The Messy Middle of Enterprise AI
Every morning, I read another article about AI.
A new model.
A new agent.
Another company claiming they've transformed the way work gets done.
Then I walk into work.
The conversations inside most enterprise organizations sound very different.
They're not asking whether AI is the future.
They're asking much more practical questions.
How do we trust it?
Where does it fit into our workflow?
How do we use it without creating more work?
How do we bring an entire organization along?
That's the messy middle of AI adoption.
And honestly, I think it's the most interesting part.
Over the last year, my team has been experimenting with AI across our design workflow.
Like many teams, we started by treating AI like another tool.
Generate some code.
Write some copy.
Summarize a document.
It was helpful.
But it wasn't transformational.
The breakthrough happened when we stopped thinking about AI as a tool and started thinking about it as another teammate.
That completely changed the questions we were asking.
Instead of asking...
"What prompt should we use?"
We started asking...
"What would a new teammate need to know to do great work here?"
That led us to build a bridge between Figma and GitLab Duo, exporting structured design data instead of static screens and connecting it directly to our design system.
Not so AI could generate code.
So it could generate our code.
The same components.
The same patterns.
The same engineering standards.
The same decisions our team had already made.
Then we kept going.
If AI could understand our design system...
Could it understand our content standards?
Our accessibility guidelines?
Our research?
Our personas?
Could it review edge cases before we did?
Could it create pull requests that matched our engineering conventions?
Instead of building one AI tool, we started building specialized teammates.
Each one understood a different part of how our organization works.
That's when something unexpected happened.
The technology almost faded into the background.
The real challenge became organizational.
Our best designers had years of experience that had never been documented.
Research lived in slide decks.
Content standards lived in PDFs.
Engineering knowledge lived in code.
Everyone knew something valuable.
Very little of it was connected.
AI didn't create that problem.
It exposed it.
Looking back, I don't think enterprise AI adoption is primarily about models.
It's about organizational memory.
The teams seeing the biggest gains aren't necessarily using the newest AI.
They're the teams investing in shared knowledge.
Connected systems.
Reusable patterns.
Clear standards.
Because that's what allows AI—and people—to work together effectively.
For years, we treated knowledge as documentation.
Something you created at the end of a project.
Something that slowly became outdated.
AI has changed that.
Knowledge is no longer just documentation.
It's infrastructure.
It powers design.
It accelerates engineering.
It improves consistency.
It scales experience.
And maybe that's the biggest lesson I've learned over the 2 years.