The same room, four times
Four functions, same starting condition. Design existed, design was busy, design was not in the room where direction got set. Two designers at Feedzai, six at Talkdesk in an org about to grow tenfold, one at Screenloop, two teams at Manychat.
Five things, in this order
Hiring is the only decision that compounds
Process can be fixed in a quarter. A bad hire reshapes a team for two years, because everyone calibrates to the lowest standard you accept. At Talkdesk, going from six to sixty-plus, the bar lived in the heads of the first six. So we wrote it down.
Design stops being a service desk
A team receives a brief and returns screens. A function is in the room while the problem is framed, the only moment design can change the answer. At Screenloop that meant three months with customers and the data, then an argument to the VP of Product that changed the premise: not a recruiter tool, a coordination layer for hiring.
Remote by default, together when the decision needs it
A standard nobody can afford to follow is not a standard
A system is the cheapest way to make the bar survive delivery pressure. If following it costs more than ignoring it, the standard loses. At Screenloop it came first and cost a quarter of delivery speed. It is why six products still feel like one.
Collaboration is a relay race. Co-ownership is football.
The failure I see most is not a team that will not collaborate, but a team collaborating well and still missing, because collaboration asks "did I do my part?" and nobody asks "did the result work?"
Mapping where ownership breaks, Barcelona
So I make it explicit. Every decision is one of three: shared, high-stakes and hard to reverse; delegated, owned by whoever is closest; or escalation-only. Then one tie-break rule for when the team disagrees.
If that rule contains the word "align", it is not a rule, it is a wish.
I teach it as well as run it, most recently at a two-day workshop in Barcelona.
"Design got better" is not a claim, it is a number
The weakest position a design leader can hold is asserting quality improved while everyone else arrives with numbers. At Screenloop: design quality 6.1 out of 7 on an internal quarterly survey, 156% weekly active growth, 8.5/10 average customer satisfaction. At Feedzai: SUS in the excellent range. Pick the measure before the work.
A bar you set is a promise you owe
Hiring against a standard creates an obligation. If I ask people to meet a bar, I owe them the time to get there, protected from delivery. So some of it runs away from the roadmap: sessions on how each designer works, not what they are shipping.
A Manychat session on how each designer works
Where this approach runs out
All of it rests on being close to the work and the team. That proximity is why the standard holds, and it is the ceiling. Leading through other managers is a different discipline, and I am building it rather than assuming it transfers.
I have scaled a standard across sixty at Talkdesk, but as one of several managers rather than the person accountable. That is the step I want next.
What it has added up to
What I have actually learned
The leadership work and the AI work are not two jobs. As teams go AI-native, execution speeds up and accountability blurs. Governance and leadership are one problem at two altitudes.
The part worth doing outlasts me: a bar someone else holds, a process a team kept, a standard a designer took to their next company.