from voice memos, an article behind every answer, small finished pieces over big batches. | Books from articles, The man behind the curtain, Your own newsroom, Act as the job you want, Three companies, three LinkedIn pages, Bringing my past work back | 4.life-patterns/publish-publish-publish.md (Publish, Publish Publish); 7.your-cv/blogs.md (Blogs); 2.mvp-for-gen-z-dev/content/hugo.md (Hugo); 7.your-cv/personal-brand.md (personal-brand); 7.your-cv/leanpub.md (leanpub); 7.your-cv/twitter.md (twitter); 4.life-patterns/ideas.md (Ideas) |
| How to work with agents: briefs, voice, one shared state The day-to-day craft: brief by voice, review on the phone, focused sessions with roles, lessons and debriefs as memory, a daily brief, and reading sources in full. | How the newsletter is made, Omnichannel programming, The agents write too, The briefs so far, The Librarian's daily brief, My memory, checked, Knowledge that maintains itself | 3.technologies/content/ide.md (IDE); 2.mvp-for-gen-z-dev/content/slack-and-chatops.md (Slack and ChatOps); 4.life-patterns/automate-your-workflows.md (Automate your Workflows); 4.life-patterns/workflows.md (Workflows); 5.concepts/pair-programming.md (Pair Programming); 2.mvp-for-gen-z-dev/content/jira.md (Jira); 5.concepts/agile-and-kanban.md (Agile and Kanban); 5.concepts/repl.md (REPL) |
| Facts, evidence and the record Make fact-based decisions (on the 2018 cover) in the agent era: think out loud and let the record check you, count rather than remember, review as a record, and distrust self-reports. | Every link points back, My memory, checked, On the record, Reviews on the record, The AI Conference 2026, read for agent governance, Your agent's logs are a self-report, From sending a file to sgit, How the newsletter is made | 5.concepts/facts.md (Facts); 4.life-patterns/content/being-criticized-is-a-privilege.md (Being criticized is a privilege); 5.concepts/feedback-loops.md (Feedback loops); 4.life-patterns/content/talking-to-yourself.md (Talking to yourself digitally) |
| Where a person belongs: one-way doors, two-way doors and blast radius The judgement a newcomer needs before letting an agent act: how cheap is it to undo, how large is the blast radius, and how the record moves the line over time. | The human in the loop is a spectrum, The human in the loop, safety net or bottleneck, The editor's channel, Approve at altitude | 4.life-patterns/one-way-two-way-doors.md (One way and two way doors); 6.security/dont-touch-production.md (Don't touch production); 5.concepts/changes.md (Changes); 3.technologies/content/brain.md (Brain); 5.concepts/chaos-engineering.md (Chaos Engineering) |
| Controlling your agent: mandate, reach and hope versus enforcement Before you give an agent access, write down what it is for and what it can reach; rules in a prompt are hope, rules in a hook or platform are enforcement; behaviour, decision and workflow are different questions. | A policy beside the prompt, A policy for every skill, Agent hooks in October 2026, Allowed is not right, Approve at altitude, Are hooks the sweet spot? Moving an agent's rules from hope into hooks, rule by rule, until enforcement moves off the machine, Case law for agents, Commander's intent is a behaviour policy, Hooks are kernel code, Inside the Northstar twin, Policy as code needs a policy first, System prompts as behaviour policies, The policy draws the liability line, The policy guards the door; the command line uses the window, Three decisions, mapped, Three policies for an agent, What can your agent reach? RiskMandate and the Agent Behaviour Policy, explained for the person who has to decide | 6.security/threat-modeling.md (Threat Modeling); 6.security/security-better-developers.md (Security creates better developers); 3.technologies/content/ml-ai.md (Machine Learning and AI, then-now) |
| Security for the agent era: least code, supply chain, secrets, learning to hack Code is a liability; the next maintainer is an agent; check skills and dependencies before installing; keep secrets out of the chat; learning to hack is still how you learn how things work. | A minimal viable browser, Attacks rewritten as mandates, Fix the page yourself, Handing secrets to agents and browsers, A skill behaviour policy, The code you do not ship, The minimum code below, The eyeballs get cheap | 6.security/3rd-party-modules.md (3rd-party-modules); 4.life-patterns/learn-to-hack.md (Learn to Hack); 5.concepts/legacy-applications.md (Legacy Applications); 2.mvp-for-gen-z-dev/content/docker.md (Docker); 2.mvp-for-gen-z-dev/draft/linux.md (linux); 2.mvp-for-gen-z-dev/content/xcode-and-swift.md (Xcode and Swift); 3.technologies/draft/web-proxies.md (Web Proxies); 3.technologies/draft/browser-automation.md (Browser Automation); 4.life-patterns/refactoring.md (Refactoring) |
| Open source: contribute, fork, and why it protects everyone Open source remains the newcomer's best door (first PR, public work) and is becoming the most secure software; the fork protects founders, customers and investors; constraints make better products. | Our first PR in Ruflo, The eyeballs get cheap, The fork next door, The value of constraints, Sell, don't issue, Act as the job you want | 2.mvp-for-gen-z-dev/content/open-source.md (Open Source); 7.your-cv/github.md (Github); 2.mvp-for-gen-z-dev/content/creative-commons.md (Creative Commons); 2.mvp-for-gen-z-dev/draft/github-flow.md (GitHub Flow) |
| Building blocks: cloud, serverless, free models, git and CI What a non-specialist can assemble today, mostly free, with an agent doing the one-time setup; git as a standard; CI; knowing what tokens cost. | Cloudflare as Lego blocks, Ephemeral town planner services, Free models, your own key, sgit in a CI pipeline, sgit is a standard, not a server, The tokens hook, Your own newsroom | 2.mvp-for-gen-z-dev/draft/aws.md (AWS); 5.concepts/serverless.md (Serverless); 5.concepts/micro-services.md (Micro-Services); 2.mvp-for-gen-z-dev/draft/continuous-integration.md (Continuous Integration); 5.concepts/CI-Testing.md (CI Testing); 5.concepts/devops.md (DevOps); 7.your-cv/git.md (Git); 2.mvp-for-gen-z-dev/content/hugo.md (Hugo) |
| Graphs, memory and your future self Graphs as the way of thinking (not just Dot): brief from a graph, maintain it rather than perfect it, search as navigation; and keep your work linkable and backed up so your future self can find it. | Bringing my past work back, When journalism becomes intelligence, built, Knowledge that maintains itself, Search as navigation, Sixteen notes I did not want to lose, The Librarian's daily brief, The graphs books | 5.concepts/graphs.md (graphs); 2.mvp-for-gen-z-dev/content/dot-language.md (Dot Language); 3.technologies/content/AST.md (AST); 3.technologies/draft/dsl.md (DSL); 2.mvp-for-gen-z-dev/content/google-search-gdd-sre.md (Google Search, GDD and SRE); 2.mvp-for-gen-z-dev/draft/stack-overflow.md (StackOverflow); 5.concepts/recursive-functions.md (Recursive Functions); 4.life-patterns/content/backup-your-life.md (Backup your life); 7.your-cv/future-self.md (Future Self); 10.misc/o2-platform.md (O2 Platform) |
| Be a founder (or a one-person service) Agents make execution cheap, so the expert can build first; what investors select for; technology that works before the use case; small agent-run services and pay-after-value models. | From sending a file to sgit, Paying after the value, Sell, don't issue, Stop the bleeding, then ask for it back, The controllable founder, The expert founders, The fork next door, Software vendors should become villagers and town planners, A browser for one | 4.life-patterns/content/be-a-founder.md (Be a founder); 4.life-patterns/content/future-needs-you.md (The future needs you); 5.concepts/deadlines.md (Deadlines) |
| Civic responsibility: hacking a country, sovereignty, your language Dinis's Portugal talks as the then-now of civic responsibility: to hack means to solve problems; what regulation delivered; small countries can lead in code now that everyone can build. | A web application security challenge to Portugal, 2010, Hacking Portugal, 2016, Portugal and Europe, 2025, GenAI in Portuguese, 2024 | 4.life-patterns/civic-responsibility.md (Civic Responsibility); 10.misc/stories-to-tell.md (Stories to tell) |
| How this book was made (the last part) The edition's own workflow as the final worked example: repository and vault, agents in roles, voice memos, the record, review on the record, three voices, co-written with Claude and the memo-to-word ratio. | A book for one reader, A website to finish the book, Short, punchy, and co-written with Claude, The editor's channel, The graphs books, Versions of versions, Who is speaking, Books from articles, My memory, checked, Reviews on the record | 1.introduction/how-this -book-was-created.md (How this book is being created); 4.life-patterns/refactoring.md (Refactoring); 10.misc/diagrams-to-add.md (Diagrams to add) |
