Developer blog
Practical notes on building reliable software with AI coding agents: context, tools, verification, and the systems around the model.
These notes come from shipping a portfolio of small iOS apps. Browse the app catalogue. Subscribe via the RSS feed.
Selected engineering work
Explore a working browser demo and small, reproducible examples. Each guide separates what was observed from what the example is intended to explain.
Jev, Open Source Experiments, and a Different Kind of AI Decision →
Try a local model in your browser and inspect its typed decisions.
How to Verify the Exact Code Your Agent Will Ship →
Follow a worked verification receipt from source revision to artifact.
Write Tool Contracts That Prevent Avoidable Retries →
Run a small contract example and compare rejected and accepted requests.
Developer essays
- 24 September 2026 · 7 minWhat to Check After You Stop a Coding AgentRecover after stopping a coding agent: inspect completed writes, handle unknown outcomes, check undo coverage, and decide when a retry is safe.
- 24 September 2026 · 9 minJev, Open Source Experiments, and a Different Kind of AI DecisionA look at Jev's hosted decision API, open-source examples, and a hands-on local model demo in your browser.
- 24 September 2026 · 9 minWrite Tool Contracts That Prevent Avoidable RetriesDesign AI agent tool schemas with visible limits, useful errors, and server validation. A practical MCP and function-calling contract checklist.
- 24 September 2026 · 8 minHow to Verify the Exact Code Your Agent Will ShipConnect your coding agent's checks to the exact source, build artifact, and deployment target with a practical verification receipt.
- 1 September 2026 · 5 minHow to Build Huge Agent Knowledge Bases Without Huge Context WindowsUse progressive disclosure, scoped Skills, and context measurements to give coding agents depth without loading every rule up front.
- 1 September 2026 · 4 minStop Putting Everything in CLAUDE.mdA practical guide to keeping CLAUDE.md small: put invariants in the root, move procedures into Skills, and keep context useful.
- 1 September 2026 · 5 minYour CLAUDE.md Needs Tests TooTreat CLAUDE.md, Skills, and agent policies as software: version them, test real tasks, and check robustness before rollout.
Apps from the studio
All appsThese notes come from the same shipping system behind Pocket Grove's active apps. If you came here looking for something to install, start with a few of the studio's current releases.
