Designing Airo — then outgrowing it

We shipped a new way of working with AI in 2025. Within a year, multi-agent architecture made it obsolete — so I proposed its successor instead of defending it.

Role
VP of Product Design & Research
Team
Product design, research, AI platform
Timeframe
2025–2026
Outcome
Multi-agent successor proposal adopted [VERIFY] status

The bet

Airo was our answer to a real question: what should working with AI in an enterprise automation platform feel like? Not a chat box bolted onto a recipe editor — a genuinely new working surface. We designed it, shipped it, and it was right for 2025.

The ground moved

By early 2026, the models had changed what was possible. Single-assistant interaction was giving way to teams of agents that could own execution end-to-end. The product I had championed was becoming the incumbent that new thinking had to route around. I wrote the critique myself — nine sections on why our own product didn't feel agent-first — because if that document was going to exist, it should come from the person with the most to lose.

Proposing the successor

I prototyped the alternative in code: a workspace where humans set goals and agent teams own execution — live activity tickers for ambient visibility, delegation instead of supervision, status you can glance instead of babysit. [CLEARANCE] Serene prototype details. The test I used for every screen: she gave goals, the agent team owned execution. If a design required babysitting, it failed.

The most useful thing a design leader can do with a product they championed is notice, first and out loud, when the ground has moved underneath it.

Outcome

[VERIFY] where the proposal landed, adoption status, any metrics

9-section critique
Working prototypes in code
[VERIFY] adoption metric
[IMAGE] Serene workspace prototype (abstracted)
[IMAGE] agent activity ticker