Eryk Orłowski · hretheumhretheum.com/questions/ai-in-the-team
Question P5 of 8
All questions5How do they bring AI into a team's work?
At Efigence, leads a team of about 10 people and introduces AI tools into its work gradually and voluntarily, in line with the company's official AI use policy. Treats the concern that AI gives little creative results as valid and shows a division of work in which creativity stays with the designer. Came up with Skills Factory, a tool library that carries a design system into code; built the first versions alone, and the team goes on extending it. One senior from the team, together with their lead, built a two-way pipeline between Claude Design and Figma.source sources: ef-01, om-04, ef-08, ef-05, ef-04, ef-07, ef-11, ef-03
Evidence
AI adoption in the team is gradual and voluntary.”
We treat the concern that AI produces uncreative work as accurate and show a division of work in which creativity stays with the designer.”
I proposed the idea and the name of AI Sagres: designers, business analysts and technology people at one table.”
In Gatefold the model proposes and a person makes every decision; the only signals that count are "I'll take this one" and "this is not it".”
Skills Factory was my idea, and I built the first versions myself. A team at Efigence now develops it further, following my idea, under a different name.”
Skills Factory is a library of tools on an adapter pattern that carries the design system into code.”
At Efigence I lead a design team and bring AI tools into its work.”
I work on AI in 3 areas: AI in the design process, a product or tool with AI, and design systems with AI.”
I took part in launching AI Sagres, the group that leads AI projects at Efigence.”
I was a panellist in a discussion on the agentic economy at ExFest, taking a sceptical position.”
At AI Lessons Unlearned I gave the keynote "Who will deliver your 2030 strategy?".”
In my talk at AI Lessons Unlearned I argued that every organisation needs its own Sagres, a space where people from different silos can think together about things that do not exist yet.”
A lesson in Gatefold only counts after a person accepts it, and the next handoff picks it up automatically.”
I set the scope, rules, ethical limits and phase acceptance gates, and the model writes the code, tests and first drafts of documentation.”
Juniors get to decent-looking craft much faster than before. […]Juniors get to decent-looking craft much faster than before. A screen that used to take a week of practice now takes an afternoon with the right tools. The catch is that it's also much easier to skip the part where you understand the problem, because the tool produces something plausible before you've asked a single good question. So I think juniors need more mentoring now, not less. Someone has to keep asking them why this solution, for whom, and what they checked, until they start asking it themselves.”
More broadly, what's overhyped is the expectation that one prompt, without a designer's creative head behind it, will produce something more than another clone of the same familiar patterns. […]More broadly, what's overhyped is the expectation that one prompt, without a designer's creative head behind it, will produce something more than another clone of the same familiar patterns. That's simply how large models work: they're trained on archetypes, while a client usually wants to see exactly what the model treats as an outlier and smooths away. That doesn't mean you can't prompt your way to something creative. You can. But it takes someone who knows what they're looking for and keeps pushing the model away from the average.”
The honest answer is that it was a different world. […]The honest answer is that it was a different world. In 2017 generative AI wasn't something every designer had on their desk, and a lot of what we did by hand could be done differently today, even with the same structure and the same people. If I were building that chapter now, I'd probably move away from fixed teams towards smaller ones that form around a problem and dissolve when it's solved, with AI taking over a share of the routine work so designers spend more of their time where judgement matters. The chapter's core, the shared standards, the career ladder and the rhythm of reviews, I'd keep exactly as it was. That part outlived me, which tells me it was right.”
Limit
It depends on what "let AI design" means. […]It depends on what "let AI design" means. Using AI to speed up the work is fine, for any group of users. Letting it decide what a vulnerable user ends up seeing is not. In my team the rule is simple: AI can help with drafts, variants and analysis, but a designer decides what goes to the user and stays responsible for it. With vulnerable users that line matters even more, because the cost of a confusing or careless screen is higher, and a model has no idea who is on the other side. The tool can be fast. The judgement has to be human.”
E-mail: [email protected] (kt-01) · LinkedIn: linkedin.com/in/eofek (kt-02)