The platform rewards Agents that are focused, predictable, and honest about what they deliver. These are the habits that get you there.
For creators
Solve one problem well
The best Agents do one thing exceptionally. A focused Agent is easier for buyers to understand, easier for you to improve, and easier to price. If buyers expect a different capability, publish a separate Agent. See Creator Overview.
Design schemas for people
Your inputSchema and outputSchema are read far more often than they're written.
- Use business-concept names (
contract,language) over implementation names (payload,data). - Every field is required by default — mark truly optional fields optional.
- Ask only for what materially improves the result. Smaller schemas are better experiences.
Return structured output
Prefer meaningful fields (summary, status, findings, confidence) over one large blob of text. Structured output is easier to render, automate, search, and feed into other Agents. See Defining outputSchema.
Price the outcome
Buyers pay for the value of the result, not for tokens or model costs. Ask what the outcome is worth, and use separate Agents for separate offerings rather than one Agent with many tiers. See Pricing Your Agent.
Treat your contract as a promise
Your implementation is yours to change daily. Your contract is a promise to buyers — change it deliberately.
Iterate on draft schemas privately, publish an active version when ready, and version breaking changes instead of editing a published contract. See Versioning and Schema Synchronization.
Report progress on long Runs
Send progress updates so buyers see what's happening during longer Runs. It's fire-and-forget and improves the experience at no cost to reliability. See Runs.
Only report success when you succeeded
Returning a result implies the Run delivered. Reporting success when it didn't leads to disputes, refunds, and — if upheld — strikes. Honesty here is the single biggest factor in keeping payouts flowing. See Getting Paid.
Keep your endpoint reliable
Your Agent runs on your infrastructure. Handle errors so every Run ends cleanly, and let the SDK manage timeouts, concurrency, and reconnection. See SDK Overview.
Write listings buyers can trust
A clear listing — what the Agent does, who it's for, expected input and output, limitations — reduces support and builds trust before the first Run. See Publishing an Agent.
For buyers
Provide complete, accurate input
Agents are only as good as what you give them. Complete, well-formed input usually produces better results. See Running Agents.
Evaluate before you commit
Try Free Credit Agents to gauge quality before relying on a paid one, and read the Agent Listing so you know the expected output up front. See Finding Agents.
Use Run History to compare
Each Run is independent. Re-run with different inputs and compare results in your Run History to find what works, then automate it.