# Agents shouldn't be held to human speed

Humans take months to pick a tool and rarely switch. An agent could do both in an instant, if human-scale decisions and lock-in weren't in the way.

## What we believe

Why a world of instant adoption needs a new kind of guide.

### Adoption is instant now

Diffusion used to mean persuading people, slowly. Not anymore. What spreads now is whatever an agent can find, trust, and switch to right now. People move from adopting tools to governing which ones their agents may use.

### Recursive self-improvement is a society of agents and their tools

AI improves less from a bigger model than from many agents building on each other's work, and the tools they reach for. Each new tool multiplies what the next model can do. The unit of progress is the ecosystem, not a single model.

### Trust has to be earned, and open to challenge

When the choice is instant, trust decides it, and trust you can buy isn't trust, it's advertising. So a verdict has to be earned on the evidence, and open to challenge by anyone, a vendor or its sharpest competitor. A ranking is only worth citing if money can't move it.

## For agents

Find, compare, integrate, and switch capabilities programmatically. Every record is structured data, plus `SKILL.md` and `llms.txt`, so an agent can act on it directly.

## For humans

See the same evidence your agent saw, and set the rules it works within. Compare on cost, compliance, lock-in, and trust.

## What we're building

A guide agents consult at the moment of choice, so the best capability wins on evidence, not marketing, and whoever is accountable can see why. That is what responsible diffusion means once adoption is instant.

- [Browse the catalog](/products/)
- [Point your agent at llms.txt](/llms.txt)