Agentic engineering is optimising the wrong metric

We're building Tempest because we think agentic engineering is heading in the wrong direction.

Today, most tools compete by running more agents. More agents sounds impressive, but it also means more duplicated context, more token usage, and more coordination problems.

We think the real challenge is different.

How do you help every agent understand the codebase without making each one rediscover it?

How do you let agents work in parallel without sharing the same workspace or database?

How do you scale useful work instead of scaling token costs?

That's the problem we're solving with Tempest.

I'd love to hear from people who build with Claude Code, Codex, Gemini CLI, Cursor, or other coding agents.

If you could fix one thing about today's multi-agent workflow, what would it be?