How Wu Uses the Leash Scott Wu draws the image from child psychology: an infant can't be left alone, a toddler for a few minutes, a teenager for a night, a college student fully on their own. AI agents have followed the same trajectory — from GPT-3's tab completion to Devon or Claude Code running for 10–20 minutes at a stretch. Each step is a smooth lengthening of the leash [^e8655330-4d35-560a-9033-baf07ebccdb0@0]. Why the Analogy Matters The leash reframes the AGI question: instead of a hard cutoff, autonomy grows continuously, and the moment worth watching is when it becomes economically profitable to never turn your AI off. Wu still pushes back on purely economic definitions — value depends on how substitutable your output is, and if everyone has an AGI, the AGIs compete with one another. The leash captures the autonomy dimension that percentage-of-labor benchmarks keep moving away from. - The leash = how long an AI agent can work unsupervised before a human must step back in.- Child development supplies the analogy: infant → toddler → teenager → college student, matched by tab completion → agents running 10–20 minutes.- The leash is lengthening over time, which is why Wu resists a single hard AGI cutoff.