https://laimonade.com/blog Notes on product ownership for teams running coding agents. en https://laimonade.com/blog/spec-driven-development-and-the-backlog https://laimonade.com/blog/spec-driven-development-and-the-backlog Thu, 08 Oct 2026 00:00:00 GMT Spec-driven development describes one change in depth. It does not say what comes next, what blocks it, or whether the last spec was ever met. https://laimonade.com/blog/tech-lead-as-product-owner https://laimonade.com/blog/tech-lead-as-product-owner Thu, 08 Oct 2026 00:00:00 GMT Tech lead as product owner works for a few weeks. Then one person writes the criteria and reviews the work, and nobody checks the criteria. https://laimonade.com/blog/engineering-metrics-when-agents-do-the-building https://laimonade.com/blog/engineering-metrics-when-agents-do-the-building Wed, 07 Oct 2026 00:00:00 GMT Engineering metrics still work once agents build, but the time inside them moves to waiting for decisions and review. What to split, add and demote. https://laimonade.com/blog/product-roadmap-when-building-gets-cheap https://laimonade.com/blog/product-roadmap-when-building-gets-cheap Wed, 07 Oct 2026 00:00:00 GMT A product roadmap used to be rationed by engineering capacity. With coding agents almost anything fits, so the roadmap has to say what not to build. https://laimonade.com/blog/definition-of-ready-for-coding-agents https://laimonade.com/blog/definition-of-ready-for-coding-agents Tue, 06 Oct 2026 00:00:00 GMT A definition of ready was called a stage-gate when people could ask questions mid-sprint. A coding agent never asks, so the gate is where they get asked. https://laimonade.com/blog/engineering-team-structure-when-agents-write-the-code https://laimonade.com/blog/engineering-team-structure-when-agents-write-the-code Tue, 06 Oct 2026 00:00:00 GMT Engineering team structure changes less than the headlines say. The roles stay; the hours move to specifying work and judging what comes back. https://laimonade.com/blog/product-owner-vs-product-manager-when-agents-build https://laimonade.com/blog/product-owner-vs-product-manager-when-agents-build Tue, 06 Oct 2026 00:00:00 GMT Product owner vs product manager: one decides which problems matter, the other turns decisions into buildable work. Agents make the second job grow. https://laimonade.com/blog/shadow-ai-coding-tools-in-a-rollout https://laimonade.com/blog/shadow-ai-coding-tools-in-a-rollout Tue, 06 Oct 2026 00:00:00 GMT Shadow AI coding tools are usually in the codebase before any policy is. A ban hides them; the people using them are a rollout's best evidence. https://laimonade.com/blog/best-ai-coding-agent-for-a-team https://laimonade.com/blog/best-ai-coding-agent-for-a-team Mon, 05 Oct 2026 00:00:00 GMT The best AI coding agent for a team is the one whose working shape fits how work is handed over. Three properties to compare before any benchmark. https://laimonade.com/blog/prompt-injection-through-tickets-and-issues https://laimonade.com/blog/prompt-injection-through-tickets-and-issues Sun, 04 Oct 2026 00:00:00 GMT AI agent prompt injection needs no exploit, only a sentence in an issue or a README. Filters miss it, so limit what the agent can reach instead. https://laimonade.com/blog/user-acceptance-testing-when-an-agent-built-the-feature https://laimonade.com/blog/user-acceptance-testing-when-an-agent-built-the-feature Sat, 03 Oct 2026 00:00:00 GMT User acceptance testing asks whether the criteria were the right ones. An agent can show its criteria pass; only the person who owns them can answer that. https://laimonade.com/blog/git-worktrees-for-running-coding-agents-in-parallel https://laimonade.com/blog/git-worktrees-for-running-coding-agents-in-parallel Fri, 02 Oct 2026 00:00:00 GMT Git worktrees for AI agents give each task its own checkout. Ports, databases, .env files and the stash are not isolated, and that is