Dev runs Opus-level coding agent locally at 2x speed for free, sees Substack arbitrage

julianharris · x · 2026-09-19

Julian Harris shares his Substack long-form piece: he claims to run an "Opus-level" coding agent locally at nearly 2x Claude Opus speed, for free — arguing Aug 2026 is for local AI what Nov 2025 (Opus 4.5) was for cloud AI coding: a tipping point.

He also spots an arbitrage: frontier AI knowledge on X takes months to reach Substack audiences, so brokering and repackaging it is a money-making opportunity — even his now-outdated local AI write-up earned respectable engagement there. He also vents about Anthropic abruptly changing quotas, rate limits and censorship, making it an unreliable business dependency.

Original post →

More from coding & agent

coding & agent channel →