Building with AI in five sprints
First-stage goals reached in five of six sprints
DEEP Measures helps pharmaceutical and biotech companies develop, validate and gain regulatory acceptance for digital measures in clinical research. Its specialists work with demanding evidence requirements and sensitive information. Some colleagues were already building prototypes and agents, and leadership took part in the work.
The opportunity was to help more people develop tools around their own expertise and learn how to judge whether the results were good enough to use. In DEEP's setting, a useful prototype still has to fit the company's responsibilities around data and quality.
How the sprints ran
- 01
Made people's goals and workflows visual and concrete, giving the coaching a personalized starting point in each person's work.
- 02
Helped department heads and specialists build around their own expertise through hands-on sessions, with individual study plans and practical follow-up material.
- 03
Used AI in preparation and follow-up to help turn session context into tailored learning resources, including visual canvases and recaps.
- 04
Built source-checking and review into the work, including specialist writing instructions and reviewer agents.
- 05
Examined what a working prototype would need before colleagues could depend on it, including examples to test against and clear review responsibilities.
DEEP Measures
“I'd say we were extremely satisfied, [it shows in the] fact that we didn't need the sixth sprint.”
The moment that surfaced
For one regulatory specialist, the work meant turning her own writing approach into reusable instructions. We worked from her documents, created a writing-rules file and set up a reviewer to check drafts against the style and evidence expectations she had defined. The first reviewer output gave us something specific to assess and improve together.
What the sprints changed
DEEP ordered six sprints. In their closing feedback, its leaders described the first-stage goals as achieved after five. Kai Langel and Rauha Tulkki-Wilke described colleagues getting started successfully and then developing different needs as their capability grew. The next phase would require more work on shared processes and coordination.
- 72% agreed or strongly agreed that coaching helped them use AI more effectively (staff-reported, DEEP's internal follow-up survey, 8 October 2026).
- 62% reported adopting at least one new technique, workflow or tool as a result of the sessions.
- Colleagues were already sharing demonstrations in biweekly meetings by the time we wrapped up.
What DEEP keeps
DEEP brought a useful combination to the collaboration: people willing to build and specialists able to explain where an answer fell short. Their questions took us into source quality, review responsibilities and how a workflow should behave for another user. DEEP owns the work and the results. My role was to help people work through obstacles and develop approaches they could continue using independently.