where runs collide. https://laimonade.com/blog/how-to-write-a-bug-report-for-a-coding-agent https://laimonade.com/blog/how-to-write-a-bug-report-for-a-coding-agent Fri, 02 Oct 2026 00:00:00 GMT A bug report for a coding agent is a failing check, not a symptom. Reproduce it, name the last good version, and make the reproduction the criterion. https://laimonade.com/blog/when-a-coding-agent-gets-stuck https://laimonade.com/blog/when-a-coding-agent-gets-stuck Fri, 02 Oct 2026 00:00:00 GMT When a coding agent gets stuck, the right move is usually to stop and say so. A clever workaround is the outcome to design against. https://laimonade.com/blog/sprint-goal-when-agents-do-the-building https://laimonade.com/blog/sprint-goal-when-agents-do-the-building Thu, 01 Oct 2026 00:00:00 GMT When agents clear scope in days, the sprint backlog becomes a queue. A sprint goal decides what gets added, what gets dropped, and what counts. https://laimonade.com/blog/sprint-retrospectives-when-agents-did-the-building https://laimonade.com/blog/sprint-retrospectives-when-agents-did-the-building Thu, 01 Oct 2026 00:00:00 GMT A sprint retrospective with coding agents sorts rework by cause and spends the hour on the biggest pile. Agent speed is not a retrospective topic. https://laimonade.com/blog/tests-written-by-the-agent-that-wrote-the-code https://laimonade.com/blog/tests-written-by-the-agent-that-wrote-the-code Thu, 01 Oct 2026 00:00:00 GMT AI-generated tests prove something about a change only if they fail without it. Run the agent's new tests against the old code before trusting them. https://laimonade.com/blog/teaching-a-team-to-write-executable-items https://laimonade.com/blog/teaching-a-team-to-write-executable-items Wed, 30 Sep 2026 00:00:00 GMT Executable backlog items are not taught by a style guide. Hand each person their own item unedited to an agent, and let the result do the teaching. https://laimonade.com/blog/when-an-agent-rollout-should-stop https://laimonade.com/blog/when-an-agent-rollout-should-stop Wed, 30 Sep 2026 00:00:00 GMT Stop an AI rollout when the constraint you set out to move has not moved after a fair trial. Without a stated constraint, you cannot stop honestly. https://laimonade.com/blog/measuring-whether-coding-agents-are-working https://laimonade.com/blog/measuring-whether-coding-agents-are-working Tue, 29 Sep 2026 00:00:00 GMT Three coding agent metrics survive contact with a real quarter: specification rate, rework caused by the item, and ready-to-accepted time. https://laimonade.com/blog/who-resists-coding-agents-and-why https://laimonade.com/blog/who-resists-coding-agents-and-why Tue, 29 Sep 2026 00:00:00 GMT Developer resistance to AI tools is usually a specific objection wearing a general one. Four of them, and what each is actually asking for. https://laimonade.com/blog/rolling-out-coding-agents-to-a-team https://laimonade.com/blog/rolling-out-coding-agents-to-a-team Mon, 28 Sep 2026 00:00:00 GMT A coding agent rollout that starts with tooling stalls in a month. Start with one team, one repository, and a backlog you have measured. https://laimonade.com/blog/the-first-two-weeks-of-an-agent-pilot https://laimonade.com/blog/the-first-two-weeks-of-an-agent-pilot Mon, 28 Sep 2026 00:00:00 GMT An agent pilot should count two things: items executed with no human edit, and rework caused by the item rather than the code. https://laimonade.com/blog/data-handling-questions-to-ask-an-ai-vendor https://laimonade.com/blog/data-handling-questions-to-ask-an-ai-vendor Sun, 27 Sep 2026 00:00:00 GMT AI vendor data handling comes down to five questions, and the answers belong in writing rather than in a reassuring call. https://laimonade.com/blog/getting-security-review-approval-for-coding-agents https://laimonade.com/blog/getting-security-review-approval-for-coding-agents Sun, 27 Sep 2026 00:00:00 GMT An AI tool security