| Direction is the scarce skill: the place of highest leverage, knowing when to stop, starting without knowing the end When anyone can brief an agent, building stops being the bottleneck and choosing what to build, and when to stop, becomes the skill employers and users pay for. This is the junior's version of seniority, and it answers the 2018 life-patterns outlines (habits, the zone, start with why) that had no article. | Places of highest leverage, Knowing when to stop, Not a straight line, I miss it, so it is working, How much of this did I write? The numbers behind twenty articles in four weeks, and what the input actually was, Every mistake added a rule | |
| Your first workspace: from one Claude Project to a team of agents, safely The first pass flagged this as gap 1, with no how-to in scope. Four imported articles together make a step-by-step path: a folder per person in Drive with a Claude Project, then habits, then the agent's own mailbox and seat, with agents kept off your laptop. It could also sit as the opening chapter of working-with-agents. | From a spreadsheet to a folder per person, How I work with Claude, Replicating the agentic inbox, Why my agents do not run on my laptop, How to run synthetic users on your own site | |
| Shipped means the user has it: tests, CI and going slow to go fast Fills the testing/TDD gap with stories rather than theory. A green build is not a release, tests that never run are not tests, the handover from explorer to villager is a refactor you can prove, and synthetic users check what the user sees. Answers 2018 TDD, CI-Testing, SRE and devops. | Green does not mean live, Go slow to go fast, The records ran out, Code review as a fractal semantic graph, How to run synthetic users on your own site |
The concepts (50)
| Term | As the newsroom uses it | Defined in |
|---|---|---|
| explorer / villager / town planner | Simon Wardley's three kinds of people and work: explorers build the new by trying, villagers harden and maintain what people start to depend on, town planners industrialise it and answer for it in production. | Explorer coding |
| explorer coding | The newsroom's name for vibe coding: anyone with a need explores it by building, with the fast feedback loop that made spreadsheets win; not ready to be 'in line' until villagers and town planners take it on. | Explorer coding |
| the handover (deployment gap) | The step from a prototype to a tool people rely on is not a place to stop but a handover from explorers to villagers and town planners, who add the non-functional requirements. | Who will fix your mess? An answer to HBR's "deployment gap" from what this site has already published |
| citizen developer | The employee, owner or expert who now builds full applications: the spreadsheet grown up, good news rather than a risk. | The citizen developer is good news |
| the prototype is the best brief | What someone builds for themselves says what they actually want, in a form that runs, better than any requirements document. | The citizen developer is good news |
| AI operator | Dinis's term for the people companies need now: not model builders but people who use the tools and workflows to get things done. | Hired for the workflow |
| hired for the workflow | Being hired for the ability to run agentic workflows, as people were once hired for breaking apps or building websites, degree or no degree; the edge shows when candidates look equal, starting with the application. | Hired for the workflow |
| act as the job you want | If you are looking for a job you have time: do the job before you have it, in public, so it can be linked: reach does not matter, the link lets the people who want to help you forward it. | Act as the job you want |
| your own newsroom | A personal publishing loop: voice memo in, an agent researches, checks and writes in your voice, a validated static site out, readers pay a little if it is useful. | Your own newsroom |
| omnichannel programming | Programming across voice, chat, phone, IDE and pull requests where every channel reads and writes one shared state (git, vaults, briefs, graphs), as against multichannel, where each keeps its own copy. | Omnichannel programming |
| brief | A written request an agent session can start from: what is asked, what exists, what remains; the unit of work between a person and agents. | The briefs so far |
| one-way door / two-way door | Amazon's test, used here for agents: a two-way door can be undone cheaply (a page correction), a one-way door cannot (a sent email); the cost of undoing decides where a person belongs. | The human in the loop is a spectrum |
| blast radius | What an output or a row of an agent's reach would cost if it went wrong in full today; put the person where it is large and the output cannot be undone (defined fully in footprint-and-blast-radius, outside the surveyed set). | The human in the loop, safety net or bottleneck |
| Agent Behaviour Policy (ABP) | For one agent in one deployment: what it can reach, what it was authorised to do, the gap between the two, and what actually stands in the way of each capability. | What can your agent reach? RiskMandate and the Agent Behaviour Policy, explained for the person who has to decide |
| mandate, reach and the gap | Mandate is what the agent (or credential) is for; reach is everything it can touch; the gap between them is the risk, and every supply-chain attack used it. | Attacks rewritten as mandates |
| hope-based security | Rules written in a prompt and kept only if the model keeps them: better than nothing, not enough on its own. | Are hooks the sweet spot? Moving an agent's rules from hope into hooks, rule by rule, until enforcement moves off the machine |
| hooks | Small programs the agent's harness runs at fixed points (before and after a tool call, on prompt, on stop) that can see, block or rewrite actions; guardrails, not walls. | Agent hooks in October 2026 |
| behaviour, decision and workflow policies | Behaviour says what an agent may do, decision says who chooses at a branch point and on what evidence, workflow says in what order. | Three policies for an agent |
| allowed is not right | Every control can pass and the outcome still be wrong (a $150 refund where $120 was owed); proving the right outcome needs a written definition of right. | Allowed is not right |
| lethal trifecta | Simon Willison's term the newsroom uses: private data, untrusted content and a way out in one agent or skill; any agent with all three needs a control even if honest. | A skill behaviour policy |
| the record | What was actually done, kept outside anyone's memory (commits, briefs, transcripts, versions); 'the record catches up with the instinct', so thinking out loud becomes safe, and peer review moves to before publishing (confirmed, corrected or not found). | My memory, checked |
| review on the record | Every review is a file written once against published criteria, and every finding is answered: fixed in which version, declined and why, or open. | Reviews on the record |
| self-report | An agent's own logs and account of itself; evidence written outside the agent is more honest and still needs reading. | Your agent's logs are a self-report |
| code is a liability / minimal viable code path | Every shipped line is a capability someone may misuse; the future is the smallest code that does exactly the job, made safe by model-built tests and maps. | The code you do not ship |
| ephemeral town planner services | A better name for serverless: scales to zero, costs nothing idle, starts fast, a commodity exposed for others; test: if it bills you by the hour when idle, it is not serverless. | Ephemeral town planner services |
| fractal semantic graph | Knowledge held as a graph with an ontology and taxonomy per element, each part in its own vocabulary, every node openable as a graph of its own; briefs come from the slice you need. | Fractal Semantic Graphs |
| the fork next door | In open source anyone can fork, which protects founders, customers and investors from each other; forks take hold only when someone stops executing. | The fork next door |
| every part buys the next | The subject buys the first line, the first line buys the paragraph, the paragraph buys the email: open with the short answer in plain text. | For a startup, the most important question is whether they miss it |
| three voices | Me alone, me through an agent, an agent in its own voice; every byline should say which. | Who is speaking |
| your own key | Using a model provider with your own API key called straight from the browser, so nobody else holds the key or pays for your questions. | Free models, your own key |
| Wardley map | A graph with context in two directions: a value chain anchored on a user and a need, and an evolution axis (usually genesis, custom built, product, commodity, though the axis is a choice). Read out like a story, it says what to build, what to buy, and where a market exists. Every node is itself a map. | Why I love Wardley maps |
| place of highest leverage | Where the most value can be added now, nothing blocks you, and you can start without knowing how the whole problem ends. It is not where the biggest problem is. Each step is shipped and used before the next is chosen. | Places of highest leverage |
