https://laimonade.com/blog Notes on product ownership for teams running coding agents. en 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.