Claude Clue #4: Your Data Manager Might Be the Weak Link

Claude Code isn't just one model—it's multiple models working together.

Claude Clue #4 - Tips for working with AI
Christopher's Claude Clues #4

Here's something suprising: Claude Code isn't just one model—it's multiple models working together. You pick your main model for the heavy lifting, but behind the scenes, Haiku handles context management by default. It summarizes conversations, decides what information to keep as your session grows, and manages data when you exceed what the main model can hold. Haiku is fast and cheap, which is why it's the default. But when you're deep in a complex codebase or a long session and Claude suddenly forgets something you discussed ten minutes ago? That's often Haiku losing track, not your main model failing.

The fix is straightforward: upgrade your data manager. Add an environment variable called ANTHROPIC_DEFAULT_HAIKU_MODEL and set it to a more capable model. At dymaptic, we swap in Sonnet for this role—going from $1 to $3 per million tokens—and occasionally even Opus when the project demands it. The difference is having a model with real thinking ability managing your context instead of just keeping up [with it]. If Claude Code has been letting you down on bigger projects, this might be why.

Another thing I made with Claude's help.

Subscribe to Almost Entirely Human

Don’t miss out on the latest issues. Sign up now to get access to the library of members-only issues.
jamie@example.com
Subscribe