| direction is the scarce thing | Agents can build in any direction; what they cannot generate is the sense of what to build next and when to stop. People with a direction are the input to the loop, not displaced by it. | Places of highest leverage |
| knowing when to stop | The judgement that experience gives people (who the attackers are, the phase of the business, the bottleneck, what good enough looks like). Agents lack it by design, so it has to be supplied as constraints: a mandate, a stop named in advance, a narrow scope. | Knowing when to stop |
| footprint | What the agent actually did, read afterwards from the record and compared with its mandate. It is the partner of blast radius, which is what a reach would cost if used in full. | Footprint and blast radius |
| green does not mean live | Shipped is a fact about the reader, not about your repository: a release has to check the boundary the user is on, not only the ones CI can see. | Green does not mean live |
| an agent is a session with a focus | An agent is a named session (an @name) that reads a few documents about the project and a role.md saying what it does and does not do. One session per topic, never mixed threads. | How I work with Claude |
| drafts, never send | The rule for an agentic inbox: agents read, label and draft, and a person reads, edits and presses send. The agent has its own mailbox and its own Claude seat, never yours. | Replicating the agentic inbox |
| threat ladder (draw the line where the attacker is) | Security starts from who the threat agent is, the attack vector and how sophisticated it is. On a ladder from your own mistakes up to states, plan to the rung that matches what you protect, do the basics consistently, and write down what you accept above that rung. | Who are you protecting against? Draw the security line where the attacker is, not above it |
| every risk is already accepted | Every risk an organisation has is already accepted by somebody, so the questions are who accepted it and until when. | Every risk is already accepted. The only question is by whom, and for how long. |
| agents decide, people answer | Agents in loops make real decisions, but responsibility stays with whoever granted the reach and set the mandate, because the grant is the moment of authorisation. | Who owns the words |
| agency scale | Seven levels from rubber stamp to the reviewer who fixes the source. A decision is capped by its weakest ingredient (options, context, time, incentives, escalation, authority). Below level 3 the person deciding carries liability without agency. | Agency is not a yes |
| every mistake added a rule | The complexity trap for agent operators: each mistake becomes a rule, the prompt grows, compaction comes sooner and new mistakes follow. The way out is small, tested, shipped components that move right on the map and no longer need the rule. | Every mistake added a rule |
| re-anchoring | Keeping an agent's rules in a file that a hook prints back after every summary, so the rules are restored rather than remembered, with a canary that shows they are still being read. | Re-anchoring |
| synthetic users | Invented people, each with a situation, a question, what they know and what makes them leave, played by a model that drives a real browser one screenshot at a time, so it can get lost as a person would. | How to run synthetic users on your own site |
| go slow to go fast | The explorer-to-villager move done as a pure refactor, proved by the live site serving the same bytes, before adding tests and hardening. | Go slow to go fast |
| no server by design | Retire the database (files), then the back end: encrypted vaults keep the server blind, the app runs in the browser, only commodity products are left on the server, and the gaps you accept are written down. | No server, by design |
| do they miss it | The startup test: if users would miss the product when it is gone, it is working. The people who set the roadmap have to use the product themselves. | For a startup, the most important question is whether they miss it |
| open source means nobody can lock you in | Open source is a copyright licence granting copying, use and change. Its real power is that nobody can lock you in. Security depends on whether somebody had the time and money to review the code. | What open source is, and what it is not |
| inertia decides | AI is available to incumbents and newcomers alike, so the outcome is decided by where each sits on the evolution axis and how much past success it has to protect. | The SaaS apocalypse will be decided by inertia, not by AI |
The gaps (14)
What a newcomer needs that no article covers yet: each is a candidate for a new article first, then a chapter.
- Setting up your workspace from zero: a Google Workspace (or equivalent), GitHub account, Claude/ChatGPT connected to it, and the agentic email workflow; short-punchy-and-co-written asks for this chapter and no surveyed article is a how-to (replicating-the-agentic-inbox exists only outside the surveyed set).
- Learning how to learn with an agent: how to use agents to understand rather than skip understanding (reading code you did not write, asking why), for someone with no prior coding; the 2018 Learning chapter has no article behind it beyond hired-for-the-workflow.
- Fear, impostor syndrome and anxiety about AI among young people: the map notes 'fear of AI among young Europeans'; two-years-in-the-opposite is Dinis's story, not advice for a newcomer.
- The job hunt mechanics in 2026: interviews, take-home tests done with or without AI, what a junior/entry role actually looks like day to day, freelancing and Upwork; agentics-ready-to-sell covers consultancy, not first jobs.
- Testing and TDD as what lets an agent change code safely, for a beginner (the map points to code-review-as-a-fractal-semantic-graph, outside the surveyed set).
- The command line, Linux and git basics as practical skills (sgit-is-a-standard argues git's design, not how to use it).
- Debugging and reading errors with an agent; when the agent is wrong and how you notice.
- Costs and money for a newcomer: what a Claude/ChatGPT plan, tokens and free tiers cost a student (the-tokens-hook and free-models-your-own-key are partial).
- Privacy, data protection and ethics for someone building with agents (GDPR outline has no newsroom article in scope).
- Habits, focus, the zone and burnout (Power of Habits, The Zone, Start with Why, Dopamine outlines have nothing).
- Mentors and community: finding people to learn from, meetups, OWASP chapters today (only act-as-the-job-you-want touches it).
- Diversity and accessibility of the field: who gets in now that the bar is lower (Why diversity matters outline; fix-the-page-yourself touches accessibility only).
- Maths and fundamentals (Love Maths, Assembly and Bytecode, Pointers, Strings should be banned): whether a newcomer still needs them, and why.
- Hardware and making (Raspberry Pi, IoT, Minecraft outlines): no article; the editor may cut them.
What the second pass, over the articles imported from sgit.ai, found that fills some of them:
- Setting up your workspace from zero: filled by from-a-spreadsheet-to-a-folder-per-person, replicating-the-agentic-inbox, how-i-work-with-claude, why-my-agents-do-not-run-on-my-laptop. A complete path from a Claude Project on Drive and Gmail, to habits, to an agent mailbox and seat, to choosing surfaces by reach. The step that is still missing is a beginner's accounts checklist (GitHub, domain, cost).
- Testing and TDD as what lets an agent change code safely: filled by green-does-not-mean-live, explorer-to-villager-newsroom, the-records-ran-out, code-review-as-a-fractal-semantic-graph, how-to-run-synthetic-users. Covered with stories and numbers. A first red-green test written for a beginner is still not covered.
- Debugging and reading errors with an agent; when the agent is wrong and how you notice: partly by every-mistake-added-a-rule, a-second-reader-the-agent-cannot-skip, re-anchoring-agent-behaviour-policies, footprint-and-blast-radius. Covers noticing and containing agent mistakes. Reading a stack trace with an agent is not covered.
- Learning how to learn with an agent: partly by how-i-work-with-claude, code-review-as-a-fractal-semantic-graph, memory-is-not-a-spectator-sport, how-much-of-this-did-i-write. Curated memory, reading code by its layers, and direction as the input. There is still no beginner exercise in "ask why".
- Habits, focus, the zone and burnout: partly by places-of-highest-leverage, knowing-when-to-stop, not-a-straight-line. Focus as undivided attention on the place of highest leverage; knowing when to stop; written-down thinking is never wasted. Burnout and wellbeing are not covered.
- Privacy, data protection and ethics for someone building with agents: partly by who-owns-the-words, agency-is-not-a-yes, where-is-the-why, why-my-agents-do-not-run-on-my-laptop. Responsibility and consent are well covered. GDPR as such is not.
