The brief
On 23 April I was a guest lecturer at the University of St. Gallen together with Oliver Gerstheimer of the Kassel-based design agency chilli mind. The setting: CAS Management for IT Executives (CAS MITF 2026). Four hours, around 17 Swiss IT heads and CIOs, and a single shared promise - no slide deck marathon, just practice.
The line on our title slide was deliberately provocative:
"What an agency used to need weeks for, one person now does in a single afternoon."
The question was not whether that statement is true. The question was whether the participants wanted to see it for themselves.
Three modes of AI work
Before anyone touches a keyboard, you have to be clear about how deeply the AI is going to be involved in your work. We divided that into three modes, because most executives only know the first one:
| Mode | Analogy | Typical use |
|---|---|---|
| Claude Chat | An intern sitting next to you | Drafting text, challenging ideas |
| Claude Cowork | A colleague with access to your project folder | Building documents, research, project work |
| Claude Code | A senior engineer in the engine room | Automations, skills, full pipelines |
Most people in the room had experience with ChatGPT. Nobody had ever worked with an AI colleague that has access to a project folder, understands files, asks its own follow-up questions, and ultimately produces an artefact that you can hand off to your next compliance review.
The case study: Novela
HSG had given us a fictional case. Two Swiss medical-technology giants, Doche and Blaukreuz, are merging into the Novela AG: 200 billion Swiss francs in revenue, 200,000 employees, headquarters in Switzerland, R&D in Europe, the US and Japan. The headline product innovation: a 3D bioprinter for kidney tissue, because ten percent of the world's population suffers from chronic kidney disease and millions die every year without affordable treatment.
The HSG case study was half a page of A4. That was all. Exactly the right amount to demonstrate what happens when you hand an idea to a good AI colleague.
What we built in four hours
We walked the participants through what happens when you treat Claude Code as a senior engineer and give it a clean brief. The sequence:
- Intake interview with Claude Code in the terminal - Claude asks back: target audience, ticket size, regulatory environment, system landscape.
- PRD generation in one go - executive summary, personas, functional and non-functional requirements, technical architecture, user stories, success metrics, risks, timeline.
- Mockups & visualisation in parallel - the first wizard screens of the configurator as visual anchors.
The result was an 8-page PRD for a kidney-printer configurator - a web-based self-service tool that lets pharma R&D, university clinics and CROs build their configuration in under ten minutes, receive a binding PDF quote, and land automatically in the CRM as a qualified lead.
Hardware step: print head technology, build volume, sensors. Live price on the right, recommendations driven by the chosen use case.
Bio-inks step: cell-type compatibility, hydrogel bases, starter kits. The recommendation engine flags compatible cell lines based on the selected use case.
Summary step: the buying committee gets a single artefact with all four configuration dimensions, the binding price, and a 30-day quote validity.
What the document contained surprised many in the room, because it included exactly the building blocks an experienced product owner would deliver:
- Three personas with concrete use cases (Principal Scientist, Lifesciences Procurement, Quality Manager).
- Four configuration dimensions: hardware modules, bio-inks, software/data management, service.
- Architecture decisions with rationale per layer (Next.js App Router, React-Three-Fiber for the 3D visualisation, Neon Postgres for the audit log, Vercel Functions in the EU region).
- Compliance layer covering the topics that bio-pharma buyers actually care about: EU MDR 2017/745, FDA 510(k), ISO 13485, GAMP 5, 21 CFR Part 11, export control.
- Risk matrix with mitigations - including the honest statement that "MVP in 6-8 weeks + all 4 configuration dimensions + CRM/ERP/CPQ integration + 3D visualisation + 2 languages + MDR/FDA/ISO compliance" is not realistic in this combination, with a proposal for a two-stage MVP.
- Eight-week timeline from discovery and alignment to a soft launch with beta customers.
This is not a toy example. This is the kind of document that goes through as a requirements brief in mid-market bio-pharma right now.
The full PRD as it came out of Claude Code. Download the 8-page PDF.
What the executives took away
Three observations I jotted down on the train back to Munich after the seminar:
The bottleneck is not technology. It is briefing. Claude delivers exactly the depth you put in. Type "make me a PRD" and you get mediocrity. Take the intake interview seriously and answer the follow-up questions cleanly, and you get a document that survives the auditors' review.
Buying committees love shared artefacts. In this case, scientists, procurement and quality assurance share the buying decision. The configurator was designed to serve exactly that three-person dynamic. The fact that Claude can pull this multi-stakeholder structure out of half a page of case study and translate it into personas plus user stories was the aha moment for many in the room.
The question is no longer "if", but "how fast". We had a slide with three data points: cost per AI output is falling roughly tenfold per year, Anthropic Enterprise is growing faster than OpenAI in B2B, and Nvidia briefings as well as seven-figure consulting boutiques use Claude Code daily. After that slide, the discussion stopped being "Does this work?" and turned into "When do we start, and who will own it internally?"
What this means for CIOs
I see the same pattern in every workshop. IT leaders sense that something fundamental is shifting, but they cannot find the language for it inside their organisation. Employees experiment with ChatGPT, communications produces posts, engineering writes scripts. But nobody knows how this scales.
The step from "chat tool" to "AI colleague with project access" is organisational, not technical. Making it requires:
- A CLAUDE.md for every project - a small constitution that gives the AI colleague context.
- Skills - reusable instructions for recurring tasks (invoice review, contract analysis, market research).
- A clean separation between "the model can read this" and "the model can act on this" - GDPR-compliant and auditable.
- Human in the loop as a binding rule, not lip service.
This is exactly the programme I run with teams over two weeks. Not a workshop that shows what is possible, but an introduction after which the team continues independently.
Materials and outcome
If you want to set up something comparable for your own organisation - from intake interview through PRD to a working prototype - here are the next steps:
- Claude Code Enablement Sprint - two weeks during which a team really integrates Claude Code
- AI workshop for executives - one or two days hands-on, like at HSG, with your own case
- Talk directly - a 30-minute scoping call to find out whether this fits
Thanks to Oliver Gerstheimer and the chilli mind team for the co-facilitation, and to Bettina at HSG for framing the methodology. It was one of those afternoons where you could feel the self-image of a profession quietly shifting in the room.