Your Next User Isn't Human

October 10, 2026

We divided engineering into front end and back end. One built the interface; the other made it work. That split shaped teams, tools, and careers.

Now another split matters: software for humans and software for agents.

Humans need interfaces that help them understand and decide. We design for attention: clear labels, visual hierarchy, useful feedback. We hide details until someone needs them.

Agents need interfaces that help them act and verify. They need to discover available actions, inspect current state, understand failures, and retry without making a mess. Hide the wrong detail and an agent has to guess.

Take an issue tracker. A human needs to see what matters and choose the next task. An agent needs to know which changes it may make, whether an update succeeded, and what to do if the connection drops halfway through. Same product. Different users. Different design work.

We’ve built APIs and command-line tools for decades. The new part isn’t that software calls software. It’s that the caller chooses its next action instead of following an integration path we wrote in advance. Discovery and recovery become part of the interface, not chores we leave to the integration author.

Agent-facing engineering isn’t human-facing engineering with a JSON flag, any more than back-end engineering is front-end engineering without CSS. It has its own constraints and deserves the same care.

This doesn’t replace front end and back end. It cuts across them. Where the software runs still matters. Who uses it now matters just as much.

Build for both. Don’t make either user scrape the other’s interface.