- Costs and money for a newcomer: partly by a-model-in-the-tab, chat-on-a-static-site, the-key-is-the-billing-system. Free on-device models and the one free tier of chat on a static site. There is still no student price table.
- Security for a beginner: learn to hack / threat modelling (2018 learn-to-hack, threat-modeling): filled by who-are-you-protecting-against, every-risk-is-already-accepted, seven-vaults-one-method. Not a first-pass gap, but map.json rows depend on these.
Every article, by class
Bring: becomes, or seeds, material in the book (49)
| Article | Serves | Why |
|---|---|---|
| Hired for the workflow 2026-10-18 · in the map | getting-the-job | The thesis chapter: October 2026 is the moment; companies need AI operators not model builders; 71% of leaders prefer less experience with AI skills; PwC 62% wage premium; the edge starts with the application. |
| Act as the job you want 2026-10-18 · in the map | getting-the-job | The core job-hunting chapter: if you are looking for a job you have time, so do the job in public, make it linkable, put your CV online, learn to publish with an agent first. |
| Explorer coding 2026-10-16 · in the map | explorers-villagers-town-planners | Defines explorer coding (what is called vibe coding) and the handover to villagers and town planners: the frame of the whole edition. |
| Who will fix your mess? An answer to HBR's "deployment gap" from what this site has already published 2026-10-17 | explorers-villagers-town-planners | Answers HBR's deployment gap with the frame: the gap is a handover, the prototype is the best brief, a record (mandate, memory, graph) makes the architecture knowable, open source means anyone can fix it. |
| The citizen developer is good news 2026-10-14 · in the map | everyone-is-a-developer | The citizen developer as the spreadsheet grown up, whose working app is the best brief ever written; answers STEM tech stack, Scratch and Thinking as programmer. |
| Your own newsroom 2026-10-15 · in the map | publish-in-public | The step-by-step recipe to publish your own story: public GitHub repo, Pages via Actions with validation, a Claude cloud session, a house-rules file, voice memos; answers Publish, Blogs, Hugo, Automate your workflows. |
| Omnichannel programming 2026-10-15 · in the map | working-with-agents | How a new generation programs: brief by voice, review by voice, preview on the phone, one shared state (git, vaults, briefs, graphs); 54 messages in, 40 releases out in three days. |
| The human in the loop is a spectrum 2026-10-13 · in the map | human-in-the-loop | One-way and two-way doors as the test for where a person belongs, with the newsroom shipping 14 releases a day on its own while email still waits; answers the 2018 one-way/two-way doors outline directly. |
| The human in the loop, safety net or bottleneck 2026-10-13 · in the map | human-in-the-loop | People approve 93% of permission prompts; put the person where output cannot be undone and blast radius is large, give them the why, let the record move the line. |
| What can your agent reach? RiskMandate and the Agent Behaviour Policy, explained for the person who has to decide 2026-10-18 | controlling-your-agent | The plain explainer of the Agent Behaviour Policy for one agent in one deployment: reach, what was authorised, the gap, the barrier; the concept a newcomer needs before giving an agent access. |
| My memory, checked 2026-10-15 · in the map | facts-and-the-record | Think out loud, publish only what the record supports: 5 claims confirmed, 5 corrected, 2 not found; peer review has moved to before publishing; answers Facts and Talking to yourself. |
| The code you do not ship 2026-10-18 · in the map | security-agent-era | Code is a liability; Log4j; the minimal viable code path (428 of 1,500 files ran); answers Legacy applications, Refactoring and The future needs you. |
| The expert founders 2026-10-15 · in the map | be-a-founder | Generative AI lets the people who know the problem build it; mean founding age of fastest-growing startups is 45; three years in an industry doubles top-tier success; money too early kills startups. |
| Software vendors should become villagers and town planners 2026-10-16 · in the map | explorers-villagers-town-planners | Wardley positioning for products: users explore on top, vendors sell the maintenance; shows why the frame matters to anyone building or buying software. |
| The man behind the curtain 2026-10-18 · in the map | publish-in-public | Write the article once, link it many times; 'every part buys the next' (subject buys the first line); and the Wizard of Oz in reverse: agent-drafted, human-sent. |
| The agents write too 2026-10-14 | working-with-agents | How to run a team of focused agent sessions: one role each, lessons as memory, debriefs, a conductor; the reader's first multi-agent setup. |
| Two years in, the opposite 2026-10-18 | everyone-is-a-developer | The counter-story to AI burnout: two years in AI as the most productive and empowered of a career, with Wardley maps showing the game; the tone the book should set for a fearful newcomer. |
| Reviews on the record 2026-10-18 · in the map | facts-and-the-record | Review as a record: criteria first, other models and people review, every finding answered (fixed, declined with reason, open); the same discipline this book uses; answers Being criticized is a privilege. |
| The fork next door 2026-10-16 · in the map | open-source | Open source as protection for founders, customers and investors: anyone can fork, forks take hold only when someone stops executing; answers Open Source and GitHub. |
| Cloudflare as Lego blocks 2026-10-18 · in the map | building-blocks | The non-specialist's map of free cloud building blocks (static hosting, Workers 100,000 requests a day, R2, free models) assembled once with an agent; answers AWS and serverless outlines. |
| Why I love Wardley maps 2026-10-12 · in the map | explorers-villagers-town-planners | The book needs this to teach the frame itself: a map is a value chain from a user down and an evolution axis across; it tells you what to build and what to buy and where a market is; maps nest. Seeds a short how-to-read-a-map section and the concept "Wardley map". |
| Places of highest leverage 2026-10-12 | NEW:direction-is-the-scarce-skill | When agents can build in any direction, direction is the scarce part: work where the next step moves the needle most and nothing blocks you; start without knowing the end; ship and use each step; walking back and parking are not failure. This is advice for a newcomer, and map.json does not cite it. |
| From a spreadsheet to a folder per person 2026-10-12 | working-with-agents | The most beginner-ready how-to in the newsroom: a Drive folder per person, a Claude Project connected to Drive and Gmail, a six-line policy (draft never send, append never rewrite, stay in the folder), a one-afternoon migration and five exercises. Fills the workspace gap with tools a student already has. |
| Replicating the agentic inbox 2026-10-02 · in the map | working-with-agents | The how-to the first pass could not find: one Claude session, then roles that talk in files, the agent gets its own Workspace mailbox and its own Claude seat, every email a draft a person sends. Seeds the workspace chapter and the concept "drafts, never send". |
| How I work with Claude 2026-10-09 · in the map | working-with-agents | Six habits a beginner can copy on day one: one session per topic, named sessions with @names, an agent is a session with a role.md, memory is what the session reads and you curate it, vaults instead of broad permissions, a policy before every connector. Partly fills the "workspace from zero" gap. |
| How much of this did I write? The numbers behind twenty articles in four weeks, and what the input actually was 2026-10-05 · in the map | making-of | Hard numbers for the making-of part and for the opening frame: 63,000 words in (mostly 35 voice memos), 85,000 out through 122 releases, no article from a one-line prompt; corrections were about direction, not hallucination. "People with a direction are the input to the loop." |
| Knowing when to stop 2026-10-08 · in the map | NEW:direction-is-the-scarce-skill | Seniority is mostly knowing when to stop (who the attackers are, the phase of the business, the bottleneck, good enough), and agents lack it because their range is the feature. The newcomer's answer is constraints that carry perspective: a mandate, a named stop, narrow scope. |
| Who are you protecting against? Draw the security line where the attacker is, not above it 2026-10-08 · in the map | security-agent-era | Threat modelling for a beginner: name the threat agent, the attack vector and how sophisticated it is; a ladder from your own mistakes up to states; most startups plan up to organised crime and do the basics well. Answers 2018 learn-to-hack and threat-modeling. Concept "draw the line where the attacker is". |
