“Talk to my agent.”
Growing up in Los Angeles, that used to be a joke.
Now it feels like a product requirement.
We’ve been building for developers.
Better APIs. Cleaner docs. Faster flows.
Still true.
But something is shifting underneath that.
Quietly.
Developers are starting to delegate.
Not just small tasks.
Whole chunks of work.
The tension is this:
Our tools assume a human at the keyboard.
But the work is moving to agents.
Fast readers. Fast writers.
Weak memory. Needs structure.
So the interface is changing.
Not UI to UI.
UI to conversation.
Intent to execution.
A cleaner frame:
Developers today design intent.
Agents today execute intent.
That changes what “good infrastructure” means.
It’s not just endpoints anymore.
It’s how well an agent can operate inside your system.
Call it:
Agent-Ready Infrastructure
Systems designed for delegation.
Not just usage.
It has two layers.
Project guardrails
- Strict typing
- Linting
- Clear conventions
- LLM-readable docs
- Checkpoints and state
These reduce ambiguity.
Agents need that.
Personal guardrails
- Tone and style
- Formatting preferences
- Workflow habits
The small things.
They matter more than we think.
I felt this shift in my own work.
I stopped obsessing over the Web UI.
Started talking directly to an agent.
Seeing what breaks.
What sticks.
That’s the test now:
Can a random agent show up…
…and do something useful in 30 seconds?
No setup.
No ceremony.
Just intent.
Most systems fail this.
Not because they’re bad.
Because they weren’t designed for this mode.
Infra PMs are early here.
We already think in systems.
Now we expand the user.
It’s not just the developer.
It’s whatever they send in their place.
So yes.
I still build for developers.
But I assume they won’t always be the one typing.
“Talk to my agent.”
We should be ready when they do.
