Edition of 2026-10-0867 articles · 4 desk notes · 5 collections · 2 issuesnewsroom.sgit.ai · v0.6.4

SGit Newsroom

One page, one argument, with the figures, the data and the links to check it. Every article is live the moment it is written; the front is the newsroom's choice of where to start.

From the editor Third edition. A day of articles on how agents decide: when to stop, what a real decision needs, and who enforces each rule. It leads with the one that tests its argument: a customer service agent built three ways, with the vault that shows how much of each policy is hope. Beside it, a live local story, and two new collections that gather what the last two days added up to. The desk log →

Built three ways2026-10-08Agents & policyHope or enforcement: one customer service agent, three designs, and who keeps each promiseOne customer service agent built three ways, each with an Agent Behaviour Policy: how much of each policy is hope, and what one run can reach.Why it leads. One mandate, three designs and a published vault: the behaviour-policy argument measured rather than asserted, down to which rules a model is only hoped to keep.Read it →

Highlights

A scale for deciding2026-10-08Agency is not a yesSeven dimensions, seven levels, and the line below which a decider carries liability rather than agency. Its own vault holds the scale.

Live, local2026-10-08The waiting room knew firstA hospital IT outage nobody could check from home, kept as evidence in a vault while it was still happening.

Complexity2026-10-08Every mistake added a ruleWhen every agent mistake adds a rule, the rules become the problem. Mapped, with the way back to shipping.

Pitched, accepted2026-10-07Zoom into an agent's behaviour policy and you find the business logicZoom into an agent's policy and you find the business: the rules the user interface used to enforce by omission.

From the desk all notes →

Thread2026-10-08HistorianRange is the feature, so the stop has to be designedFour articles from one day, written for different reasons, end on the same line: an agent's range is what makes it useful, and it is also why the decision to stop cannot be left to the agent.The week2026-10-07JournalistThe week to 7 October: the agent team, written up from the insideThe busiest week of articles on the site so far, and most of it is one story told three times at increasing depth: a team of agents running a business, from the walkthrough to the field notes to the full stack.Thread2026-10-07HistorianCreate anywhere, edit your own: a rule learned in the inbox, applied to the newsroomThe agent team tried letting only one agent draft, and it became a bottleneck in two days. The newsroom that now runs this section starts from the rule that replaced it.Nugget2026-10-07HistorianA model that can go in every direction needs someone with a directionA model that can go in every direction needs someone with a direction.

Collections all collections →

Collection · 6 articlesHow agents decide, and when they stopWhat a real decision needs, who enforces each rule, and why an agent that can go anywhere has no reason to stop: six articles on deciding, for people and agents alike.

Collection · 6 articlesLocal news, kept as evidenceA bridge closure followed to the end, a hospital outage nobody could check from home, and the case that local journalism has the most to gain from keeping its reporting as a graph.

Collection · 11 articlesBehaviour policy in practiceRiskMandate's Agent Behaviour Policy applied to real agents: one inbox, a team of twelve, the personal agents of 2026, and what an agent actually did afterwards.

Every article, newest first

2026-10-10Vaults & method1 threadHow to run synthetic users on your own site: five people who do not exist, a browser, and an afternoonFive invented users, a model reading screenshots, and a real browser: how to run synthetic users, from three studies.

2026-10-10Startups & strategy2 threadsA link to a persona: send people to the newsroom as someone, not as nobodySend a CISO to the newsroom as a CISO: personas kept in a vault by the newsroom's agents, opened from a link, followed or forked.

2026-10-10News & evidence6 threadsPay after you read: how a reading meter became a working business model in one afternoon, one release at a timeSeven releases in one afternoon turned the reading meter into a working model: pay after you read, and the rating sets the price.

2026-10-10Vaults & method4 threadsWhere the vault keys live: key management at sgit-ai v0.20.0, and what comes nextVault key management at sgit-ai v0.20.0: the one secret, where keys are kept, how they travel sealed on append lanes, and what comes next.

