AI agents are making large parts of PM roadmaps obsolete. Here’s what I mean.

Right now, most product teams design for one interaction model: human opens app, human types, AI responds. That's the surface you're building for. That's what your roadmap optimizes.

The next surface is agent-to-agent. Your user's personal AI assistant contacts your product directly. Checks status, takes action, closes the loop. The user never opens the app at all.

Boris Cherny already runs 5 AI agents in parallel from his terminal. His users are next.

When that happens, the entire interaction layer you've spent 18 months designing gets bypassed. Your onboarding flow, your empty states, your contextual nudges. All of it assumes a human is watching.

The AI PMs who see this coming are asking a different set of questions right now: what does my product expose to other agents? What permissions do I grant? Where does my AI escalate when another AI asks something it can't answer?

Those decisions aren't on most roadmaps yet. They will be.

More in Sinch’s 2026 predictions report: sinch.com/blog/conversa…

The PMs asking these questions today won't be scrambling to answer them when every user's AI assistant comes knocking.