Phase 1 · Weeks 1–2
Foundations
All roles together. Shared language, mental-model shift, the vibe-coding-to-agentic-engineering spectrum, and the first hands-on agentic workflow. Verification-first thinking introduced before anything else.
Digital Product Team Transformation
Your team builds software the old way. Here's the new operating model — installed in 12 weeks.
A 12-week program that migrates your product team from traditional development to governed agentic engineering. Not a course. Not a workshop. A full operating model migration — spec, verify, generate, review, ship — rebuilt and installed so it sticks after I leave.
Your engineers use AI. Your delivery system is from 2015.
Your engineers accept AI suggestions without verification. Code reviews can't keep up with the volume. Quality drifts in directions nobody is watching.
Your PM still writes PRDs. The engineer prompts an agent with them. Nobody checks whether the output actually solves the right problem.
You've invested in tools. Usage is up. Throughput isn't. Quality isn't. Something is missing between the licences and the results.
The tools aren't the problem. The operating model is.
"What sets Raphael apart is his pragmatism. Where other consultants stay theoretical, he immediately dives into technical feasibility and implementation. You don't stay stuck in ideas — you move forward concretely."
The shift
Code generation is cheap now. The bottleneck moved to everything around it — specifying intent, verifying correctness, holding architecture coherent, governing autonomous agents.
Teams that still organise around writing code are optimising for the wrong thing. Teams half their size who rebuilt the operating model around directing agents ship faster, with higher quality, and recover from failure sooner.
Three phases. Twelve weeks.
Foundations together, then role-specific tracks, then a real pilot under governance. Same arc in every format — only the cadence changes.
-
-
Phase 2 · Weeks 3–8
Role-specific tracks
Engineers, PMs, Designers, QA, and DevOps each take a dedicated track. Two cross-role integration sessions in weeks 4 and 6 keep the team practicing as a system, not five parallel disciplines.
-
Phase 3 · Weeks 7–12
Governed delivery
Real pilot project — your codebase, your tickets, your product. Governance stress-tested in practice, EU AI Act compliance applied to actual artefacts, evidence report and scaling plan delivered at week 12.
"Raphael's approach was spot on: pragmatic, focused, and perfectly balanced between coaching and hands-on workshops. In just a few sessions, we achieved real improvements that already show in our daily efficiency. It's rare to see such impact from such a compact collaboration, I strongly recommend the format!"
What changes for each person on the team
Five roles, five distinct shifts. Engineers don't get the same training as designers; PMs don't get the same as QA. Track depth is calibrated to the discipline.
Software Engineer
From code writer to engineering director.
- Write specs that agents can execute against.
- Run TDD cycles with AI agents — red, green, refactor.
- Review AI-generated code for architecture coherence, not just correctness.
Product Manager
From definer to orchestrator.
- Write outcome-based specs that machines can read.
- Run AI-assisted discovery and parallel experiment design.
- Make autonomy-routing decisions: what to delegate, what to keep human.
Designer
From maker to director.
- Explore 10+ prototype variants in the time you used to make 2.
- Evaluate design at volume against intent rubrics.
- Specify design intent so agents can implement and iterate.
QA Engineer
From tester to evaluation engineer.
- Design eval suites and grader calibration.
- Build harnesses that run continuously, not at release.
- Review traces and transcripts to debug agent behaviour.
DevOps
From infrastructure to platform owner.
- Design the agent environment — capabilities, permissions, sandboxes.
- Stand up trace and observability pipelines.
- Adapt CI/CD for agentic workflows that branch and self-verify.
"Raphael came to the office to help us further leverage AI into our daily workflows, and the impact was immediate. He guided us from a high-level brainstorming on AI capabilities down to very specific, individual sessions where we tackled real operational bottlenecks. He gave me a specific framework that has since become my rule of thumb for any AI-assisted task."
Three delivery formats
Match the cadence to the team's operational reality. The arc is identical — what changes is how often we meet and how dense each session runs.
Intensive
12 weeks continuous
Dedicated transformation teams. Full focus, fastest time-to-impact.
