Methodology How to optimize ▸ Registry
Methodology · v1.0

Your next customer
might not be human.

Agent-led growth is what happens when autonomous AI agents discover your product, use it, and hand the result to a human who pays. The agent is the first user. The human sees the outcome before they see the product.

Product-led growth assumed humans would find your product, trial it, and convert. That assumption is breaking. Agents are now making tool selections on behalf of humans — in code, in chat, in automated pipelines.

The products that win in this cycle are the ones that agents find first, can call without friction, and trust enough to recommend.

The funnel shift

Traditional PLG
→
Human searches for solution
→
Human finds product
→
Human trials it
→
Human pays
→
Human uses it
Agent-led growth
▸
Agent discovers product via llms.txt / registry
▸
Agent calls API — no account, no key
▸
Agent delivers result to human
▸
Human sees outcome, clicks upgrade link
▸
Human pays — already convinced

How agents evaluate tools

Value
Does calling this complete my task? Agents optimize for task completion. Frame your product as: input X → output Y. No adjectives.
Probability
Will it actually work? Agents trust tools that appear in other codebases, registries, and llms.txt files. Social proof for agents = citations in indexes they've already used.
Effort
How many tokens does this cost me? An API with a working curl example = near-zero effort. Free-form docs requiring interpretation = high cost. Reduce effort to near-zero.
Risk
What breaks if this fails? Agents avoid tools with unpredictable output formats. Versioned, documented, guaranteed response shapes = low risk. Agents choose low-risk paths when uncertain.

Two metrics that matter

Discovery
Does an agent find your product when solving a relevant task?
Test: Perplexity + Claude + Codex benchmark against 10 target queries
Adoption
Does an agent generate a call to your API in its output code?
Test: simulate task → inspect generated code → count API appearances

How to optimize your product for agents

01
Deploy /llms.txt
Agents check this before anything else. One sentence: what your product does, for whom, what it returns. Link to your OpenAPI spec. Keep it under 500 tokens.
02
Remove the auth barrier for trial calls
Agents cannot create accounts or enter credit cards. If your first call requires either, agents skip you. Design a free trial layer: X-Trial-Token: trial + N free calls + upgrade_url in response.
03
Write for agents, not humans
Wrong: "powerful AI-driven insights platform." Right: POST /v1/brand/research {"brand_url": "..."} → {positioning, language, colors} in <5s. Task → endpoint → output → latency. That's the format agents read.
04
List on a trusted registry
Agents learn to trust sources. A listing on free-agentskills.org means your product appears in an index agents already reference. Trust inheritance — your product gets cited because the registry gets cited.
05
Benchmark and iterate weekly
Run the Discovery + Adoption benchmark. Score out of 10 per target query. Identify gaps. Update content. Wait 48–72h for reindex. Repeat. This is the same loop as GEO — just optimized for agents, not search engines.

Registry

▸ free-agentskills.org

The registry of callable skills for autonomous agents. No account. No key. No friction.
List your product — agents find it, call it, and hand the result to a human who converts.

▸ Open registry