Case study
Measuring, then moving, a 200-developer workforce
Cefalo is a Norwegian software consultancy whose 200+ developers work embedded in client teams. Their concern was strategic: AI is changing what clients expect from a development partner, and they wanted to lead that change rather than react to it.
The baseline: 172 interviews, one consistent scorer
The baseline ran in September 2025, when 172 of the 180 developers invited sat a voice interview with Discovery Dialogues and each conversation was scored against the same five-level capability model across six skill areas. The picture was typical of the industry at the time: overall capability at level 2.2, with 69 per cent of developers at level 2 and one developer working at level 4.
Eight months of deliberate change
Cefalo drove the change and deserves the credit for that. Our role was AI leadership on demand: a standing seat at their weekly AI task force, guidance sessions, and advice on adoption, tool choices and their transformation plan, including identifying the champions who would carry it. We also supported their company-wide hackathon, where an automated pipeline we built judged 35 teams' repositories and demo videos. The baseline set the starting point, a change workshop equipped their engineering managers and the reassessment measured the movement.
The reassessment: same questions, same scoring
In May 2026 Discovery Dialogues ran again: 191 interviews across seven days, most of them completed on the busiest single day, with a median conversation of 20 minutes. Of those interviewed, 138 developers could be paired with their own September baseline for a like-for-like comparison.
September 2025 to May 2026, paired cohort of 138
+0.9
capability levels gained, cohort average
80%
of returning developers moved up
~30%
now at level 4+, from under 1%
The largest gain came in agentic development, up 1.5 levels. Tool adoption shifted with it: agentic coding assistants went from near zero to widespread, and every major coding tool at least doubled its reach across the cohort.
Some of that movement is industry tailwind; 2025 to 2026 was a fast year everywhere. What the programme adds is measurement you can trust, with the same model and scoring grid applied to paired individuals, and guidance that makes the movement deliberate.
"Paul's work has been exceptional – not just insightful, but actionable. His conversations with 172 of our developers revealed that AI isn't just hype, it's already boosting quality and productivity."
Petter Gustafson, CEO, Cefalo, on LinkedIn
The measurement method
Change programmes usually measure themselves with self-assessment surveys, which drift with mood and incentive. A 20-minute structured conversation with every developer, scored consistently by the same system both times, gives leadership an evidence base: who moved, where the stalled group is and what to do next.
Published with Cefalo's agreement. Figures come from the paired assessment data; developer productivity claims are deliberately excluded because they were self-reported.