review stalls on missing information far more often than on real objections. What to bring to the first meeting. https://laimonade.com/blog/audit-trails-when-an-agent-changed-something https://laimonade.com/blog/audit-trails-when-an-agent-changed-something Sat, 26 Sep 2026 00:00:00 GMT An agent audit trail needs three things: attribution, the commit that moved the status, and the criteria the work was accepted against. https://laimonade.com/blog/what-an-agent-should-never-be-able-to-do https://laimonade.com/blog/what-an-agent-should-never-be-able-to-do Sat, 26 Sep 2026 00:00:00 GMT Three agent guardrails are worth enforcing rather than requesting: it cannot close its own work, widen its own access, or erase the record. https://laimonade.com/blog/is-it-safe-to-let-coding-agents-into-your-backlog https://laimonade.com/blog/is-it-safe-to-let-coding-agents-into-your-backlog Fri, 25 Sep 2026 00:00:00 GMT Coding agent security is a question about capability, not intention. Judge a connection by what it cannot reach, not by what a vendor promises. https://laimonade.com/blog/what-permissions-should-a-coding-agent-have https://laimonade.com/blog/what-permissions-should-a-coding-agent-have Fri, 25 Sep 2026 00:00:00 GMT Agent permissions come down to one line: enough to fetch work and hand it back, and no authority to close it. Where that line sits, and how to review it. https://laimonade.com/blog/migrating-a-backlog-without-losing-context https://laimonade.com/blog/migrating-a-backlog-without-losing-context Thu, 24 Sep 2026 00:00:00 GMT A backlog migration that imports everything moves rows and loses reasons. Migrate the next two sprints properly and leave the archive where it is. https://laimonade.com/blog/running-a-product-single-handed-with-coding-agents https://laimonade.com/blog/running-a-product-single-handed-with-coding-agents Thu, 24 Sep 2026 00:00:00 GMT Solo founder coding agents do not remove a job, they collapse two. You are the product owner and the reviewer, and both want the same hour every day. https://laimonade.com/blog/connecting-claude-code-to-your-backlog https://laimonade.com/blog/connecting-claude-code-to-your-backlog Wed, 23 Sep 2026 00:00:00 GMT One command, one authorisation, and you connect Claude Code to backlog items it can fetch itself — with criteria attached. Plus what to do when it drops. https://laimonade.com/blog/what-to-look-for-in-an-agent-native-project-tracker https://laimonade.com/blog/what-to-look-for-in-an-agent-native-project-tracker Wed, 23 Sep 2026 00:00:00 GMT An AI feature bolted onto a tracker is not an agent-native project tracker. Four capabilities that decide it, and the ones that turn out not to. https://laimonade.com/blog/auditing-what-your-agents-actually-shipped https://laimonade.com/blog/auditing-what-your-agents-actually-shipped Tue, 22 Sep 2026 00:00:00 GMT An engineering audit compares three lists: items marked done, commits merged, and behaviour a user can observe. Work in only one of them is the finding. https://laimonade.com/blog/jira-alternatives-for-teams-running-coding-agents https://laimonade.com/blog/jira-alternatives-for-teams-running-coding-agents Tue, 22 Sep 2026 00:00:00 GMT The usual Jira alternatives differ in speed and taste, not in kind. What to test a tracker for when the thing reading your backlog is a coding agent. https://laimonade.com/blog/quality-gates-that-fail-quietly https://laimonade.com/blog/quality-gates-that-fail-quietly Mon, 21 Sep 2026 00:00:00 GMT Quality gates fail quietly when they exit zero for a reason unrelated to what they check. Four patterns, and how to prove yours is not one of them. https://laimonade.com/blog/what-evidence-should-a-coding-agent-return https://laimonade.com/blog/what-evidence-should-a-coding-agent-return Mon, 21 Sep 2026 00:00:00 GMT Agent evidence is what a reviewer can re-run: the commands and their exit codes, the