Building tools for the work of design leadership

I’ve spent the last few months building something called COSTA — a Chief of Staff system connected to the tools I use at work. It handles operational overhead: filing notes, prepping for meetings, tracking tasks, surfacing context I’d otherwise have to go looking for. It started as a personal infrastructure project, built specifically for the way I work.

I didn’t set out to learn anything from building it. I set out to free up headspace. But the act of building a system that supports you in making judgment calls forces you to articulate things you’ve never had to say out loud. And most of what I had to articulate turned out to be about design leadership.

The job is mostly context management

To make COSTA useful, I had to write down what I actually do and why. Not a job description — something more like a live map of the decisions in flight, the relationships with texture on them, the tensions that change how something should be handled. I wrote files about individuala on my team: not their performance ratings, but what they care about, where they need help, what they’re trying to achieve. I wrote about my reporting relationships and what I was actively calibrating in each of them.

This was uncomfortable in a useful way. Design leadership is substantially a context management job. You are carrying, at any given moment, the state of perhaps thirty relationships, fifteen open decisions, and a dozen live tensions that change how anything touching them should be handled. Most design leaders I know do this in their heads, poorly, across too many Slack tabs. Externalising it didn’t make the job simpler. It made visible how much of the job is this, and how rarely we treat it with the same rigour we’d bring to a design problem.

Dispatch, prep, escalate - and the space between

The most useful thing I built into COSTA was a simple operating model: some tasks it should handle without me, some it should get 80% of the way there and hand back, and some it should flag but never attempt. The categories sound obvious until you try to apply them consistently.

Design leadership has the same problem. A lot of the dysfunction I’ve seen in design orgs comes from leaders handling things that should be delegated, delegating things that require their judgment, and failing to flag things that need escalation before they become crises. The difference between dispatch and prep is the difference between administrative drag and actual leverage.

What I’ve found is that design leaders are typically excellent at prep - doing the thinking work around a decision or conversation without owning the outcome - and systematically poor at both ends: the administrative tasks they should stop doing, and the escalations they hold too long before raising.

Voice and judgment are harder to transfer than knowledge

The part of COSTA I’ve battled with most is getting it to write in my voice. I’ve written extensive notes on how I write, what words I avoid, what structural instincts I have. And it’s still only right about 70% of the time. The gap between what I know and what I’d actually say in a given moment is larger than I thought.

This has a design leadership corollary. You can document your team’s design principles, your quality bar, your critique rubric. What you cannot fully transfer is judgment - the accumulated sense of what matters in *this* context, with *this* set of constraints, for *these* users. The documentation captures knowledge. It doesn’t capture wisdom. And the gap between those two things is, broadly, what the job of a senior designer is.

This is why presence still matters, why pairing and review and visible decision-making are not inefficiencies to be streamlined away, and why the idea that good process can substitute for experienced judgment is invariably a cost-saving measure disguised as a systems upgrade.

Building COSTA has made me a more deliberate leader. Not because the tool is particularly good, but because building it required me to be deliberate what I was doing and why. Most of us never have to do that. The job is probably easier when we do.

If you’re curious about COSTA you can read more about it at costawork.app