Mike's post explains why AMP is for a new kind of work. The least flashy, but necessary foundation: agents get their own identity. Not a shared login. Their own permissions, and a trail you can audit.

We've been hearing that AI is collapsing PM, design, and engineering into one 'AI builder' role. This can work in a small company. Inside a 10,000+ person company, the story is more nuanced. When you have complex codebases, lots of dependencies, and hundreds of thousands of customers, the roles don't converge - they expand. The goal of the PM job is unchanged: set direction, exercise judgment, hold the quality bar, drive outcomes. What's changed is how PMs get there. AI changes how fast you move and how much more you can do. At Lenny & Friends Summit, I shared real examples from inside Atlassian: what's working, what's not, and where we're still figuring it out. Thanks to @lennysan for bringing PM leaders together! Check it out here: 📷

Tomorrow I'll be speaking at the Lenny & Friends Summit 🏕️ I'll be sharing how we're seeing the PM role evolve inside a company of 10k+ people, building enterprise products at scale. Amazing lineup too: Anthropic, OpenAI, Lovable, Linear, and more. The programming will be livestreamed starting at 9:30am PT:

Jira customers have been orchestrating work in automations for years. Now those same automations can trigger a coding agent: an event in your system of record automatically packages the context, fires the agent, and logs the outcome. Same workflow. Much more powerful. Works with Claude Code, Cursor, and Copilot and with the built-in Jira Coding Agent.

Just shipped: Claude Agent for Jira. Built on Anthropic’s Claude Managed Agents infrastructure, Claude Agent for Jira lets you assign work items directly to @claudeai. It reads the task, writes the code, opens a PR. The work item is the prompt. No copy-pasting specs, no branch setup, no prompt crafting. Built with @AnthropicAI x @Atlassian. Learn more: