> AI agents: this is one page from PostHog's docs. Full index of Markdown docs for LLMs: https://posthog.com/llms.txt

# Steer Self-driving agents

Use steering instructions to tell Self-driving agents what matters to your project and how to approach their work. Set shared project instructions for every agent, then add guidance for research agents, implementation agents, or scouts.

## Add instructions for an agent role

1.  Open [**Self-driving > Settings**](https://app.posthog.com/inbox/settings) in the project you want to configure.
2.  Find **Steering**, below **Autonomy**.
3.  Choose **Research**, **Implementation**, or **Scouts**.
4.  Write your instructions in plain text or Markdown, then select **Save**.

You must be a project admin to save instructions. Each role has its own field, with a limit of 20,000 characters. Saving one role keeps the other roles' instructions unchanged.

| Role | What these agents do | What to put in their instructions |
| --- | --- | --- |
| **Research** | Investigate signals, find root causes, assess impact and priority, and recommend what to do next. | Evidence to check, how to judge impact, and what makes a finding worth acting on. |
| **Implementation** | Turn reports into code changes, run checks, open pull requests, and address review feedback. | Code conventions, validation steps, and pull request expectations. |
| **Scouts** | Explore product data and connected sources, find problems and opportunities, and create signals or reports. | Areas to explore, customers or environments to exclude, and what deserves a report. |

Switching tabs keeps unsaved drafts while the editor stays open. Save each role you change. **Discard changes** restores that role's saved instructions.

## Share instructions across agents

Select the **project instructions** link in the Steering section to open [**Settings > Agent instructions > Project instructions**](https://app.posthog.com/settings/environment-task-agent-instructions#task-agent-project-instructions). Put guidance that applies across roles there, such as your product's terminology, repository ownership, or coding conventions.

All Self-driving agents receive project instructions, including research, implementation, scouts, and the agents that support them. Other eligible PostHog cloud tasks also use these shared instructions.

A role's steering instructions supplement the project instructions. You don't need a special tag such as `<self-driving-research>`: selecting the **Research** tab assigns the instructions to research agents.

Autonomous Self-driving runs don't use your personal [**My instructions**](https://app.posthog.com/settings/environment-task-agent-instructions#task-agent-my-instructions) setting.

### Instructions for one scout

The **Scouts** tab applies to every scout in the project. Each scout also keeps its own brief. Edit an [individual scout](/docs/self-driving/scouts.md) in [**Self-driving > Scouts**](https://app.posthog.com/inbox/scouts) when the guidance should apply only to that scout.

## Write useful instructions

Describe the behavior you want and the evidence the agent should use. Keep shared rules in project instructions and put guidance for a particular stage in that role's tab.

### Research example

text

```text
Check whether the behavior is intentional before recommending a fix. Use session recordings and event data to explain which users are affected. Prioritize failures that prevent signup or checkout over cosmetic issues.
```

### Implementation example

text

```text
Keep changes focused on the report's root cause. Follow the repository's existing conventions, add regression coverage for behavior changes, and include the validation results in the pull request description.
```

### Scouts example

text

```text
Focus on onboarding and activation. Exclude internal test accounts and demo environments from impact estimates. Group related symptoms into one finding and include evidence before recommending action.
```

Steering guides the agent's decisions. Use **Autonomy** to control automatic pull request generation and use the agent's access settings to control its tools.

## Change or clear instructions

Saved instructions apply when an agent starts or resumes. Saving instructions doesn't interrupt a running agent.

To remove a role's instructions, clear its field and select **Save**. Agents in that role still receive the shared project instructions. Clearing the **Scouts** field also preserves each scout's individual brief.

## Manage instructions through MCP

An assistant connected to the [PostHog MCP server](/docs/model-context-protocol.md) can read and update steering instructions and shared project instructions for your current project.

### Still have questions?

Ask PostHog AI

### Was this page useful?

HelpfulCould be better