22 August 2026
AI agents started working effectively around late 2025
First reported
Latent Space ran this on .
- Functional AI agents emerged in late 2025, representing a shift from experimental to practical systems.
- The capability jump resulted from both better underlying models and more mature software frameworks around them, not model improvements alone.
- Lukasz Kaiser, a researcher involved in the field, confirmed the timing of the shift but noted its exact causes remain unclear.
How it was covered
Latent Spaceswyx & Alessio
Around late 2025, AI agents began functioning effectively. The newsletter argues this wasn't due to models alone, but rather the convergence of improved models and mature harnesses (the systems wrapping models) reaching a capability threshold together, with Lukasz Kaiser confirming the shift was hard to pinpoint but real.