Pulse
Tool UpdateApr 2026

Anyone can build agents now

The shift

Building agents means starting from scratchFork a production agent and adapt it

How it happened

Apr 1
Claude Code source code published

Full production codebase on GitHub — not a demo, the actual shipping product

Apr 2
Community dissects the architecture

Developers find clean patterns: tool loops, context management, error recovery

Apr 4
First forks and custom agents appear

Teams adapt Claude Code patterns for customer support, data analysis, content pipelines

The players

Anthropic

Published the full production agent — not a tutorial, the real thing

Open source

Immediately forked and adapted for dozens of use cases

What you used to think

  • "Agent architecture is a secret sauce only big labs have"
  • "I'd need a team of engineers to build an agent"
  • "I'll just wait until agents come built into the tools I use"

What's actually true now

  • The full source code of a production agent is free to study and fork
  • The patterns are surprisingly simple — plan, act, observe, retry
  • Solo developers are shipping custom agents within days of reading the code

What to do about it

Read the Claude Code source to understand how agents actually work. Even if you don't build one, knowing the loop (plan → tool call → observe → decide) makes you a better user of every AI agent.

AgentsHarness Engineering