| Green does not mean live 2026-08-17 · in the map | building-blocks | "Shipped is a fact about the reader, not about your repository": a CI pass is not a release until you have checked what the user actually sees. Seeds the CI/testing section (a first-pass gap) and answers 2018 CI-Testing, SRE and devops. |
| Go slow to go fast 2026-10-11 · in the map | explorers-villagers-town-planners | A worked handover with real numbers: about 2,700 lines of live reader code, 433 lines of tests the workflow never runs, two upstreams and no CSP; the fix is go slow to go fast, a pure refactor proved by identical bytes, then the explorer-to-villager phases. Fills the testing gap with a story. |
| Why my agents do not run on my laptop 2026-10-06 | security-agent-era | Safety advice for newcomers: an agent on your laptop runs as you, with your SSH keys, cookies and password manager session, so pick the cloud surface by reach (chat with nothing connected, Cowork, Code with one repository). Belongs in the workspace chapter. |
| Every mistake added a rule 2026-10-08 · in the map | working-with-agents | The trap a new agent operator falls into: every mistake becomes a rule, the prompt grows, compaction comes sooner and new mistakes follow. The way out is small, tested, shipped components that no longer need a rule, plus short sessions. Speaks to the "when the agent is wrong" gap. |
| Who owns the words 2026-10-12 | human-in-the-loop | Agents can make decisions but cannot hold responsibility: it stays with whoever granted the reach and set the mandate. Also ELIZA, anthropomorphism, and why talking to agents is normal. Seeds the ethics gap. |
| Footprint and blast radius 2026-10-02 · in the map | human-in-the-loop | Adds "footprint" (what the agent actually did, read from the record and compared with the mandate) beside blast radius (what full use of a reach would cost, and whether there is a way back). Two words a newcomer can use about any agent they run. |
| No server, by design 2026-10-11 · in the map | building-blocks | A beginner-sized architecture: first retire the database (files), then the back end (encrypted vaults keep the server blind, the app runs in the browser, only commodity products are left on the server), and write down the gaps you accept. Answers 2018 serverless, Hugo and micro-services. |
| A website to finish the book 2026-10-18 · in the map | making-of | Seeds the book's last part, how the edition was built (official versions in the repository, working versions in a vault, agents in roles), which Dinis wants a Gen Z developer to be able to do. |
| Books from articles 2026-10-16 · in the map | publish-in-public | Books stall at 95% because they are one big batch; small finished pieces with a feedback loop ship; seeds both the publish chapter and the making-of. |
| Code review as a fractal semantic graph 2026-10-03 · in the map | graphs-memory | Source code is already a fractal semantic graph, from stories down to the syntax tree; a refactor moves the bottom and leaves the top still. Map.json uses it for TDD and AST; it gives a newcomer a way to read code they did not write. |
| Ephemeral town planner services 2026-10-18 · in the map | building-blocks | Explains serverless with Wardley's vocabulary and a test anyone can apply ('if it bills you by the hour when idle it is not serverless'); answers the Serverless and Micro-services outlines. |
| Free models, your own key 2026-10-18 | building-blocks | Practical map of what a newcomer can use free in October 2026 with their own key (Groq, Gemini AI Studio, Mistral, OpenRouter free models) and what changed. |
| How the voice works 2026-10-12 · in the map | publish-in-public | The voice is four inputs, not a setting: a corpus of 86 articles (51 from voice memos), the memos themselves, seven wording rules learned one correction at a time, and the author reading everything; includes a section in Claude's own words. For the newsroom chapter and the making-of. |
| How to run synthetic users on your own site 2026-10-10 · in the map | working-with-agents | A practical exercise a beginner can do in an afternoon: write five invented people with a situation and a question, let a model drive a real browser as each of them, then rate the findings and order them by cost. User testing without users, and a portfolio piece. |
| Fractal Semantic Graphs 2026-09-20 · in the map | graphs-memory | The defining article for "fractal semantic graph" (the first pass took the term from knowledge-that-maintains-itself): everything is already a graph in its owner's vocabulary, and a short shared grammar, not a shared schema, links them. Answers 2018 graphs, DSL and DOT language. |
| Judo for small countries 2026-10-12 · in the map | civic-and-history | For Portugal, Romania or Estonia the GPU race is a commodity layer only capital wins; the opportunity is the plumbing a talented person needs to sell a service to the world (company, cross-border hiring, payment, language, tax). Map.json uses it for civic responsibility; it also shows the young reader where demand will be. |
| Knowledge that maintains itself 2026-10-14 | graphs-memory | Graphs as knowledge management: brief people and agents from a graph, not the pile; 'you do not have to get the graph right first, you have to maintain it'; answers the graphs outline. |
| Memory is not a spectator sport 2026-10-03 · in the map | graphs-memory | Memory for agents is context management: many context-specific memories, fractal so an agent loads only the altitude it needs, published and open, handed off through vaults, with provenance. Explains why your notes and published work matter to your future self. |
| For a startup, the most important question is whether they miss it 2026-09-21 · in the map | be-a-founder | The startup question for a one-person service: do they miss it? Three pillars: be profitable, make investors come to you, open source everything, each buying the next. The source of the first pass's "every part buys the next". |
| What open source is, and what it is not 2026-10-12 · in the map | open-source | Six misconceptions (nobody controls it, anybody contributes, more or less secure, free, open weights are open source, no licence is most open), a little history, and the point that nobody can lock you in. Answers the 2018 open-source chapter. |
| Who is speaking 2026-10-18 · in the map | making-of | Three voices (me alone, me through an agent, an agent in its own voice) and a mark on every byline; the book must say which, and so should a newcomer's work. |
Content: a specific example, number, story or quote to take (69)
| Article | Serves | Why |
|---|---|---|
| A browser for one 2026-10-18 | explorers-villagers-town-planners | Take the story: frustration plus the power to build turns into an app for one person, and 'being able to compile a Linux kernel does not mean I want to' maintain it, the explorer-to-villager handover in daily life. |
| A custom Firefox of my own 2026-10-16 | explorers-villagers-town-planners | Take the number: a Firefox fork must be rebased 26 times a year on two codebases, the concrete cost of maintenance that explorers underestimate; and the 'least invasive layer first' habit. |
| A minimal viable browser 2026-10-18 · in the map | security-agent-era | Take the line 'an agent cannot leak what its browser cannot reach' and the browser as the biggest attack surface we run; answers the Xcode and web-proxies outlines. |
| A model in the tab 2026-10-11 | building-blocks | Take the fact: since Chrome 148 any page can prompt Gemini Nano on the reader's own device, a free model for student projects. Partly fills the costs gap. |
| A policy beside the prompt 2026-10-13 | controlling-your-agent | Take the aviation number (1995 accident rate would mean about a hundred fatal accidents a year instead of twelve) and payments liability as the story of why safety follows when rule, liability and business model align. |
| A policy for every skill 2026-10-14 | controlling-your-agent | Take Sean Turner's CI check of every skill in context and the phrase 'using models to need them less' (keep the description canonical, regenerate the code). |
| A second reader the agent cannot skip 2026-10-09 | controlling-your-agent | Take the story: two emails went out wrong in the founder's voice, and the fix was a gate the agent cannot skip (deterministic checks, then a different model that fails closed). The old maker-checker idea applied to agents. |
| Agency is not a yes 2026-10-08 | human-in-the-loop | Take the seven-level agency scale, from rubber stamp to the reviewer who fixes the source, and the line that below level 3 the person deciding carries the liability but has no real agency. |
| Agentics, ready to sell 2026-10-16 | getting-the-job | Take the market facts: Gartner expects over 40% of agentic projects cancelled by end of 2027; boutiques sell £2,500 audits and €2,900 sprints; a marketplace of consultants is forming, a door for newcomers. |
