AI-augmented software and product engineer · agent pipelines run under a written harness · the Next.js and TypeScript products they ship
I lead the frontend of a multi-tenant real-estate platform: Next.js and TypeScript sites on one shared component package, each with its own headless CMS. Most of that work is delivered through AI agents running under a written harness. The agents write code; the deciding happens first, and it happens on paper.
Five stages. The first four are specification; only the fifth is code.
- Grill the problem. What is actually being asked for, and what happens if we do nothing. Most of what I cut gets cut here.
- Write the master priority document. One file a cold reader can run the next wave from.
- Cut tickets. The finding in the title, the receipts in the body.
- Set the fence. Which files each agent lane owns, and which shared files are serialised.
- Dispatch waves, then review every one. Nothing lands without a review pass.
Four invariants keep it safe: two-axis review before acceptance (standards and spec) · a capability guard decides whether a lane runs attended or autonomously · evidence labels on every claim (measured, source, inferred) · agents never merge.
The full write-up, including an audit of how I direct agents by one of my own agents, is at brandonperfetti.com/how-i-work. The agreements themselves are public: agent-working-agreements.
One program is the clearest illustration: a fleet-hardening initiative around incremental static regeneration. Roughly 150 tickets and 426 files across a shared component package, the CMS, the portal and the tenant sites, first commit to fleet promotion in four weeks, in ten reviewed waves, with the fleet end-to-end suite green at every wave.
The number I care about there is not 426. It is ten: the number of times the work stopped and got looked at before any of it reached a release branch.
Before that, a decade of data integrations built the depth: helping build a React and GraphQL ingestion platform for 250+ MLS feeds and 10M+ records, SAML/JWT SSO across 100+ platforms, and a re-architecture that cut data-source integration time by 80%.
|
The harness itself: evidence labels, the two modes and the guard that picks between them, the git ritual, the stop-list. The changelog shows which rules changed after something bit.
|
14 charts on PR throughput, cycle time, flow health and release cadence, read server-side from the GitHub API. Nobody asked me to build it. Source.
|
|
The site and its content platform: a block page-builder on Payload CMS, Storybook for the components, Vitest and Playwright in CI.
|
An interactive macOS-inspired portfolio: windows, a dock, a menu bar. What frontend craft looks like when the constraints are removed.
|
|
Event timelines made simple for teams and organizations. A SaaS product built around clear information architecture and fast, intuitive UX.
|
A modernized web presence for a forensic engineering firm. Accessibility-first, CMS-driven, and built for long-term maintainability.
|
- Four-Week Context Reset: Recover Focus After an Interrupt-Heavy Sprint · Sep 2026
- Your Runbook Is Rotting. Teach It to an Agent Instead. · Aug 2026
- The Cheapest Database Migration Is the One You Do Before Production Exists · Aug 2026
More at brandonperfetti.com/articles.
Core
AI-augmented engineering
UI
Backend and data
Testing
Delivery
22 certificates in three groups (click to expand)
| Certificate | Issuer | Issued |
|---|---|---|
| AI Coding Crash Course | AIHero.dev | Sep 2026 |
| AI Coding for Real Engineers | AIHero.dev | Jun 2026 |
| Claude Code for Real Engineers | AIHero.dev | Apr 2026 |
| AI SDK v6 Crash Course | AIHero.dev | Mar 2026 |
| Build Your Own AI Personal Assistant in TypeScript | AIHero.dev | Dec 2025 |
| Master the Model Context Protocol (MCP) | EpicAI.pro | Nov 2025 |
| Certificate | Issuer | Issued |
|---|---|---|
| Certificate of Interface Design | Shift Nudge | Apr 2026 |
| The Complete Next.js Testing Course | JS Mastery | Feb 2026 |
| Database Mastery: SQL to Prisma | JS Mastery | Jan 2026 |
| Full Stack Foundations | Epic Web | Jul 2024 |
| Professional Web Forms | Epic Web | Jul 2024 |
| Data Modeling Deep Dive | Epic Web | Jul 2024 |
| Authentication Strategies & Implementation | Epic Web | Jul 2024 |
| Pixel Perfect Figma to Tailwind | Epic Web | Jul 2024 |
| Certificate | Issuer | Issued |
|---|---|---|
| Web Application Testing | Epic Web | Jul 2024 |
| Testing Fundamentals | Epic Web | Jul 2024 |
| Automated Accessibility Testing | testingaccessibility.com | Jul 2024 |
| Coding Accessible Interactions and Mechanics | testingaccessibility.com | Jul 2024 |
| Semantic Markup with HTML and ARIA | testingaccessibility.com | Jul 2024 |
| Manual Accessibility Testing | testingaccessibility.com | Jul 2024 |
| Design Thinking & People Skills for Accessibility | testingaccessibility.com | Jul 2024 |
| Foundations of Accessibility | testingaccessibility.com | Jul 2024 |
Each link opens the certificate itself. The full set is on LinkedIn.
Courses completed, by platform (click to expand)
AI Hero
- AI Coding for Real Engineers (cohort)
- Claude Code for Real Engineers
- AI Coding Crash Course
- AI SDK v6 Crash Course
- Build Your Own AI Personal Assistant in TypeScript
Epic Web and Epic AI
- Epic Web: full-stack foundations, forms, data modeling, authentication, testing, Figma to Tailwind
- Master the Model Context Protocol
- Epic React
- Testing JavaScript
Testing Accessibility
- Testing Accessibility: the six workshops, foundations through automated testing
JS Mastery and Shift Nudge
Recently merged
- agent-working-agreements: Release: wave 4 — the OpenClaw appendix's added-file readback (#103) and the Claude Code appendix's EnterWorktree line (#101) · Oct 2026
- agent-working-agreements: OpenClaw: how row 3's added-file readback passes the path (#104) · Oct 2026
- agent-working-agreements: Claude Code: EnterWorktree is not the lane-isolation command (#101) · Oct 2026
View stats
The card is regenerated daily by index.js in this repository, from the GitHub API, and committed as an SVG. No third-party stats service is involved.
Time spent this year
If you are solving the same problem, I would like to compare notes.
