How I use AI (Oct 2026)

Salesforce AI Research engineer Simon Willison, formerly of Microsoft’s Windows AI team, disclosed he now spends roughly $3,000 per month on AI tokens and manages five to six agentic coding agents—Claude Code, OpenCode, Codex and Pi—daily, having largely stopped writing code by hand as agents now implement features, reproduce bugs, run tests and draft and review pull requests, prototyping new features in about 20 minutes. He said coding is “solved” and that GitHub Copilot autocomplete, which he praised a year ago, has been made obsolete by agentic workflows. Willison also drafted a personal AI policy to set norms for day-to-day use, arguing that humans lack conventions for the tools and are “messing up,” citing BrainPOP’s apparent use of AI art and researchers delegating thinking to machines. The account matters as a signal of how fast agentic coding is displacing autocomplete in professional workflows and of emerging demand for explicit usage policies at both individual and organizational levels. Watch for whether token spending of this scale becomes typical for engineers, how agent-orchestration tooling consolidates, and whether formal AI policies spread from individuals to employers and publishers.