Heard an interesting podcast with Shishir Mehrotra — CEO of Grammarly — on the future of workflows. He broke down agents into four components:
- Knowledge — what the agent knows
- Skills — what it can do
- Assignments — what it’s been asked to handle
- Soul — its personality and approach
That framing stuck with me. Because it’s exactly how I think about building agents in client workspaces.
Agents as Digital Teammates
We’re not building automation scripts. We’re building digital teammates. And a good teammate doesn’t just execute — they understand context, work within constraints, and know when to ask before acting.
The soul component is what most builders skip. But a soulless agent is just a script with extra steps. Agents need to be tailor-made to client use cases — capturing their unique workflow, their terminology, their decision points.
The Human in the Loop
I’m a fan of keeping humans in the loop. Not as a safety net, but as a design principle. As an end-user, I need to understand what happens. As a consultant, I make sure my clients do too.
Helpful where tasks are repetitive. Essential where judgment is required.
Fascinating times. Still learning and evolving in this — which is exactly how it should be.