Very important agents (Friends)

What It Covers

Nick Nisi discusses his workflow using Claude Code for daily development at WorkOS, the Anthropic acquisition of Bun, emerging browser wars around AI agents, and predictions that developers will write significantly less code in 2025 while producing more output.

Key Questions Answered

• •Anthropic Acquires Bun: Anthropic purchased the Bun JavaScript runtime to accelerate AI coding tools like Claude Code and Claude Agent SDK. Bun remains open source under MIT license with the same team, but will now optimize specifically for making AI coding tools faster and smaller while maintaining node compatibility focus.
• •Context Window Management: Claude Code Opus 4.5 reduced API costs from $25 to $15 per million input tokens and $75 to $25 per million output tokens. Users can run slash context command to monitor token usage, with MCP tools consuming significant baseline tokens (112,000 before any work). Amp's control-O handoff feature creates fresh context windows while referencing old threads.
• •Custom Output Styles: Claude Code's output styles feature lets developers customize system prompts to control code generation behavior. Setting styles to pause at business logic while auto-generating boilerplate keeps developers engaged with core functionality. This prevents the problem of debugging code written weeks ago that scrolled past without review or understanding.
• •Skills Over MCP: Developer-created skill marketplaces using marketplace.json files in git repos enable sharing custom agents and commands. Nick's code simplifier agent automatically removes AI-generated comments and cleans code without breaking functionality. Skills provide better tool discovery than Model Context Protocol servers, which consume excessive context tokens and lack discoverability.
• •Migration from GitHub: Zig programming language moved from GitHub to Codeberg, a nonprofit EU-based GitHub alternative, citing platform bloat and Microsoft neglect of core features. GitHub Sponsors remains the primary retention mechanism. Multiple projects follow this trend as developers seek alternatives to increasingly AI-focused, action-unstable GitHub platform with excessive JavaScript bloat.