Wintercircus — Vlerick & Donna: the AI-powered organisation

Tuesday night (15/06/2026) at Wintercircus Ghent, Steven Muylle from Vlerick teamed up with Jonas Deprez and Xander Berkein from Donna to talk about how AI is actually changing the way companies operate. Enya Stessens did the intro and framed it as the third in a series: first the basics of AI, then building agents, now optimizing your operational model.

Steven opened with the macro picture, the claims and the data on AI’s impact on work. The part I found more valuable came after, sorry Steven, when Jonas and Xander showed how their company, a 40-person AI startup selling to enterprise, actually runs on this stuff day to day.

Setting the scene: claims vs data

Steven’s framing was useful as a backdrop. The narrative says AI is coming for the jobs. When Anthropic launched Cowork in February 2026, the sector lost $300 billion in market cap and people called it the SaaSpocalypse. But take ServiceNow: revenue up 28% year over year, highest operating cash flow ever at $100 million. The pain was in the stock market, not in the financials.

The research has the same distortion. An OpenAI study from 2023 claimed 47 to 56% of tasks could get an AI productivity boost. Anthropic’s actual usage data from March 2026 came in way lower. And employment data shows no systematic increase in unemployment since ChatGPT launched. The prediction now isn’t that hiring declines, just that it grows slower.

A framework of Steven’s worth keeping is weak bundles versus strong bundles. A weak bundle is a job that’s essentially a list of separable tasks — data entry, ticket-queue programming. AI picks those apart one by one. A strong bundle is where the tasks are intertwined and require governance and judgement. His example: radiologists were supposed to be obsolete by now, yet they’re in higher demand and better paid, because reading an image is one task inside a much larger bundle.

What Donna sees in enterprise

Jonas, CEO, co-founded Donna, an AI co-pilot for field sales teams and to a wider extent all outside people in service and so on. Think of the reps still taking notes on paper and filling in CRM on an admin day or after hours. Donna calls you ten minutes after a meeting to follow up and capture what happened.

Because they sell to large enterprises, they see a lot of AI adoption attempts up close. The number Jonas quoted: 95% of AI pilots fail (FORCE 2025). Not because the tech isn’t ready, but because the mindset, the process and the people aren’t. They are too often treated as a bolt-on and experiment, rather than a transformation.

To be successful you need a clear scope, starting with one team to learn before going across the whole organization. Leadership that stays engaged and actually asks whether people are using the tool, a feedback loop to continuously refine if needed and measurement of pre-defined KPIs, so there’s a real business case for expansion. The failures do the opposite: they automate a process that’s already broken, run one training session and expect adoption, get a leadership sign-off without any behavior change, and never iterate.

What I realized is that this isn’t an AI problem. It’s the same thing that kills most technology adoption aka digital transformation. AI just makes it harder, because the urgency creates pressure to move fast and skip exactly the steps that make it work.

Running a marketing department with one person

This was what I was hoping to learn.

Donna runs their entire marketing function with one marketeer. She came from a sales background where she built teams of specialists: a digital marketeer, a product marketeer, a designer. Now she is forced to do it alone with AI.

They showed a slide that captured it well, the opportunity to rethink the marketing org. The traditional model is a head of marketing hiring specialist roles to add capacity, one box per discipline. Their point: adding AI into every one of those buckets gives you limited leverage. You’re just bolting a co-pilot onto each silo.

The agentic model restacks the whole thing into three layers. At the top, leadership and orchestration: someone who sets direction, defines the data foundation, makes the judgement calls, coordinates the agents and feeds everything back into context. That’s a marketing generalist plus GTM strategy plus a bit of data and AI architecture, a human working with agents. In the middle, execution: content, campaign and research agents running in parallel, reading from context, with a GTM engineer and what they call superior craft specialists elevating the output where it matters. And underneath it all, context: the GTM brain that every insight, result and learning feeds back into. Built from the ground up that way, they put the productivity gain at around 45%.

So one person, but well supported: the GTM engineer and the craft specialists aren’t extra headcount bolted on top. They’re roles inside the new structure, each paired with agents. The shape of the org changed, not just the tooling.

AI didn’t replace the marketers, it’s that the shape of the role changed. You need someone senior enough to recognize good, who can direct tools instead of doing everything by hand. The junior specialist roles, the ones that were basically task lists, are the hard ones to justify now.

Building is cheaper, context is expensive

Xander, the CTO, talked about the engineering side.

His take: building got cheaper, but context got more expensive. You need engineers who go find context themselves, who talk to sales, product and support, who understand what they’re building rather than just executing tickets. And you need seniors who can guard against slop, because AI will happily generate code that ignores your frameworks and infrastructure. Product-minded engineers become more valuable, not less, because they iterate faster with less back-and-forth.

And then there was Franky

A Frankenstein version of Donna, their internal AI agent. The team built an internal agent called Franky that lives in Slack. It started because engineers and solution engineers were each building separate automations and skills, none shared, none with safe production access or shared memory.

Franky became the shared knowledge system with clear access rights depending on who’s asking. She does first-line triage on incoming bugs, checking customer configs and logs and suggesting what’s wrong, and she can investigate code paths to reproduce issues. For solution engineers with exotic cases, she can look across other customers’ configurations and surface similar problems.

The tell: Franky now costs about as much as their AI coding tools. That’s how heavily the team leans on her. And she doubles as a testbed, what works in Franky eventually makes it into Donna.

The pricing question

Someone asked how Donna handles the tension between rising AI costs and their model. Their answer: seat-based, not usage-based. Enterprise buyers are used to fixed license rates and annual budgets, and usage-based pricing makes that harder to approve. They’re often compared to CRM pricing, which sets the range. For now token costs are manageable, with margin subsidizing the heavier users. Xander added that coding-tool costs have jumped in the last two months as models get more capable and expensive, and that bigger companies with stricter budgets will probably crack the cost question first.

My take: seat-based works for Donna precisely because the product is personal. The value is directly linked to the efficiency of the individual user. The tool follows one person, captures their meetings, prepares their day. When value creation maps that cleanly onto a single user, seat-based pricing still holds. It’s only in that case that it does. The moment value decouples from the individual and starts flowing from volume or outcomes across an organization, the seat model starts leaking. What I’m curious about is whether they’ll add an MCP-style service that other systems can call, and how they’d monetize that. The personal-efficiency logic that justifies the current model doesn’t transfer to machine-to-machine usage, so that would force a different answer.

What I took away

The gap between AI narrative and AI reality is still large. The market panics, the financials hold. The research predicts mass displacement, the employment data shows nothing yet.

But the more interesting insight wasn’t in the macro slides, it was in how Donna actually operates. They’re not replacing people, but changing what people do. One marketeer runs a department. Engineers spend less time writing code and more time gathering context. Support gets first-line triage from an agent before a human steps in.

In every case the pattern is the same: the executor role shrinks, the orchestrator role grows.

---

Wondering whether your pricing model still fits how AI changed your cost structure? Let’s talk.