My AI Did 16 Things
Simon Willison amplified David Wilson's May 31 essay calling AI a "thermonuclear ADHD amplifier." Wilson lists 16+ projects spawned by Claude sessions that started "write a quick script for X" and ended in unfinished sprawl. The Hacker News thread split sharply: ADHD users said agents finally let them ship.
Yesterday's operator essay isn't about model quality. It's about what unlimited cheap output does to attention.
Wilson's piece ("The solution might be cancelling my AI subscription") describes the new failure mode of 2026: you ask Claude for a quick script, one hour later you have a tested, documented project you didn't need, and your original task is still untouched. Willison: "I'm finding that coding agents can take me from a vague idea to a working solution... in less than an hour. Even if the code is rock solid, there's a limit to how many projects like that I can sensibly care for." The Hacker News thread is the interesting part: ADHD readers said the opposite - agents finally let them ship.
For PMs: the new product surface is discipline tooling, not capability. Limit-setters, focus modes, abandon-tracking. For ops: measure project completion rate, not project velocity. For builders: a coding agent that says "are you sure this is what you needed?" is now a feature, not a snub.
⚡ Why this matters
- First widely-discussed operator essay in 2026 that frames AI as a productivity tax, not a productivity boost.
- Names the second-order effect we've all been feeling - cheap output multiplies abandoned work.
- The split reaction (Wilson's frustration vs ADHD readers' relief) maps to a real product opportunity: AI for hyperfocus vs AI for completion.
🔍 What happened
- May 31, 2026. David Wilson publishes "The solution might be cancelling my AI subscription" on thoughts.hmmz.org.
- Wilson's lede: he lists 16+ projects he spun up with AI tooling and concludes "I didn't mean to build most of these things."
- Key quote: AI as "a thermonuclear ADHD amplifier. I have seen the same effect in every single one of my adult friends."
- Hacker News thread #48345896: hundreds of comments, prominent front-page placement same day.
- Simon Willison reposts (May 31, 4:31pm) and adds: "there's a limit to how many projects like that I can sensibly care for."
- Counter-voices in the HN thread: multiple ADHD users say agents let them finish side projects for the first time.
💬 Smart takes
- David Wilson: "This technology is horrific for attention. It's a thermonuclear ADHD amplifier. Folk running 3 screens simultaneously working on totally unrelated projects they have little hope of maintaining."
- Simon Willison: "I'm hopeful that the critical skill to develop here is discipline. That's not great news for me: I've been trying to figure that one out for decades!"
- HN counter (ADHD reader): "For those of us prone to hyperfocus, working with AI can provide the kinds of stimulation we crave. I can hardly remember a time when I've felt more engaged with my work."
- HN counter (ADHD reader): "I'm finishing side projects for the first time ever because I can actually get them working before I get bored of them."
🧭 Where this goes
- A wave of "AI hygiene" essays from operator voices (Mollick, Shipper, Karpathy, Krieger) by end of Q3.
- Product surfaces emerge for tracking abandoned projects, capping concurrent sessions, requiring a written goal before code generation.
- Hyperfocus-as-feature gets explicit positioning by Cursor, Claude Code, or a new entrant. One of them ships an "intent lock" by Q4.
- Education curricula start adding "agent discipline" as a required skill alongside prompt engineering.
- Cancel-rate becomes a tracked metric for AI subscription products. Net retention starts to depend on how often users finish what they start.
🎯 Implication
- For PMs: add session goals + abandonment metrics to your product analytics. Discover where the meaningless work happens.
- For builders: the next valuable feature isn't more capability. It's a confirmation step before the agent runs another hour of work.
- For individual operators: if you have 5+ unfinished AI-spawned projects, the product isn't broken. Your filter is.