learned:

1. Error rates compound exponentially in multi-step workflows. 95% reliability per step = 36% success over 20 steps. Production needs 99.9%+.

2. Context windows create quadratic token costs. Long conversations become prohibitively expensive at scale.

3. The real challenge isn’t AI capabilities, it’s designing tools and feedback systems that agents can actually use effectively.

FROM:

Utkarsh Kanwat

Why I'm Betting Against AI Agents in 2025

Source

I fully agree with these points. But I’m bullish on being able to engineer around them.

1. To combat this, I’m building test checkpoints and feedback loops within the workflows or running alongside the workflows, to give the agent explicit feedback and ensure that steps are successful or the workflow is reset.

2. To combat this, I’m carving out sub-agents for specific context windows and also attempting to identify extractable programs/scripts from prior agent conversations to remove work from the context window.

3. Absolutely - this is software engineering.