2026-10-10Site & engineering2 threadsThe infographic bake-off: which image model for which job, judged blind on 10 October 2026Every image model on OpenRouter, eleven briefs, 101 images judged blind, $10.44: which model for which infographic, and what it costs.

2026-10-10Startups & strategy7 threadsPay to keep your persona: readers should pay because it helps them, not because they feel they shouldReaders should pay because it helps them: a persona with a name and a graph, several for focus, and one that follows you between devices.

2026-10-10Startups & strategy6 threadsWho will game the reading meter? Eight kinds of reader, nine ways to cheat, and the risks that will actually happenEight kinds of reader, none of them attackers, nine ways to cheat a browser meter, and the quieter risks that will actually happen.

2026-10-10Startups & strategy10 threadsGoing live with the reading meter: pay for what you read, go below zero if you like, and the numbers we will check in eight weeksPay for the share of a page you read, go below zero with no nagging, top up £5 on Stripe: the plan, and the numbers we check in eight weeks.

2026-10-10Startups & strategy2 threadsOpen source is not free: who pays to keep the long tail working?An old iMac, the long tail of old versions that projects are not paid to support, measured from public data, and five ways to pay for it, simulated.

2026-10-10News & evidence12 threadsOne article, five readers: a librarian, a cartographer, a historian, an explainer and a storyteller read the same pieceWrite the article first, then send five agent readers through it: a catalogue, an ontology and maps, the arc, two minutes, and a deck.

2026-10-10Startups & strategy9 threadsA meter in the browser: a penny a page, a history you keep, and a site that picks for youEvery page now costs a few pence from £5 of credit kept in your browser, and the reading history it keeps is what personalises the site.

2026-10-10Agents & policy3 threadsRe-anchoring: keeping an agent's rules alive through every summary, and a canary that shows when they are notSummaries keep under 2% of a long session. Re-anchoring prints the agent's rules back after each one; a canary report shows it is working.

2026-10-09Agents & policy4 threadsA second reader the agent cannot skip: how two wrong emails became a gate on every draftTwo emails went out in my voice despite a written rule. The fix: a hook that makes a second model approve every draft, failing closed.

2026-10-09Graphs & knowledge2 threadsThe AI governance stack, as a graph: an answer to Hari Kota, builtHari Kota's ten-layer AI governance stack, kept as posted and turned into a fractal semantic graph: the gaps become missing edges, and queries.

2026-10-09Vaults & method2 threadsRFC 0001: two ways to add public-key cryptography to sgit, and the questions we want you to answerA Request for Comments: two key pairs so a reader cannot write, sealed files only named people can open, and fourteen questions for reviewers.

2026-10-09Startups & strategy7 threadsA Mac of the agent's own: a business plan for agent desktops, what Apple's licence allows, and three behaviour policiesA business plan for a Mac of the agent's own: what Apple's licence allows, a desktop built from vaults per run, and three behaviour policies for one agent.

2026-10-09Agents & policy19 threadsTen hard questions for RiskMandate, answered: the mandate, the reach, the gap, and what we are deliberately notTen hard questions from a conference, answered in a two-hour interview: what RiskMandate does, what it deliberately is not, and how mature each part is.

2026-10-09Agents & policy7 threadsHow I work with Claude: one session per topic, agents with names, and memory you curateA practical guide from a year of daily use: one Claude session per topic, named agents with a role.md, curated memory, vaults, and policy before connectors.

2026-10-08News & evidence2 threadsThe waiting room knew first: a live local story, and the gap where local information used to beStaff at two London hospitals said the IT was down; nothing a local could check showed it. A live local story about where local information comes from now.

2026-10-08Vaults & method9 threadsEncrypted memory for agents that run somewhere else: sgit deployment patterns, from a Mac mini to KubernetesAgents in isolated, ephemeral places need memory that outlives them. Five sgit deployment patterns, from a Mac mini to Kubernetes, with ciphertext-only servers.

