Department agentified.
The bottleneck wasn't speed. It was decisions. I put our standards into the tools, so designs stopped looping and started shipping.
- Role
- Head of Product Design & Research
- Reported to
- CPO
- Team
- 9 designers across 3 product tracks
- Mix
- Two-thirds external, no shared way of working
- Mandate
- Agentify the design workflow, a company goal for 2026
- Scope
- Design org behind DMP, Randstad's talent app (web + app)
- 6 wk → 1 dayOnboarding flow, from open-ended iteration to dev-ready, well received in the demo
- 1 → 4Directions explored per problem, instead of betting on one design
- WeeklyEvery designer runs the AI design review in their files
Nine designers, no finish line.
I took over the design org in May 2026, reporting to the CPO, with an explicit company goal: agentify the design workflow.
Nine designers across three product tracks, two-thirds external. On the surface, work was moving. Underneath, it looped. Delivery ran waterfall while design was expected to cover the full double diamond, so discovery had no place. Requirements kept moving, nothing had a finish line, and some features took a full quarter to reach launch.
Seniority titles came without senior expectations, and design was positioned as a feature-request factory: no shared process, no QA loop, no clear owner per feature.
Put the standards into the tools.
Instead of more crits and more process, I put the standard into the tool.
First came design principles and a design vision, written in weeks instead of the usual one to two quarters. They guided the uplift of legacy flows onto DMP patterns.
Then an AI design review inside Figma, loaded with our principles, best-in-class examples and the consumer-experience bar. Designers get actionable feedback in their own file, without waiting for a crit. I also built UX review skills and agents across Figma, Claude and GitHub, so designers can explore in code and Artifacts as well as in frames.
- Hypothesis
If the standard lives in the tool, designers can make and defend decisions themselves, and work stops looping.
- How I'd know it failed
Usage drops off after launch, or reviewed designs still bounce back from product and development.
What I chose, over what.
| I chose | Over | Because |
|---|---|---|
| Standards built into the tools | More crit rounds and more process | Crits don't scale, and the team was waiting on feedback instead of deciding |
| Four directions per problem | One polished bet | AI makes extra directions cheap. Building the wrong one is expensive |
| Raising the bar for the team | Training everyone up to it | The calibration showed a gap too big for a quick upskill |
| Shipping the workflow in progress | Waiting for perfect adoption | Adoption follows proof, so the workflow had to ship first |
What I did with the team.
- Adoption
I had to push adoption. The review skill only stuck after three or four nudges. Now it's part of every designer's week.
Some designers ran with it straight away, a middle group followed once they saw proof, and a few need a different path.
- Feedback loop
There was no feedback culture, so I asked product for it. Two PMs gave a candid, structured read, and it matched the diagnosis: no shared guidelines, no time for discovery, diluted ownership.
- Career framework
I wrote a career framework for the team, with input from product and engineering, and calibrated the team against it. The team averages 2.9 against an expected 3.5 for their level. The biggest gaps are AI and UI.
- Process
Now co-designing a new delivery model with our development partners and stakeholders, so discovery gets a real place before delivery.
What changed so far.
The onboarding flow: about six weeks of open-ended iteration in the old way of working, against one day from brief to dev-ready with the new tooling. Product partners called out the result after the demo.
Exploration: designers moved from betting on one design to four directions per problem, each pushed further than the last.
Product partners see clearer handovers: review rounds and questions per handover roughly halved, and typical features now launch in weeks instead of a quarter. Early signals, tracked from here.
Investment: more budget for AI tooling next year, AI-native designers joining the team, and a shared AI workflow harness for design and our external development partner. I'm also making the case for a dedicated design system product team, because agents build with whatever system they're given.
PM, RANDSTAD DMP
Still in progress.
This is a live case. The process redesign, the design system team and the new hires are all in motion.
What I'd do differently so far: start the feedback loop with product on day one. I only asked after four months, and the PMs had the same diagnosis I did.