Writing
What I learned shipping, written down.
18 articles in four series: building products with AI agent teams, the gates between an agent and production, conversion experiments on real funnels, and design systems that match the code. Each series has a place to start.
Kit7 templates, checklists and prompts from the method, ready to copy: CLAUDE.md, tickets, Notion board, release gates, MVP scope, checkout audit.Open the kit →Building with AI agents 6
November 4, 2026 / 12 min / start hereHow to build an MVP with AI agents: the full method, from scope to production
A step by step method for building a real MVP with AI coding agents: how to scope it, the project rules file, tickets agents cannot misread, the team structure, the gates before production, and what it costs. From a build that shipped 445 PRs in seven weeks.
November 4, 2026 / 7 minHow to scope an MVP for AI agents: the one workflow rule and the list of what you will not build
Scoping an MVP that AI coding agents will build is different from scoping one for a team. Four parts: the thesis, one workflow, invariants marked by where they are enforced, and the explicit list of what is not built. With a worked example.
November 4, 2026 / 8 minCLAUDE.md example: the project rules file that made 445 pull requests mergeable
What goes in a CLAUDE.md, with a neutral example: thesis, fixed stack, numbered hard rules with the incident behind each, what counts as proof, task states and an escalation format. Why rules without reasons get argued with and rules with reasons get followed.
November 4, 2026 / 7 minWriting tickets for AI agents: the template that stops them inventing requirements
A ticket template for AI coding agents: a measurable acceptance criterion, the explicit out of scope list, decisions already taken, and frontmatter a script can validate. The three fields that prevent the most expensive failures, with an example ticket.
November 4, 2026 / 8 minAI agent team structure for a solo builder: roles, handoffs and who merges
How to structure a team of AI coding agents when one human runs it: a lead that decomposes but does not merge, workers with a mandate and a file boundary each, a reviewer, a security agent and a verifier that can only read. Why the orchestrator never writes code and no agent holds production credentials.
November 4, 2026 / 9 minWhere AI coding agents fail on real products: the eight failures I hit and the gate for each
Eight failure patterns from seven weeks of building a production product with AI coding agents, each with the gate that now catches it: the green check that did not run, the self contradicting ticket, the test that protected a bug, the fact declared instead of produced, and four more.
AI agent teams 7
November 4, 2026 / 9 min / start hereAI agent teams for software development: what actually works in 2026
How to run a team of AI coding agents that ships to production: roles with mandates, isolation, gates, a decision log, and the numbers from seven weeks of doing it.
November 5, 2026 / 9 minNotion for AI agent teams: the board is the control plane, the repo is the source of truth
How to run an AI coding agent team from a Notion board without losing control: instructions only in the repo, decisions in their own database, cards drafted by the orchestrator and owned by a person, and Done only after a merged pull request with a reviewer PASS.
November 4, 2026 / 7 minHuman in the loop for coding agents: the two gates that matter
Where a person must sit when AI agents write and ship code: the decision gate (one id, one letter) and the release gate (one approval with a security report). Everything else can be automated.
November 4, 2026 / 7 min445 PRs in seven weeks: what a one person team with AI agents actually looks like
Real numbers from building Acervo with an AI engineering team: 484 pull requests, 1,456 CI runs, 15 gated releases, and what the human did all day.
November 4, 2026 / 6 minWhy a green check is not proof: the gates between an AI agent and production
On October 6 a navigation redesign merged on 8 of 8 green checks. The e2e job had not run. What that taught me about verification when agents write the code.
November 4, 2026 / 7 minRow level security with AI agents: tests that run as the user who must not get in
How to keep multi tenant isolation honest when agents write the migrations: every rule lives in the database, every policy has a test that runs as the wrong user, and the reviewer breaks a policy on purpose.
November 4, 2026 / 6 minFrom ticket to production in five hours: a release process for a one person team
One real change, end to end: ticket at 14:35, pull request at 16:56, fourteen checks, pen test PASS, merge, promotion at 19:50, served at 19:54. The process that makes that routine and safe.
Conversion and funnels 2
November 4, 2026 / 6 minCheckout conversion: the ten experiments behind +11%
At GetYourGuide, 38% of converters left checkout and came back with no re-entry experience. Ten experiments, four EU markets, +11% conversion, and the ones that lost.
November 4, 2026 / 6 minInstrument the gaps, not the stages: funnel analytics that find the real leak
Stage by stage funnels hide where people actually drop: the back button, the tab switch, the return visit. How to rebuild a funnel per user and find the leak nobody measured.
Design systems 1
November 4, 2026 / 7 minA Figma design system that mirrors the codebase: tokens first, then components, then screens
How to build a design system so that Figma and code say the same thing: a single token source exported to CSS by script, a component library that consumes it, and checks that fail when a raw value sneaks in.
Product and consulting 2
November 4, 2026 / 6 minHow much does it cost to build an MVP with AI agents? Real numbers from one build
What a production MVP built with an AI agent team actually costs: the agent bill, the infrastructure, the human time, and where the money really goes. Numbers from seven weeks on Acervo.
November 4, 2026 / 5 minFractional product manager vs product consultant: which one do you actually need?
A consultant diagnoses and recommends. A fractional product manager owns outcomes inside your team for a period. How to tell which one your company needs, and what each should cost you.
Nothing matches. Try a shorter word, or ask me.