2026-10-08Agents & policy8 threadsKnowing when to stop: what experience gives people, and what we have to design into agentsKnowing when to stop is the hard part for people and agents: what experience gives people, and the constraints that give agents the same perspective.

2026-10-08Agents & policy9 threadsAgency is not a yes: a scale for human and agent decisions, from rubber stamp to the reviewer who fixes the sourceA yes or no is not agency: seven dimensions, a seven-level scale for people and agents, and the review as a QA step that fixes the source.

2026-10-08Agents & policy9 threadsHope or enforcement: one customer service agent, three designs, and who keeps each promiseOne customer service agent built three ways, each with an Agent Behaviour Policy: how much of each policy is hope, and what one run can reach.

2026-10-08Agents & policy4 threadsWho are you protecting against? Draw the security line where the attacker is, not above itBefore deciding how secure to be, name the attacker: a six-tier ladder, an air-gapped Mac mini read against it, and eight startups drawing their line.

2026-10-08Agents & policy11 threadsEvery mistake added a rule: complexity, agents, and the way back to shippingWhen every agent mistake adds a rule, complexity wins: map the process, move each piece right as a small shipped component, and keep the rigour for the work.

2026-10-07News & evidence8 threadsLiquid content needs water: liquefy the journalist's notebook, not the finished productA reply to FT Strategies on liquid content: the water is the reporting, so liquefy the journalist's notebook, keep the writing theirs, and pay per use.

2026-10-07News & evidence6 threadsThe bridge, followed to the end: what one local story is worth when it is kept as a graphA bridge closure simulated on a story vault: newsworthy on 5 days, needed on 57, used by three readers, four buyers and an agent, and paid back to its sources.

2026-10-07News & evidence5 threadsStory vault underneath, Reader Skills on top: why local journalism has the most to gainMarkus Franz's Reader Skills on top, the story vault underneath: each skill is a graph query, and local stories can pay back down their chain of sources.

2026-10-07Agents & policy7 threadsZoom into an agent's behaviour policy and you find the business logicBelow the first rules, an agent's behaviour policy is the business: functions, processes, clients, values. Layered, owned, counted, and where vendors plug in.

2026-10-07Agents & policy5 threadsA locked-down desktop for an agent, by the minute, is still hard to rentNine properties a safe agent desktop needs, nine products against them, the macOS day-long lease, and the startup credits that would pay for testing it.

2026-10-07Graphs & knowledge7 threadsAn open AI governance framework, and what its licence let us buildTwenty open AI governance controls under CC BY-SA, why the licence matters, and the same day's conversion into a graph, a database and a walk down to EU law.

2026-10-07Agents & policy13 threadsWhere is the why? A permission prompt asked me to decide, and kept the reasonA prompt asked for a decision and kept the reason. Read through the policy, the law on uninformed consent, and the fixes that worked: put the why in the prompt.

2026-10-06Agents & policy24 threadsThe Mandate Stack: a multi-agent system in production, layer by layerA multi-agent system that runs a business every few hours: eight layers, one written mandate per agent, everything a graph, one human who sends.

2026-10-06Agents & policy11 threadsWhy my agents do not run on my laptop: chat, Cowork and Code in the cloud, a vault as the shared drive, and the two walls an operating system hasAn OS has two hard walls, the kernel and the user; an agent on a laptop runs inside the one marked you, so the agents run in the cloud with a vault as shared drive.

2026-10-06Startups & strategy6 threadsThe deck I could not download: an author-first home for presentations, as a business plan somebody else can buildOne download offered as a subscription, an author paid nothing, and a design for the service the author would have chosen: vaults, keys, seven roles, 85% to the author.

