Field notesOn engineering with AI agentsLast entry 27 Oct 2026
Agents do the work.
People make the calls.
I'm Daniel Broadhurst, a full-stack engineer. These are my notes on building real software with AI agents in the loop: how to structure the work, where a person should decide, and what to leave out.
Fig. 1 · How a feature moves
agentyou
- 01ResearchAn agent writes the brief and picks the lane: feature or fix.
- 02PrototypesThree genuinely different takes. You pick one.
- 03PlanA spec, or questions. You only answer if it asks.
- 04BuildAgents implement, run the checks and open a draft PR.
- 05IterateYou try it running. Five minutes beats a green test run.
- 06ReviewTwo reviewers on different models argue. You tick what’s worth fixing.
- 07LaunchDocs and the demo ship alongside it.
Contents
Every post →- 01Lambda Memory Size for Node.js: What 160 Benchmarks SayDoes more Lambda memory cut cold starts or just raise the bill? My 160 benchmark runs, split into cold start, warm duration and cost, and why 512 MB won.27 Oct 2026 · 6 min read
- 02Retiring WordPress with a Cloudflare Worker: 301s and 410sOne Cloudflare Worker in front of a static site: 301s for old WordPress permalinks, 410 Gone for wp-admin and wp-json, and tests before DNS cutover.27 Oct 2026 · 8 min read
- 03Migrate WordPress to Astro with the REST API, Not XML ExportAn 84-line Node.js script that moves WordPress posts into Astro content collections via the REST API, images downloaded and captions kept.20 Oct 2026 · 7 min read
- 04Working with AI agents · Part 2From Idea to Pull Request, With Humans at the GatesA staged workflow for building features with AI agents: research, prototypes, plan, build, iterate and review, with people deciding only at the gates.13 Oct 2026 · 6 min read
- 05Working with AI agents · Part 1One Inbox for Everything Your Agents Need From YouA method for working with AI agents: audit what they ship, have each keep a diary, and derive one inbox of everything that needs a human decision.6 Oct 2026 · 5 min read
- 06The Guardrails That Stop My Publishing Agent Before It ShipsAI agent guardrails in real code: a leak scan before every push, a build check that fails the run, two human gates in GitHub and the cost printed in each PR.29 Sept 2026 · 6 min read
- 07Direct, SQS, EventBridge or Step Functions? What 160 Lambda Benchmarks SayI benchmarked four ways to chain AWS Lambda functions for my MSc. Managed orchestration adds a ~1.5 s routing tax, 512 MB is the memory sweet spot, and SQS catches up.27 Sept 2026 · 6 min read
- 08How AI Agents Scale Software Engineering: The Agentic WorkflowThe agentic workflow: how a pipeline of AI agents acting as product manager, architect, engineer and QA can scale software engineering without cutting quality.12 Jan 2026 · 9 min read
Liked this?
Get the next field note
New posts and tools on engineering with AI agents and serverless, when they're ready. No schedule, no spam.
About
I build robust back ends and interfaces people enjoy using, mostly in TypeScript on AWS.
Lately I spend my time on the question every team is asking: how do you get real, shippable work out of AI agents without becoming the bottleneck? This is where I write up what's worked. New here? Start with the series.
Say hello —
hello@danielbroadhurst.co.ukSet in Newsreader, Inter and Caveat.
Built with Astro, served from Cloudflare.