Skip to content
Blog
- The phone keyboard covers your form field: test every input Chrome on Android no longer shrinks the page when the keyboard opens, so one Playwright test should check every input stays visible and on top.
- Keep up with web platform changes without a reading list Have an agent turn current web guidance into checks on your code, review them in a PR, and let a nightly job file a few tickets.
- Turn your agent's repeated mistake into a lint error A rule your agent keeps breaking is in the wrong place. Group mistakes that happened twice and enforce each in code, lint, a test or a doc.
- Let Postgres refuse the double booking An app-side availability check races with itself; a Postgres exclusion constraint makes the database reject the second overlapping booking.
- Give a UI-reviewing agent screens to judge against Without references, an AI UI reviewer passed a broken screen. Given two screens the app already ships, it blocked all six problems.
- Turn on CSP reporting without flooding your own API Send CSP violation reports to a small collector off your API, start in Report-Only, and count reports instead of storing each one.
- Upserts skip your optimistic lock unless they bump it An upsert that leaves the version column alone lets a stale copy overwrite it with no error. One clause fixes it, and one test catches it.
- Swap a heavy MCP server for a script your agent calls For a few fixed tracker calls, a short script beats an MCP server: smaller responses, only the writes you wrote, and runs you can replay.
- Review your coding agents' spend like a slow query Read yesterday's agent transcripts for the patterns that burn money and the outlier runs, then turn each one into a rule or a ticket.
- Ship a security fix without locking out real users Count which real callers a security fix would reject before it ships, then compare each user's 4xx after deploy and roll back new refusals.
- Postgres migrations that don't stall the table Start with lock and statement timeouts, add a NOT NULL column with a constant default in one statement, and know when CONCURRENTLY cannot run.
- Evidence or nothing: an AI bug sweep your team won't mute Every finding needs a screenshot, log line or database row, a nightly cap, prod writes only on a demo account, and unproven ones reported apart.
- Build the bug repro from the real failing case Have the agent pull the exact failing case before writing a repro test, and call a bug confirmed only when the test fails with the reported value.
- Turning a job done twice into a nightly ticket How to turn a job your agent did well twice into a scheduled command that files tickets into the normal queue, with the rule checked at deploy.
- Spec first or build first with a coding agent Spec-first versus build-first compared on cost, feedback and risk, and why build-first fits anything you can hide behind a flag.
- The setup that stops AI slop before code review Five layers a tech lead owns so AI-written code arrives clean: a rules file, lint rules, saved commands, a check on them and an AI reviewer.