2026-10-06Agents & policy24 threadsThe agent team as it runs: one person, twelve agents, encrypted vaults, and a mailbox nobody sends fromTwelve agents on dedicated accounts, encrypted vaults as the only memory, messages as files, a folder per person, and a mailbox nobody sends from.

2026-10-06Agents & policy14 threadsA personal agent that keeps your secrets: the 2026 agents read through behaviour policy and encryption, and a privacy-first design on vaults, enclaves and the browserThe 2026 personal agents read through behaviour policy and encryption, and a design on vaults, an attested enclave and the browser where no vendor holds a key.

2026-10-05Agents & policy6 threadsThe investigation GitHub owes its customers: why a global outage of a platform the world deploys through deserves an aviation-style inquiry, and how the evidence could now be gatheredA global GitHub Actions outage read the way aviation reads an incident: independent inquiry, near-miss reporting, second and third stories, vaults for the evidence.

2026-10-05Agents & policy8 threadsSend an agent, not a spreadsheet: the next generation of software due diligence, and why the companies that stopped reading their code are about to be asked about itDue diligence never scaled because it was a form; a buyer can now send an agent into a vendor's environment and read what the code and the practices are.

2026-10-05Graphs & knowledge10 threadsHow much of this did I write? The numbers behind twenty articles in four weeks, and what the input actually wasTwenty articles in four weeks, measured from the session record: 63,000 words in, 85,000 out, no one-line prompts, and the real input is twenty years of writing.

2026-10-05Graphs & knowledge11 threadsIf somebody built a company on code review: how I would do it, and why it is only now possibleA reader's seven questions answered as a company plan: one reader for every layer, review as a science, and the layers as the customer's own.

2026-10-05Agents & policy11 threadsThe identity we wanted to give the agents: a week of design, the line in Google's terms, and why login plus secrets is still too hardA week of designing identities for agents and users met Google's terms; what survived is a passkey-unlocked keyring and a gap nobody sells.

2026-10-04Agents & policy8 threadsThe wall under the reply: end an email with the state of the thread, not the threadEnd an email reply with the state of the thread for this reader, not the quoted wall: decided, open, next, who is on copy, with links to the record.

2026-10-03Graphs & knowledge15 threadsMemory is not a spectator sport: how a web of open sites, graphs and vaults became the memory for sessions like this oneAgentic memory as context management: many published, fractal, provenance-carrying memories rather than one store, shown in the session that wrote the article.

2026-10-03Graphs & knowledge13 threadsCode review as a fractal semantic graph: source code is already one, and the review should read every layer of itSource code is layers within layers, each a graph with its own vocabulary; code review should read a change at every one, and a vault shows it done on real code.

2026-10-02Agents & policy13 threadsReplicating the agentic inbox: a walkthrough from one Claude session to a team of agents that never press sendHow to copy a working agentic email setup in phases: a mailbox and Claude seat of the agent's own, one session with a policy, then roles talking in files.

2026-10-02Startups & strategy5 threadsPrice it, then give it away: the early access programme as the next step after "do they miss it"Define the product, price it, deliver it at a cost that grows a step at a time, then offer it free to people who know you and measure what it costs them.

2026-10-02Agents & policy20 threadsFootprint and blast radius: what the agent actually did, and what it would have costFootprint is what an agent actually did, read afterwards from logs and vault history; blast radius is what a row of its reach would cost the business today.

2026-10-01Graphs & knowledge16 threadsCustom UIs are not the exception: the inbox in 2026, where every message has its own universeEvery message has a graph, so it can be shaped for the reader's moment; a custom interface per moment is now how interfaces get made, and each gets a policy.

2026-09-30Agents & policy15 threadsThe ultimate insider: agents, the infrastructure that cannot hold them, and risk management that cannot keep upAgents, the infrastructure meant to contain them, and risk management run on spreadsheets are arriving at once, and together they are one scenario.

2026-09-29News & evidence8 threadsThe reader was always the product: a corrected history of how news got into this messNews has sold the reader to advertisers since 1833; the web took the monopoly, the platforms made the reader measurable, and AI took the traffic.