Compressed
6 weeks
Teams that can't pause operations. Higher session density, same outcomes.
Modular
16 weeks half-pace
Teams under heavy operational load. Practice cadence stretched, repetition preserved.
The methodology
Built on primary research, not vibes.
The methodology is synthesised from a cross-analysis of independent research reports drawing on primary sources from Anthropic, OpenAI, NIST, OWASP, METR, DORA, and practitioners including Simon Willison and Andrej Karpathy. Every method taught has evidence from multiple independent sources. The underlying theory is the Bicycles Framework (Agrawal, Gans & Goldfarb, 2025): AI substitutes implementation but complements judgment. Every exercise is designed to build judgment, not tool fluency.
Intent Design
From requirements writing to outcome specification.
Verification-First Development
From review-after to prove-before.
Agentic Workflow
From linear coding to orchestrated agent loops.
Discovery & Experimentation
From sequential research to parallel exploration.
Governance & Risk
From afterthought security to first-class agent governance.
Measurement
From velocity metrics to quality-adjusted outcomes.
Learning & Adaptation
From informal growth to structured practice evolution.
"Raphael is a very passionate, skilled and hard-working UX strategist and architect, with a keen attention to detail for design systems, UX strategy and design operations. I absolutely loved to see Raphael bridging the gap between design concepts and technical implementation."
Investment
Fixed format, fixed price, no surprise add-ons. Three tiers scoped to your team size and the level of post-program support you want in place.
Core
Single team up to 8 people. Phase 1 + Phase 2 + Phase 3, facilitator-led end to end.
Extended
Core, plus additional coaching hours and async support through the post-program embedding window.
Multi-team
2–3 teams in parallel. Shared Phase 1 across all teams, parallel role-specific tracks, coordinated Phase 3 with cross-team integration sessions.
Tier anchors shared during the discovery call so we can scope against your team size and operational constraints.
"A lot to say about Raphaël, but something that is not already mentioned is that he always finds a way to deliver the expected value, whatever the format or circumstances, mobilising his vast general culture and AI knowledge for direct, actionable insights. Strongly recommended!"
From 28 years of installing the new way
"The launch of the MyIRISnet customer selfcare tool was of a quality and efficiency rarely encountered for such complex projects. This success demonstrated that working with a small team of experts actually provided a high guarantee of results and budget control."
"Raphaël is probably the most digital person I have met in the course of my career. In business, we often have a tendency to create complexity and his greatest strength is that he simplifies things so that the customer receives the best possible experience."
Talk to me before you scope this.
Thirty minutes, no slides. We pressure-test whether your team is the right shape for this program, which format fits your operational load, and what evidence you'd need to know it worked.
Questions I get every time
We already did AI training. How is this different?
Training teaches tools. This installs an operating model. After training, people know how to prompt. After this, your team specs, verifies, reviews, and ships differently — at the team level, not just per individual.
Twelve weeks is too long. Can we go faster?
There's a 6-week compressed format. But practice needs repetition, and teams that rush skip the verification habits — which is where most of the value compounds. Compress the calendar, not the practice density.
Our engineers will resist being told how to work.
The program is built by engineers, for engineers. The SOPs are optional but tested. The content respects technical intelligence — it's rigorous, not patronising. Engineers tend to push back when the work is shallow; not when it's serious.
Can't we just hire AI-native engineers instead?
You could. But your PMs, designers, QA, and DevOps still need to change. New hires also need to integrate with your existing team — and that integration is where most AI-native talent gets ground down. This transforms the system, not one role.
What if we're already using AI?
Using AI isn't the same as having an agentic engineering practice. The question isn't whether you use AI; it's whether you have verification-first development, outcome-based specs, governed agent workflows, and quality-adjusted metrics. If not, you have tools without a system.
How is this different from the AI Activation Sprint?
The Sprint is horizontal adoption — any team, any role, AI in daily work. This is vertical depth — specifically for product teams, specifically rebuilding the delivery system. Sprint: 'we use AI.' This: 'we build software differently because of AI.'