PostHog Self-driving
AI agent · Coding agent · Engineering & development by PostHog
Automate the loop of diagnosing, fixing, and verifying software improvements.
PostHog Self-driving is an autonomous agent that identifies bugs from product signals and automatically generates pull requests for engineers to review. It helps engineering teams automate the loop of diagnosing, fixing, and verifying software improvements.
Agent specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Price
- From 20 USD value)
- Languages
- English
Category: AI Agents
- Channels
- Web app, PostHog Desktop, GitHub
- Integrations
- GitHub, Zendesk, Linear, Stripe, HubSpot
- Editors and tools
- PostHog Desktop, Cursor, Claude Code, VS Code, Zed
- Deployment
- Cloud (US and EU regions)
- Primary Function
- Automated bug fixing and PR generation
- Input Signals
- Errors, replays, support tickets, GitHub issues
- Automated pull request generation for bugs and UX issues
- Monitors errors, session replays, and support tickets 24/7
- Verifies fixes automatically after they are shipped
- Human-in-the-loop review for all proposed code changes
Overview
PostHog Self-driving is an autonomous software maintenance agent designed for product engineering teams to automate the bug-fixing lifecycle. It proactively identifies issues by monitoring signals from error tracking, session replays, and support tickets, then researches the root cause using context from external tools like GitHub and Zendesk. Once a problem is diagnosed, the agent automatically generates a pull request with a proposed fix for engineers to review and merge. After a change is deployed, the agent monitors the product to verify the fix was successful and restarts the loop if new issues are detected. This system allows teams to automate the repetitive tasks of diagnosing and fixing software improvements while maintaining human oversight.
Issue Diagnosis and Research
- Signal Monitoring: The agent monitors signals from error tracking, session replays, and support tickets to proactively identify bugs and UX friction.
- UX Issue Detection: PostHog Self-driving identifies issues such as dead clicks, quick backs, and long stalls within session recordings.
- Error Grouping: Exceptions and stack traces are automatically grouped into actionable issues for the agent to investigate.
- External Tool Integration: The agent pulls context from external tools including Zendesk, Linear, and GitHub issues to research problems.
Code Generation and Shipping
- Automated Pull Requests: When a report is ready, the agent automatically writes the pull request to fix the identified issue.
- Feature Instrumentation: The agent builds new features with proper PostHog instrumentation included in the code.
- Human-in-the-loop Review: Engineers maintain control by reviewing and merging the proposed pull requests before they ship.
- Parallel Agent Management: PostHog Desktop allows users to manage multiple AI agents in parallel during the development process.
Verification and Monitoring
- Post-deployment Verification: After a change is shipped, the agent automatically checks whether the fix worked as intended.
- Continuous Improvement Loop: Any new problem detected after a ship starts the diagnostic and fixing loop again.
- Impact Validation: The agent checks whether a feature worked after shipping to validate the impact of the change.
What it does
- Monitor signals from errors and replays\nResearch root causes using external context\nGenerate pull requests with proposed fixes\nVerify fixes after deployment
Use cases
- Engineering teams use the agent to automatically generate pull requests for high-priority sign-in errors identified in logs.
- Product managers monitor the Inbox for researched reports on checkout issues found through session replay analysis.
- Developers use the agent to investigate root causes of support tickets by pulling context from Zendesk and Linear automatically.
- Teams verify the success of a bug fix as the agent monitors product signals post-deployment to ensure the issue is resolved.
Specifications
- Channels
- Web app, PostHog Desktop, GitHub
- Integrations
- GitHub, Zendesk, Linear, Stripe, HubSpot
- Editors and tools
- PostHog Desktop, Cursor, Claude Code, VS Code, Zed
- Human control
- Engineers review and merge every pull request before it ships
- Deployment
- Cloud (US and EU regions)
- Setup
- Setup wizard installs PostHog and connects agents via terminal command
- Primary Function
- Automated bug fixing and PR generation
- Input Signals
- Errors, replays, support tickets, GitHub issues
- Control Model
- Human-in-the-loop (Review and Merge)
- Supported Platforms
- Web, GitHub, PostHog Desktop
- Integration Protocol
- Model Context Protocol (MCP)
- Free Tier Allowance
- 2,000 credits per month for PostHog Desktop usage
How it works
- Data Ingestion: PostHog ingests signals from users and AI agents, combining them with revenue, CRM, and support data from tools like Stripe and Zendesk.
- Issue Reporting: Scouts monitor the product 24/7 and file researched reports in the Inbox, ranking them by priority with full context.
- Pull Request Generation: When a report is ready, the agent writes a pull request that an engineer reviews and merges.
- Verification and Feedback: After the change ships, the agent checks whether it worked and restarts the loop if new problems arise.
Pricing
- PostHog Desktop (includes agent credits)
- 2,000 credits free per month ($20 value)
FAQ
What signals does the agent use to find bugs?
PostHog Self-driving monitors error tracking, session replays, and support tickets to identify problems. It also pulls context from external tools like Zendesk, Linear, and GitHub issues.
Does the agent ship code without human review?
No, engineers maintain control by reviewing and merging every pull request the agent generates. The agent only writes the code; a human decides what actually ships.
How does the agent verify that a fix worked?
After a change ships, the agent automatically checks whether it worked. If a new problem is detected, it starts the loop again to find a fix.
Is there a specific editor for using these agents?
Yes, the agent can be managed through PostHog Desktop, which is an AI code editor designed for building features with proper instrumentation.
Added by AntSpace from public sources (posthog.com).