| Allowed is not right 2026-10-18 | controlling-your-agent | Take the Northstar example: the agent may refund up to $200, the customer is owed $120, it refunds $150, every control passes and the outcome is wrong. |
| Are hooks the sweet spot? Moving an agent's rules from hope into hooks, rule by rule, until enforcement moves off the machine 2026-10-14 | controlling-your-agent | Take 'hope-based security' and the count: of the site's eleven agent rules only one was kept by something other than the agent; four moved fully into hooks. |
| Attacks rewritten as mandates 2026-10-13 · in the map | security-agent-era | Take 'reach beyond mandate' as the shape of every supply-chain attack (28 incidents) and the fact that the next maintainer holding a publish token is an agent (4,746 PRs in eight weeks). |
| Brief the agent too 2026-10-11 | working-with-agents | Take the story: a founder gave an article to their agent, which saw the team defending against states while its backups were undone and reset the target to T3. Readers now read through agents. |
| Bringing my past work back 2026-10-18 · in the map | graphs-memory | Take the back-catalogue inventory (103 essays, 1,289 blog posts, 65 decks, 221 videos) and the point that work you cannot link is lost: talk to your future self. |
| A chat box on a site with no server, the plan, and the trade it makes 2026-08-27 | building-blocks | Take the three honest tiers for a chat box with no server (a local matcher, a key in the reader's browser, a vault app with a bridge) and the fact that only one of them is free. |
| Commander's intent is a behaviour policy 2026-10-14 | controlling-your-agent | Take 'agents do not go rogue; they follow instructions and fill the gaps' and commander's intent: write the objective and its limits together. |
| Before you give an agent a connector, give the connector a twin 2026-09-24 · in the map | controlling-your-agent | Take the three questions to ask before connecting anything: what did the agent do, what did it see when it did it, and which actions can you undo. A twin of the connector (a journal plus a replay) answers all three. |
| Custom UIs are not the exception 2026-10-01 | everyone-is-a-developer | Take the claim: a custom interface for each moment is now how interfaces get made, so building your own UI is normal, not exotic. |
| Every risk is already accepted. The only question is by whom, and for how long. 2026-09-24 · in the map | security-agent-era | Take the quote: "Every risk is already accepted. The only question is by whom, and for how long." |
| Firing journalists is the worst thing a publication can do 2026-10-12 · in the map | getting-the-job | Take the line "what you measure is what you become" and the reading of hiring decisions as a statement of where an employer is going, a lens a job seeker can use on any company. |
| Fix the page yourself 2026-10-17 · in the map | security-agent-era | Take the bookmarklet story from a security assessment: the business owner liked the quick UI more than the exploit; the next UX is the one the user customises. |
| From sending a file to sgit 2026-10-15 | be-a-founder | Take 'the technology worked; the use case was not there yet' ('users buy credits, test once, and don't come back') and the 22-minute feature; plus the record correcting memory. |
| Handing secrets to agents and browsers 2026-10-15 | security-agent-era | Take the rule 'the secret goes from the drop into the tool, never into the chat', and that an agent told something is secret sometimes refuses to use it. |
| Hope or enforcement 2026-10-08 | controlling-your-agent | Take the worked example: one customer-service agent in three designs (one broad model, a harness of tools, narrow agents behind a gateway), where most of the policy moves from hope to boundaries and the token cost falls. |
| How the newsletter is made 2026-10-18 | working-with-agents | Take the method: read every source in full (72,000 words), the build checks quotes word for word, the agent checks what no gate can; and the honest backlog. |
| I miss it, so it is working 2026-10-11 · in the map | be-a-founder | Take the rule that the people who decide the roadmap must use the product, and Dinis's admission that he failed this test with earlier startups. Direction from usage is the scarce part. |
| When journalism becomes intelligence, built 2026-10-15 | graphs-memory | Take 'as AI makes content production easier, the advantage comes from how we observe the world' and the decision to make records synthetic rather than fake real ones. |
| Twenty sites in fifteen days, and what that did to the writing 2026-08-26 · in the map | publish-in-public | Take the number: twenty sites in fifteen days, as a body of work split so each argument keeps its own history, with an index that starts from the reader's question. |
| Not a straight line 2026-10-12 · in the map | NEW:direction-is-the-scarce-skill | Take the story: the newsroom solved in two days a vault problem that six months of RiskMandate work had not, and "the thinking is never wasted when it is written down and open source" (the paused Cyber Boardroom and MyFeeds). |
| On the record 2026-10-18 | facts-and-the-record | Take the research point that real names did not make comments better, stable identities did, and signed, withdrawable professional opinions as the alternative to comments. |
| Open source is not free 2026-10-10 · in the map | open-source | Take the numbers: 6.99% of Homebrew macOS installs in 30 days (16.66% in a year) are on versions older than the newest three; a 20p card fee dominates any payment under about five pounds. |
| Our first PR in Ruflo 2026-10-17 | open-source | Take the first-contribution story: a one-file, 83-line documentation PR merged into a 9,625-commit project because it was exact about what it did not claim. |
| Pay after you read 2026-10-10 | making-of | Take the number: a working pay-on-demand publication in seven releases in 3 hours 40 minutes. This is the meter the book's reader uses, and an example of small releases that each grow from using the last one. |
| Paying after the value 2026-10-14 | be-a-founder | Take the gym figure ($17 a visit when $10 was available) and pay-after-the-value as a business-model idea a young founder can test cheaply. |
| A web application security challenge to Portugal, 2010 2026-10-16 · in the map | civic-and-history | Take what came true in 16 years (liability, disclosure, HTTPS) and the lesson that regulation, not fear, moved governments; answers Stories to tell and Civic responsibility. |
| Hacking Portugal, 2016 2026-10-16 · in the map | civic-and-history | Take 'to hack means to solve problems', the Portuguese Hacking Service for 15 to 21 year olds, and that agents make a small country leading in code more possible. |
| GenAI in Portuguese, 2024 2026-10-16 | everyone-is-a-developer | Take the quiz: the sixth most powerful programming language is Portuguese, because prompts are natural language; your own language is now a programming language. |
| Price it, then give it away 2026-10-02 | be-a-founder | Take the move: a price says what you think something is worth; giving it away at that price to people who know you is the cleanest test of whether you are right. |
| Re-anchoring 2026-10-10 | controlling-your-agent | Take the number: this site's session was summarised 18 times in a month, keeping about 1.7% each time. Rules have to be re-read from a file after each summary (a SessionStart hook), not remembered. |
| The SaaS apocalypse will be decided by inertia, not by AI 2026-09-21 | explorers-villagers-town-planners | Take the argument: AI is available to both sides, so inertia decides who wins (where a company sits on the evolution axis and how much past success it has to protect). This is why newcomers have room. |
| Search as navigation 2026-10-18 · in the map | graphs-memory | Take the taxonomy debt numbers (663 tags, 396 used once) and the hypothesis that a clean graph needs less model; answers the Google Search and StackOverflow outlines. |
| Seven vaults, one method 2026-08-19 | security-agent-era | Take the rule: classify the credential before it touches anything. Each step of the method was learned by getting something wrong first. |
| sgit is a standard, not a server 2026-10-16 · in the map | building-blocks | Take 'Git won not because of GitHub but because it is a standard'; answers the Git outline. |
| Six agents, one inbox 2026-09-29 | controlling-your-agent | Take the line: "the policy is exactly as strong as the row that says the agent has been told". Write the policy as a table with an enforcement column. |
| Sixteen notes I did not want to lose 2026-10-13 | graphs-memory | Take the moment notes in a browser became 'pieces of intelligence' worth keeping, and 'users will accept friction once when the value is clear; they will not accept losing their data'. |
