Claude as My External Brain: Autistic, ADHD, and Supported (zackproser.com)

A software engineer with autism and severe ADHD describes using Claude as an "external brain" to turn voice‑first thinking and biometrics into a safe, high‑velocity development loop. By orchestrating background agents (isolated VMs/branches) rather than micromanaging them, the author dictates briefs in the morning, designs and reviews on mobile while walking, and completes PRs back at the desk—while CI/CD gates (pre‑commit/secret scanning, dependency checks, lint/test/build) enforce a defense‑in‑depth safety lane. Claude reconciles task state from sources like Linear and biometrics (MCP/Oura), offers compassionate pushback to counter perfectionism, and holds unlimited patience for repeated queries—reducing shame, burnout, and cognitive overload. Technically, the post lays out what makes tasks agent‑friendly (narrow scope, deterministic validation, minimal coupling) and concrete workflows: WhisperFlow streams speech to Claude to generate tickets/PRs; agents run with strict acceptance criteria and return diffs/previews for human review. Reusable acceptance checks include build success on the target Node version, key unit tests green, zero findings from secret scanners, no high/critical dependency vulnerabilities, previews showing intended before→after behavior, and a PR with changed files, tests, and risk notes. The writeup is a practical blueprint for untethered, auditable development with ML agents—highlighting gains in productivity and mental health while flagging the need for secure pipelines and careful handling of biometric/privacy tradeoffs.