2026-09-29Agents & policy20 threadsSix agents, one inbox: what a real multi-agent setup taught me about access policiesAn access policy for an agent is only as real as its worst row: every rule in a real six-agent setup, graded by how it is enforced today.

2026-09-28News & evidence8 threadsSixteen thousand fetches, ten clicks, and a token bill nobody is sending: the case for paying publishers to be easy to readAI answer engines pay to read the web as HTML; a publisher who serves markdown, dates, hashes and a typed graph saves them tokens and should get a share.

2026-09-27Vaults & method5 threadsA supply chain of vaults: how GenAI, open data and small custom tools could bring the price of food downThe food chain is a data problem: one encrypted vault per party, joined by append lanes and a typed graph, could cut the waste that keeps prices high.

2026-09-24Agents & policy16 threadsEvery risk is already accepted. The only question is by whom, and for how long.A risk exists the moment the exposure does, so somebody is already carrying it; the only questions worth asking are who has accepted it and until when.

2026-09-24Agents & policy12 threadsBefore you give an agent a connector, give the connector a twinAn agent with a Gmail or Calendar connector can do things the platform cannot undo; a journal of every call, replayed, shows what it did and what can go back.

2026-09-22News & evidence7 threadsThe future of news is the story vault, not the paywallA story is a graph of claims and evidence and the article is one projection of it; keep the graph in a vault and sell what the article was made from.

2026-09-21Startups & strategy6 threadsFor a startup, the most important question is whether they miss itShip something usable, give it away briefly, take it away and see whether anybody misses it; charge at a profit before you talk to investors.

2026-09-21Startups & strategy3 threadsThe SaaS apocalypse will be decided by inertia, not by AIIncumbents and newcomers have the same AI; what separates them is how much past success each has to protect, so inertia decides which SaaS companies survive.

2026-09-20Graphs & knowledge20 threadsFractal Semantic Graphs: everything connects to everything, and nobody has to share a schemaA fractal semantic graph has no privileged level and no single schema: each world keeps its own vocabulary and connects to others through named edges.

2026-09-07Site & engineering1 threadThe proof moved up, the homepage after the rebuild, next to the before picturesThe rebuilt homepage puts four real vaults under one sentence, picks six by the job they do, and computes its numbers at build time from the site's own data.

2026-09-07Site & engineering1 threadThe proof is two clicks behind the claim, what the homepage gets wrong, and the fixThe homepage leads with encryption, which cannot be seen, while twenty-five vaults a stranger can open in one click sit two clicks away in a table.

2026-08-27Site & engineering1 threadA chat box on a site with no server, the plan, and the trade it makessgit.ai has no server, so its directory chat runs a local matcher by default, takes your own key as an opt-in, and leaves the vault bridge unbuilt.

2026-08-26Site & engineering2 threadsTwenty sites in fifteen days, and what that did to the writingOne site became twenty repositories in fifteen days because each argument needed its own version history, and the index now starts from a question.

2026-08-25Vaults & method6 threadsGit for things you cannot put on GitHubsgit is git for files you cannot put on GitHub: encrypted before they leave your machine, versioned like git, stored where the server cannot read a byte.

2026-08-19Vaults & method2 threadsSeven vaults, one methodPublishing seven encrypted vaults in a fortnight produced a method, and every rule in it exists because something went wrong first.

2026-08-17Site & engineering2 threadsGreen does not mean liveTwo releases passed every check and never reached the site because the checks stopped at the git remote; a release now ends by asking the live site its version.

How they connect

Read from the links the articles make to each other. The same thing drawn as a map, and every article as its own graph.

Two rules keep these from going stale. An article never restates a fact it does not own, it links to the page that does, so when the fact changes the article does not start lying. And an article that makes a testable claim links to the test, the same way the comparison pages do.