The future of knowledge work isn’t fully autonomous.
Over the last few weeks, I’ve been experimenting heavily with agents, custom setups, background tasks, layered automations running inside Notion. On the surface, it looked impressive. Workflows running on their own, information processed in seconds, tasks getting handled without me touching them.
But the more I built, the clearer something became.
Building the agent is actually the easy part.
Evaluating its performance inside a complex workflow? That’s where it gets tricky. The more connected your system is, the more tuning it requires. You’re constantly adjusting constraints, refining prompts, testing edge cases to see where things break.
Iteration becomes the real work — because the ongoing calibration to make sure it’s actually doing what you need it to do is where it’s at.
What I Learned About Drift
Without structure, automation doesn’t scale — it drifts. And drift in a knowledge work system means bad outputs, missed context, and decisions made on incomplete information.
There is no question that AI in knowledge work is powerful. But the human in the loop isn’t a safety net — it’s part of the design.
Agents can assist with repetitive tasks, surface patterns, handle the grunt work. But they shouldn’t replace thinking. Just because something can run on its own doesn’t mean it should.
The Question Worth Asking
Still figuring out where that line is for my own workflows.
Where does your system still require judgment?