Even With AI Agents, Someone Still Has to Do It

Even with AI agents, someone still needs to do it.

In this Agentic AI era, the pitch is clear: AI agents will run your business while you sleep. Don't get me wrong: the progress is real. If you've been using LLMs for a while, you've felt it coming. Then agents arrived, and suddenly you could build your own tools.

What really changed for many of us is the cost of trying. Features that were too long or expensive to even test. Tools that would take weeks to build, only to be rejected by your team for poor usability. Now you can test four ideas in parallel and iterate on custom tools in days.

But here's the important part: agents aren't (yet) autopilot. Someone still has to explain, steer, test, steer again, find the bugs, and know when to stop. This is the new bottleneck.

That "someone" needs three things at once: fluency across the whole software lifecycle to avoid wasting time and prevent obvious architecture mistakes, the business sense to know what's actually worth building, and the flexibility to work surgically, fast, and close to the person who needs it.

Agents made building cheap. Judgment and follow-through are where the value is.

(And if that "someone" is what you're missing, you know where to find us. 😉)