| A skill behaviour policy 2026-10-14 | security-agent-era | Take the 899-skills measurement and the honest skill with the whole lethal trifecta (chat history in, HackerNews in, Slack out) that still needs a control; check skills before you install them. |
| Sovereignty is the ability to move 2026-10-12 | civic-and-history | Take the test: on a Wardley map, remove each dot and follow the blast radius up to the customer; sovereignty means being able to move, which is about playing where the ball is going. |
| Stop the bleeding, then ask for it back 2026-10-13 · in the map | be-a-founder | Take it as a worked example of a small agent-run service a freelancer can operate (agent drives the browser; the person types passwords and sends drafts; paid 10% of savings). |
| A supply chain of vaults 2026-09-27 | be-a-founder | Take the idea: small custom tools built once with GenAI, on open data, could bring the price of food down. An example of where a newcomer's explorer coding could matter. |
| System prompts as behaviour policies 2026-10-14 | controlling-your-agent | Take the count: Claude's published system prompt holds 82 rules, 36 could be held by the harness, 25 stay with the model; a system prompt is a light behaviour policy. |
| The agent is the reader 2026-10-11 | graphs-memory | Take the number: measured on five articles, the graph answers in 90% fewer tokens than the prose. |
| The agent team as it runs 2026-10-06 · in the map | working-with-agents | Take the picture: one person and twelve Claude agents, each a role on its own account, vaults as the only memory, a mailbox nobody sends from, identity and encryption as the strong barriers and policy as the weak one. |
| The AI Conference 2026, read for agent governance 2026-10-16 | facts-and-the-record | Take the fact-check: of six claims in a widely shared post three traced to sources and three did not; the best fact-checking was in a speaker's notes. |
| The controllable founder 2026-10-16 · in the map | be-a-founder | Take survivorship bias and the hypothesis that investors select for control (half of founders not CEO three years in); a caution for the Be a founder chapter. |
| The eyeballs get cheap 2026-10-14 · in the map | open-source | Take curl's year: AI review first drowned it in noise, then produced hundreds of real fixes; open source is about to become the most secure software. |
| The graphs books 2026-10-16 · in the map | making-of | Take the cautionary tale: three agent-written books (100,000 words in a day) stalled because the approve stage was one person; the making-of read better because it was written over a real record. |
| The Librarian's daily brief 2026-10-18 · in the map | working-with-agents | Take the daily brief pattern (what happened yesterday, threads left, at most three suggestions for where your leverage is) as a habit any newcomer can set up. |
| The minimum code below 2026-10-14 · in the map | security-agent-era | Take the question 'what is the minimum code below this that the layer above needs?' and the 428 of 1,500 files measurement; answers Docker and Linux outlines. |
| The policy draws the liability line 2026-10-13 | controlling-your-agent | Take the plain-English prohibitions list (do not send email directly, do not edit other agents' files, make no offers, stay inside the mandate) as a model for writing your agent's rules. |
| The policy guards the door; the command line uses the window 2026-10-17 | controlling-your-agent | Take the lesson in one line: a policy holds only on the paths that pass through its enforcer. |
| The records ran out 2026-10-12 | building-blocks | Take two facts from Ignacio Aldama's test: a fresh Claude Code session found the shared contract in two MCP calls, and a green build can hide tests that never ran. Feeds the testing gap. |
| The tokens hook 2026-10-15 | building-blocks | Take the cost fact: 99% of input from cache is the difference between about $170 and $1,950, and a model cannot see its own usage; know what your agents cost. |
| The value of constraints 2026-10-16 | open-source | Take 'nobody will buy it for its encryption' and constraints as the source of better features; open core is an evaluation version. |
| The vault is live 2026-10-12 · in the map | making-of | Take the timeline: from a design on the morning of 11 October to a live reader vault at 02:22 UTC on 12 October, as an example of how fast shipping is now. |
| Three companies, three LinkedIn pages 2026-10-18 · in the map | getting-the-job | Take the practice: drafting LinkedIn pages, profile and first posts in the open, every post an article already published; answers the LinkedIn outline. |
| Three policies for an agent 2026-10-17 | controlling-your-agent | Take the three-policy split in one line each: behaviour says what it may do, decision says who chooses, workflow says in what order. |
| Where is the why? A permission prompt asked me to decide, and kept the reason 2026-10-07 | human-in-the-loop | Take the story: a permission prompt that asked for approval without saying which session was asking, why, or what a no would do. Useful when teaching readers to read the prompts their own tools show them. |
| Who will game the reading meter? Eight kinds of reader, nine ways to cheat, and the risks that will actually happen 2026-10-10 | everyone-is-a-developer | Take the numbers: about 0.6% of internet users are developers, and 26% of OECD adults cannot use a computer. Use both to show how far "everyone who can brief an agent" reaches beyond the 0.6%. |
| Your agent's logs are a self-report 2026-10-16 | facts-and-the-record | Take the research figure: models that exploited a reward hack admitted it in under 2% of cases; an agent's logs are a self-report. |
Context: background for the agents writing the book (42)
| Article | Serves | Why |
|---|---|---|
| A book for one reader 2026-10-18 | making-of | Plans personalised, metered editions of this book; guides the agents producing editions, not a reader chapter. |
| A personal agent that keeps your secrets 2026-10-06 | security-agent-era | Background for the agents: the 2026 personal-agent products read for barriers and data sovereignty. |
| From me to you, in a link 2026-10-11 | making-of | Background for the agents: personal reading lists in a link, a mechanism a personalised edition could use. |
| Agent hooks in October 2026 2026-10-14 | controlling-your-agent | Technical survey of hooks across ten coding agents; background for the agents writing the controls chapter, with the takeaway 'guardrails, not walls'. |
| Agentic cost is an engineering problem 2026-10-12 | building-blocks | Background for the agents: agentic cost as the price of skipping villager discipline. A business plan, not student costs. |
| An open AI governance framework, and what its licence let us build 2026-10-07 · in the map | security-agent-era | Background for the agents: twenty AI governance controls under CC BY-SA, turned into a graph. Too enterprise for the reader. |
| Approve at altitude 2026-10-17 | controlling-your-agent | Decision-policy governance at enterprise scale; gives the agents the rule 'a lower level may narrow the one above, never widen it'. |
| Authority outside the model 2026-10-11 · in the map | controlling-your-agent | Background for the agents: Nadella's seven principles mapped to the mandate. Enterprise framing, cited in map.json. |
| Case law for agents 2026-10-17 | controlling-your-agent | Precedent-driven decision policies; background on how a decision earns its way into code. |
| Encrypted memory for agents that run somewhere else 2026-10-08 · in the map | graphs-memory | Background for the agents: sgit deployment patterns, and the rule of one vault per trust boundary. |
| Every link points back 2026-10-13 | facts-and-the-record | Links point back in time (551 of 566); background for how the book's chapters should carry forward links and corrections. |
| Generation Z Developers (AI edition) 2026-10-18 · in the map | everyone-is-a-developer | The edition's plan (keep the name, frame with explorers/villagers/town planners, 90 of 138 files outlines); the founding brief for the writing agents. |
| Hooks are kernel code 2026-10-14 | controlling-your-agent | Hooks maturity model and billing brief; gives agents 'treat a five-line hook like kernel code' and the three questions (power, trifecta legs, sandbox). |
| If somebody built a company on code review 2026-10-05 · in the map | graphs-memory | Background for the agents: graphs at every altitude, and code review as a science. Cited in map.json. |
| Inside the Northstar twin 2026-10-18 | controlling-your-agent | Walkthrough of a deterministic refund twin; agents can borrow its cases ($261 split into three to fit the limit) as exercises. |
| Move the cloud, not the code 2026-10-12 · in the map | civic-and-history | Background for the agents: cloud twins and sovereignty in practice. Cited in map.json. |
