Agent workflows
A coding-agent handoff should survive a fresh session
Keep coding-agent work resumable with a compact handoff: current files, verification results, open decisions, permissions and the next bounded action.
6 min readNO FEED / Articles
Articles
Practical guides and considered arguments for building with AI.
32 articles
Code quality
Before an AI refactor, record what the old code actually does
Capture legacy behavior before an AI refactor. Choose fixtures, control snapshot noise, review changed outputs, and separate preservation from bug fixes.
6 min readEnvironments & integrations
Build a reproducible development environment for your coding agent
Build a coding-agent environment with a declared toolchain, locked dependencies, repeatable bootstrap and a real readiness check. Start from a clean checkout.
5 min readTools & models
Claude Code vs Codex vs Cursor: choose the workflow you can finish
Compare Claude Code, Codex and Cursor by execution environment, permissions and review handoff. Use a small task-based trial instead of a universal ranking.
6 min readCode quality
Diagnose flaky tests before a coding agent “fixes” CI
Investigate flaky Playwright tests with first-failure evidence, controlled reproduction and explicit retry policies before an AI agent changes your CI.
5 min readTools & models
Evaluate coding agents on your repository with a reusable task card
Build a fair coding-agent comparison with frozen tasks, trusted graders, repeated trials and complete cost records. Includes a reusable task-card checklist.
6 min readAgent workflows
Give a coding agent a monorepo task without loading the whole monorepo
Give a coding agent focused monorepo context without missing downstream checks. Compare scoped instructions, dependency direction and affected-test selection.
6 min readSecurity & privacy
Give a coding agent credentials without handing over production
Scope coding-agent access by task, resource and lifetime. Check inherited credentials, GitHub App tokens, OIDC trust and the actual sandbox boundary.
5 min readCode quality
How to test AI-written code without copying its assumptions
Test AI-written code against independent expectations. Compare worked examples, properties, metamorphic checks and mutation testing with one interval example.
6 min readCode quality
Review an AI-generated database migration before it touches real data
Review AI-generated PostgreSQL migrations with compatibility checks, lock analysis, resumable backfills and recovery checkpoints before touching real data.
6 min readSecurity & privacy
Review an MCP server before connecting your coding assistant
Review an MCP server’s publisher, tools, permissions, data flow and update behavior. Use a disposable trial and verify how to revoke the connection.
6 min readSecurity & privacy
Run coding agents in GitHub Actions without trusting every issue
Review coding-agent workflows in GitHub Actions: untrusted issues and PRs, token permissions, runners, artifacts and the boundary before a privileged job.
6 min readAgent workflows
Skills, MCP or repository instructions? Start with the missing piece
Choose repository instructions, an agent skill or MCP by the task’s missing piece. Compare context, procedures, access and enforcement with one worked example.
5 min readCode quality
Test accessibility in an AI-built interface beyond a green axe scan
Test AI-generated interfaces with Playwright and axe, keyboard journeys, focus checks and screen-reader evaluation. Know what a clean scan cannot prove.
6 min readCode quality
Verify AI-generated frontend changes with Playwright visual checks
Verify AI-generated frontend changes with stable Playwright screenshots, reviewed baselines and meaningful diffs, while keeping behavior checks separate.
5 min readSecurity & privacy
Verify AI-suggested packages before installing them
Check AI-suggested dependencies before installation. Separate package identity, provenance, hashes and script controls with practical npm and pip guidance.
6 min readSecurity & privacy
What leaves your repository when you use an AI coding assistant?
Compare specific Claude Code, Cursor and Copilot data paths: code transmission, indexing, training, retention, cloud sessions and feedback exceptions.
6 min readTools & models
When a local coding model makes sense
Decide whether a local coding model fits your work. Check real context, CPU/GPU placement, tool access and privacy before buying hardware or expanding scope.
6 min readAgent workflows
Write a coding-agent task brief that can be checked
Turn a vague coding request into a checkable task brief. Use concrete examples, acceptance criteria, boundaries and evidence, with a worked CSV-export case.
5 min readAgent workflows
Write an AGENTS.md that earns its place in the context
Write useful AGENTS.md and CLAUDE.md instructions: verified commands, clear scope, maintained references and a practical check of what your agent actually loads.
5 min readCode quality
Your agent finished. Can you explain the change?
A music app its author stopped understanding. A decoder checked against a reference. Two accounts make a useful case for keeping comprehension inside the work.
4 min readCost & efficiency
Cheap subagents need a result and a stop.
A developer chose cheaper models for delegated tasks and still saw most usage attributed to subagents. Measure the accepted result, including handoffs, retries and review.
5 min readEnvironments & integrations
Customize the harness. Retest the boundary.
A developer built a mod to gate deployments. Another tested whether a different mod could approve a blocked command. Their accounts show why the installed hook chain matters.
5 min readEnvironments & integrations
Give each agent its own application state.
One developer’s worktrees stopped file conflicts but left browser tests reaching shared services. Separate the database and runtime each task actually uses, then test cleanup.
5 min readEnvironments & integrations
The setting you wrote is no longer read
A clean startup can hide a missing instruction file, an ignored exporter setting or a shorter context limit. Follow one hypothetical invoice workflow through the September changes.
5 min readCost & efficiency
The harness is a pricing decision
A developer choosing an agent also chooses how that agent spends. A September comparison kept the model fixed and found a large cost gap. Its evidence is useful; its limits matter before a team switches tools.
5 min readEnvironments & integrations
The hop nobody tests
A coding session can fail before the model sees a useful prompt. September’s Claude Code regressions exposed two different contracts between the client and an intermediary: how it signs in, and which tool schemas the endpoint accepts.
5 min readCost & efficiency
The discount is a cache hit
A cheaper cache read helps only when the client actually reuses the context. Claude Code’s September rollout of Fable 5.1 shows how a working agent can finish its task while missing the economics used to justify it.
5 min readEnvironments & integrations
Stable is not safer
A team can choose a cautious update channel and still miss a fix for its own production path. The August release sequence shows why the installed version matters more than the reassuring name.
5 min readSecurity & privacy
The session left the laptop
Closing a laptop can stop access to its folders while a cloud task keeps running. The August releases moved work across that boundary; the permissions need to make the same journey.
6 min readSecurity & privacy
The click was never the control
August’s switch to automatic command screening addressed approval fatigue. A tax-calculation fix shows why the place where code reaches production still needs its own decision.
5 min readTools & models
Open weights inside a closed harness
A lower token rate is useful only if the agent can still reach its tools. An August Codex/DeepSeek integration shows where that comparison can go wrong.
4 min read