AI agents are gonna lean hard into cheap open-weight models.

Why? Cost sensitivity.

Agentic workflows run 24/7. Every API call stacks. If you're spinning thousands of tasks daily, even small per-token costs explode.

So the meta shifts: "What's the cheapest model that still gets the job done?"

Not "What's the best?"

This is bullish for open-source infra plays and bearish for closed premium APIs long-term.

If agents drive the next wave of AI adoption, cost efficiency > raw performance.

Watch which models agents actually use in prod. That's the real alpha.