files changed, and the criterion each change satisfies. https://laimonade.com/blog/sprint-length-for-teams-running-agents https://laimonade.com/blog/sprint-length-for-teams-running-agents Sun, 20 Sep 2026 00:00:00 GMT Sprint length should be the larger of two numbers, and coding agents collapse one of them while leaving the other exactly where it was. https://laimonade.com/blog/why-done-and-the-repository-disagree https://laimonade.com/blog/why-done-and-the-repository-disagree Sun, 20 Sep 2026 00:00:00 GMT Status accuracy is a data-integrity problem wearing a discipline costume. A tracker holds no reference into the repository, and the error points one way. https://laimonade.com/blog/what-to-do-when-an-agent-returns-work-that-does-not-meet-criteria https://laimonade.com/blog/what-to-do-when-an-agent-returns-work-that-does-not-meet-criteria Sat, 19 Sep 2026 00:00:00 GMT A large share of agent work rejected at review is a specification defect, not a coding one. How to tell, and how to send it back so it fails differently. https://laimonade.com/blog/do-story-points-still-work-with-coding-agents https://laimonade.com/blog/do-story-points-still-work-with-coding-agents Fri, 18 Sep 2026 00:00:00 GMT Story points still work, but they stop measuring effort and start measuring specification risk. Why the cheap-looking items are now the expensive ones. https://laimonade.com/blog/standups-and-status-when-agents-report-their-own-work https://laimonade.com/blog/standups-and-status-when-agents-report-their-own-work Fri, 18 Sep 2026 00:00:00 GMT The reporting half of an agile standup is redundant once the repository says what happened. What breaks is the coordination that was riding along with it. https://laimonade.com/blog/how-many-coding-agents-can-one-team-run https://laimonade.com/blog/how-many-coding-agents-can-one-team-run Thu, 17 Sep 2026 00:00:00 GMT How many coding agents one team can run is set by two human queues rather than by compute, and most teams only ever count one of them. https://laimonade.com/blog/how-sprint-planning-changes-when-agents-do-the-building https://laimonade.com/blog/how-sprint-planning-changes-when-agents-do-the-building Thu, 17 Sep 2026 00:00:00 GMT Sprint planning with AI agents stops asking how much a team can take and starts asking how much of the work is genuinely ready to be executed. https://laimonade.com/blog/code-review-when-an-agent-wrote-the-code https://laimonade.com/blog/code-review-when-an-agent-wrote-the-code Wed, 16 Sep 2026 00:00:00 GMT In AI code review the summary is a claim and the diff is the evidence. Five things agents get wrong that people rarely do, and one no reviewer automates. https://laimonade.com/blog/how-to-give-a-coding-agent-a-work-order https://laimonade.com/blog/how-to-give-a-coding-agent-a-work-order Wed, 16 Sep 2026 00:00:00 GMT A coding agent work order has six parts, and a prompt has none of them. What to include, what to leave out, and what should come back when it is finished. https://laimonade.com/blog/dependencies-and-sequencing-in-an-agent-backlog https://laimonade.com/blog/dependencies-and-sequencing-in-an-agent-backlog Tue, 15 Sep 2026 00:00:00 GMT Two sensible items that each need the other output cannot both be satisfied. Backlog dependencies fail quietly, as an unfinished checkbox. https://laimonade.com/blog/running-coding-agents-in-a-team-workflow https://laimonade.com/blog/running-coding-agents-in-a-team-workflow Tue, 15 Sep 2026 00:00:00 GMT A coding agent workflow is four stages: specify, dispatch, execute, verify. The stage teams skip is the first one, and the bill arrives at the fourth. https://laimonade.com/blog/how-to-audit-a-backlog-before-pointing-agents-at-it https://laimonade.com/blog/how-to-audit-a-backlog-before-pointing-agents-at-it Mon, 14 Sep 2026 00:00:00 GMT A backlog audit is twenty items and four questions. Whatever you cannot