> For the complete documentation index, see [llms.txt](https://docs.zerve.ai/zerve-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.zerve.ai/zerve-docs/superagents/best-practices.md).

# Best practices

How to get the best results from a superagent.

These practices come from how superagents are designed to work. They'll help you get reliable results, and spend your budget where it counts.

## Starting out

**Set up your data first.** The planner looks at your organization's [data connections](/zerve-docs/data-and-assets/connections.md), [constants and secrets](/zerve-docs/data-and-assets/constants-and-secrets.md) and [skills](/zerve-docs/agents-in-the-workspace/skills.md) when it drafts a plan. A superagent that can see your warehouse and your data provider keys makes a far more grounded plan than one that has to ask for them.

**Describe the outcome, not the steps.** "Find which momentum signals still work in European equities, and why" gives the superagent room to plan well. A list of queries to run doesn't. See [Writing a good mission](/zerve-docs/superagents/create-a-superagent.md#writing-a-good-mission).

**Start focused.** One objective, done well, beats five done badly. The planner usually proposes one objective for this reason, so resist adding more until the first is working.

**Set a first milestone as the KPI.** A target you can hit and measure soon, such as "a signal with an out-of-sample Sharpe above 0.8 on the last three years," is more useful than an ambitious end state. Raise the bar as the work matures.

## Staying in control

**Size autonomy to the risk.** Use **Full autonomy** for low-risk research, and **Ask me first** for anything irreversible or client-facing. Autonomy is set per objective, so you can mix them. See [Autonomy](/zerve-docs/superagents/objectives-kpis-and-autonomy.md#autonomy).

**Write guardrails for the things that must never happen.** For example: "Never trade or recommend trading," "Only use licensed data," or "Never send anything outside the firm." See [Persona and guardrails](/zerve-docs/superagents/settings.md#persona-and-guardrails).

**Respond to decisions promptly.** A decision marked **Need input** is holding up work. A quick answer keeps the superagent moving. See [Decisions and recommendations](/zerve-docs/superagents/decisions-and-recommendations.md).

## Trusting the results

**Ask how it checked.** For any result that matters, ask the superagent how it verified it. It can have a number independently re-derived from source, or have a whole notebook reviewed for problems. See [They check the results](/zerve-docs/superagents/superagents/how-superagents-approach-work.md#they-check-the-results).

**Look at the work.** Every worker has a real notebook. Open it from the Map to see exactly how a result was produced.

**Ask for a held-back test set.** When the superagent is optimizing against data, as in a [Hill-climb](/zerve-docs/superagents/batches/hill-climb.md), ask it to hold back part of the data as a test set, so you can tell whether the result will hold up.

## Making it last

**Let it compete ideas.** When you're not sure which approach is best, ask the superagent to try several and keep the winner, rather than betting on one. See [Batches](/zerve-docs/superagents/batches.md).

**Tell it your preferences once.** State a preference or constraint in a conversation and the superagent remembers it. See [Memory](/zerve-docs/superagents/memory.md).

**Use schedules for ongoing work.** A weekly briefing or a daily check belongs in a [schedule](/zerve-docs/superagents/schedules.md), so it happens without anyone asking.

**Keep objectives up to date.** Mark objectives achieved or abandoned as things change. A superagent with a small, current set of objectives works better than one with a long, stale list.

## Managing cost

**Match the model to the task.** Set a cheaper model for small-effort worker tasks and a stronger one for large-effort tasks. See [Model and effort](/zerve-docs/superagents/settings.md#model-and-effort).

**Set a daily budget that fits.** The default is the maximum, 25,000 credits a day. For a superagent doing light, recurring work, a lower budget avoids surprises. See [Budgets and credits](/zerve-docs/superagents/budgets-and-credits.md).

**Pause what you're not using.** Pausing stops all work and scheduled runs immediately, and a message resumes it.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.zerve.ai/zerve-docs/superagents/best-practices.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.