Goals
Q3 2026 Objectives
Everything you need to build a really good agent
PostHog users have a vision of what they want to automate – in their work, in their life – and today the path from that vision to a working agent is too long and too lumpy. If we close that gap, we unblock a huge category of users who otherwise give up or build something fragile themselves. The win condition is: someone with an idea can get to a working agent from the app, Slack, MCP, or PostHog Code without hitting a wall.
What we'll ship:
- Agent builder platform core – the underlying primitives every agent builder surface depends on
- Agent builder user experience – could be concierge, could be PostHog Code, could be MCP; the surface that turns an idea into a working agent
- Integrations with Slack, GitHub, Discord, and other useful tools – agents need to read from and act on the places where work actually happens
- End-to-end smoothness – connections work as close to first-try as possible; we don't ship something that requires babysitting to set up
- MCP parity – anything you can do in the agent builder, you can do via MCP (e.g., from Claude Code)
The agent improves itself based on product data
The better the agent builder gets, the more people will build agents – and the more important it becomes for those agents to improve themselves. If every agent needs a human to fix it every time it goes off the rails, building agents collapses into babysitting agents, and we lose. We want a loop where product data flows back into the agent and makes it better without the builder having to intervene.
What we'll ship:
- Agent observability loop with AI Observability integration – every agent emits the traces, evals, and signals needed to know what's working and what isn't
- Agent self-improvement loop – agents use that data to improve themselves, so builders aren't the bottleneck on every regression
Handbook
We build the tools that let anyone go from "I wish I had an agent for that" to a working agent in production – without having to babysit it.
As of June 2026, that covers two things:
- Agent builder – a smooth, end-to-end experience for creating agents from wherever people already work (their CLI, the PostHog app, Slack, MCP, PostHog Code)
- Self-improving agents – agents that get better on their own using product data, so builders aren't stuck babysitting problems the agent could solve itself.
Slack channel
Feature ownership
You can find out more about the features we own.