answer from the item own text is the defect, and it is cheaper to find now. https://laimonade.com/blog/splitting-stories-for-coding-agents https://laimonade.com/blog/splitting-stories-for-coding-agents Mon, 14 Sep 2026 00:00:00 GMT One observable outcome per item, not one component per item. Why splitting user stories properly makes the estimate go up rather than down. https://laimonade.com/blog/acceptance-criteria-that-an-agent-can-verify https://laimonade.com/blog/acceptance-criteria-that-an-agent-can-verify Sat, 12 Sep 2026 00:00:00 GMT Acceptance criteria are verifiable when they name something you can run and an outcome you can observe. Five that fail that test, and their fixes. https://laimonade.com/blog/what-product-ownership-costs-a-small-engineering-team https://laimonade.com/blog/what-product-ownership-costs-a-small-engineering-team Sat, 12 Sep 2026 00:00:00 GMT The product owner cost on a team of three to eight engineers is four to six hours a week, paid by whoever has least protection over their calendar. https://laimonade.com/blog/ai-product-owner-vs-human-product-owner https://laimonade.com/blog/ai-product-owner-vs-human-product-owner Fri, 11 Sep 2026 00:00:00 GMT A human product owner decides what the product should become. An AI product owner turns those decisions into work an agent can execute, then checks it. https://laimonade.com/blog/signs-your-team-has-a-product-ownership-gap https://laimonade.com/blog/signs-your-team-has-a-product-ownership-gap Fri, 11 Sep 2026 00:00:00 GMT Six symptoms of a product-ownership gap that read as ordinary team friction, and are one thing underneath: work is built faster than it is specified. https://laimonade.com/blog/why-agent-adoption-stalls-after-the-first-month https://laimonade.com/blog/why-agent-adoption-stalls-after-the-first-month Wed, 09 Sep 2026 00:00:00 GMT Month one of coding-agent adoption spends a backlog written over the previous year. The stall that follows is a supply problem, not a model problem. https://laimonade.com/blog/what-is-mcp-and-why-it-matters-for-backlogs https://laimonade.com/blog/what-is-mcp-and-why-it-matters-for-backlogs Fri, 04 Sep 2026 00:00:00 GMT The Model Context Protocol turns a backlog from a page someone reads into a service an agent can call. What that changes, and what to withhold. https://laimonade.com/blog/definition-of-done-when-an-agent-wrote-the-code https://laimonade.com/blog/definition-of-done-when-an-agent-wrote-the-code Tue, 01 Sep 2026 00:00:00 GMT A commit message is a claim and the diff is the evidence. What a definition of done must say once the author is an agent, and why the last step is human. https://laimonade.com/blog/who-owns-the-backlog-when-agents-write-the-code https://laimonade.com/blog/who-owns-the-backlog-when-agents-write-the-code Wed, 26 Aug 2026 00:00:00 GMT Backlog ownership splits into deciding what should exist and keeping each item executable. Only one of those two halves is still a meeting. https://laimonade.com/blog/how-to-write-a-backlog-item-an-agent-can-execute https://laimonade.com/blog/how-to-write-a-backlog-item-an-agent-can-execute Wed, 19 Aug 2026 00:00:00 GMT A backlog item for AI agents needs five fields a human reader never asks for. Here they are, with one real item rewritten from unusable to executable. https://laimonade.com/blog/what-an-ai-product-owner-actually-does https://laimonade.com/blog/what-an-ai-product-owner-actually-does Wed, 12 Aug 2026 00:00:00 GMT An AI product owner turns decisions into work an agent can execute, sequences it, hands it over, and checks what comes back. Deciding stays human. https://laimonade.com/blog/a-backlog-an-agent-can-read https://laimonade.com/blog/a-backlog-an-agent-can-read Wed, 05 Aug 2026 00:00:00 GMT Four backlog failure modes that stand between a normal backlog and an agent-ready backlog. Each reads as an ordinary unfinished ticket.