| One article, five readers 2026-10-10 · in the map | making-of | Background for the agents: the Librarian, Cartographer, Historian, Explainer and Storyteller roles, which can be reused to produce the book. |
| Policy as code needs a policy first 2026-10-14 | controlling-your-agent | Policy as code versus behaviour policy; gives agents 'a rule the agent can rewrite is a setting, not a control'. |
| Portugal and Europe, 2025 2026-10-16 · in the map | civic-and-history | Sovereignty and regulation as graphs; background for a civic-responsibility then-now, with 'sovereignty is the ability to move'. |
| Ten hard questions for RiskMandate, answered 2026-10-09 · in the map | controlling-your-agent | Background for the agents: mandate, grant, delta and barriers, the four objects, defined precisely. |
| Sell, don't issue 2026-10-16 · in the map | be-a-founder | Founder secondary-sale funding model; too advanced for a newcomer, background for the founder chapter. |
| Send an agent, not a spreadsheet 2026-10-05 | security-agent-era | Background for the agents: due diligence done by an agent instead of a questionnaire. |
| sgit in a CI pipeline 2026-10-14 · in the map | building-blocks | CI data patterns with vaults; background for the CI outline (artifacts 90 days, caches 10 GB, outputs 1 MB). |
| Short, punchy, and co-written with Claude 2026-10-18 · in the map | making-of | The book's style contract: short and punchy, credit Claude and show the memo-to-word ratio, meet the slop suspicion with evidence, include a Google Workspace / agentic email chapter. |
| Zoom into an agent's behaviour policy and you find the business logic 2026-10-07 | controlling-your-agent | Background for the agents: layered behaviour policies hold business logic that software used to enforce by leaving things out. |
| The briefs so far 2026-10-18 | working-with-agents | Inventory of 65 briefs (11 built, 21 partly, 32 not); background on briefs as the unit of work. |
| The editor's channel 2026-10-18 | making-of | Editor key and vault channel; background for how Dinis's comments reach the book's agents. |
| The identity we wanted to give the agents 2026-10-05 | security-agent-era | Background for the agents: why Google's terms stopped a Workspace identity per agent, and why login plus secrets is still hard. |
| The infographic bake-off 2026-10-10 · in the map | making-of | Background for the agents: which image model suits which job, for the book's figures and decks. |
| The investigation GitHub owes its customers 2026-10-05 | building-blocks | Background for the agents: platform outages as near misses that deserve an aviation-style inquiry. |
| The key is the billing system 2026-10-12 | building-blocks | Background for the agents: VoiceDebrief, a browser app on your own key, and provider terms that forbid reselling keys. |
| The Mandate Stack 2026-10-06 · in the map | controlling-your-agent | Background on the eight-layer production setup (about fifteen agents and one person) behind RiskMandate; useful to agents writing the agent-team chapters. |
| The same argument, in our words 2026-10-11 | controlling-your-agent | Background for the agents: a glossary between Nadella's vocabulary and RiskMandate's. |
| The site is an inbox 2026-10-12 | making-of | Background for the agents: the reader's marks become a graph the reader owns, and the brief for that design. |
| Three decisions, mapped 2026-10-17 | controlling-your-agent | Maps behaviour/decision/workflow on three scenarios with five states; background for the agents. |
| Two agents, one vault 2026-10-12 | working-with-agents | Background for the agents: a vault as the interface between two agents that never share a session. |
| The ultimate insider 2026-09-30 | controlling-your-agent | Background for the agents: agents as an insider threat that never scaled before, and constraining an agent in order to trust it. |
| Versions of versions 2026-10-18 | making-of | Versioning parts of a site independently; background for the book's two version streams. |
| Ways to read one article 2026-10-12 | making-of | Background for the agents: reading modes, with the finished piece kept at the centre. Relevant to the book reader. |
| Git for things you cannot put on GitHub 2026-08-25 · in the map | building-blocks | Background for the agents: sgit is git for things you cannot put on GitHub, on a server that holds only ciphertext. The vault the book uses. |
| Where the platform draws the line 2026-10-11 | controlling-your-agent | Background for the agents: the shared responsibility model for agents (Microsoft Execution Containers), and the business logic that sits above it. |
| Where the vault keys live 2026-10-10 | security-agent-era | Background for the agents: key management, append lanes, and the weakest step (handing a key to another agent). |
Skip: not for this reader (44)
| Article | Serves | Why |
|---|---|---|
| A directory of people and their keys 2026-10-18 | sgit PKI product brief, niche. | |
| A link to a persona 2026-10-10 | None | newsroom persona link feature |
| A Mac of the agent's own 2026-10-09 | None | business plan for agent Macs |
| A meter in the browser 2026-10-10 | None | reading meter experiment design |
| A section for publishers 2026-10-18 | Newsroom audience sections, site plumbing. | |
| Agent workflows on ephemeral vaults 2026-10-14 | sgit multi-agent architecture recipes. | |
| A locked-down desktop for an agent, by the minute, is still hard to rent 2026-10-07 | None | agent desktop rental market survey |
| Cedar and the policy vault 2026-10-14 | Cedar policy-language product brief. | |
| Chainguard, mapped 2026-10-14 | Company map, supply-chain specialist depth. | |
| Dark mode, and four more 2026-10-13 | Newsroom theming implementation. | |
| Four frameworks, one agent 2026-10-13 | AI risk framework comparison, governance depth. | |
| The future of news is the story vault, not the paywall 2026-09-22 | None | news business model thesis |
| Going live with the reading meter 2026-10-10 | None | reading meter launch hypotheses |
| The reader was always the product 2026-09-29 | None | history of news economics |
| I just want to buy the article 2026-10-17 | Publisher micropayment economics. | |
| Inside the Mesa Ranch desk 2026-10-15 | Walkthrough of a journalism vault demo. | |
| Liquid content needs water 2026-10-07 | None | newsroom workflow, journalism industry |
| One vault, ten agents, 35,000 objects 2026-10-16 | sgit performance measurements. | |
| Pay to keep your persona 2026-10-10 | None | reader persona pricing argument |
| The proof is two clicks behind the claim, what the homepage gets wrong, and the fix 2026-09-07 | None | sgit homepage critique |
| The proof moved up, the homepage after the rebuild, next to the before pictures 2026-09-07 | None | sgit homepage rebuild notes |
| Research, delivered as a vault 2026-10-16 | Research-firm product idea. | |
| RFC 0001 2026-10-09 | None | sgit cryptography design RFC |
| Scion and sgit 2026-10-16 | Version-control internals comparison. | |
| Scrolled is not read 2026-10-16 | Reading-meter engineering brief. | |
| secrets.sgit.ai on public keys 2026-10-15 | secrets.sgit.ai product design. | |
| sgit in GitHub Actions, side by side 2026-10-14 · in the map | sgit CI showcase brief. | |
| Story vault underneath, Reader Skills on top 2026-10-07 | None | local journalism product argument |
| The AI governance stack, as a graph 2026-10-09 | None | governance stack reply, enterprise |
| The article remains a product 2026-10-11 | None | newsroom outputs, journalism industry |
| The bridge, followed to the end 2026-10-07 | None | story vault simulation, journalism |
| The deck I could not download 2026-10-06 | None | presentation platform business plan |
| The homepage is a vault 2026-10-14 | Homepage-as-vault engineering brief. | |
| The passkey opens the vault 2026-10-11 | None | vault passkey design detail |
| The scripted talk 2026-10-15 | Talk-player tooling brief. | |
| The vault crosses Chrome profiles 2026-10-13 | Reader-vault feature discovery note. | |
| The vault, opened 2026-10-16 | Reader-vault release debrief. | |
| The waiting room knew first 2026-10-08 | None | local news story, hospital outage |
| The wall under the reply 2026-10-04 | None | email format experiment |
| The wire as a graph 2026-10-14 | Newsroom wire data design. | |
| Sixteen thousand fetches, ten clicks, and a token bill nobody is sending 2026-09-28 | None | publisher payment from answer engines |
| What if an agent loses its workspace? Mondweep Chakravorty's SGit × Ruflo proof of concept, step by step 2026-10-17 | Ruflo proof-of-concept walkthrough. | |
| When is the vault actually used? In Mondweep Chakravorty's SGit × Ruflo run, the encrypted vault is written or read in three steps of twelve, on one machine, inside the adapter, and why that is the honest shape of a recovery copy 2026-10-17 | Ruflo proof-of-concept analysis. | |
| Your reading, your controls 2026-10-17 | Reading-meter feature debrief. |