GSD Venture Studios Podcasts by Gary Fowler

There are a handful of genuinely useful ideas - the three-layer architecture (system of record, system of engagement, agentic layer), the warning against building AI on unreconciled data, and the long-tail spend insight - but they are buried under extensive host filler, a meandering Tesla/Fast and Furious tangent, and padded transitions that waste several minutes of a 29-minute episode.

“if you build AI on top of bad data, uh, if the operating system underneath is not clean and structured, the output of the AI agent or whatever, AI capability will be very poor”

“what we used to have procurement team being able to focus only on the 10% biggest contract that a company had... With the automation that we bring and all the work that is done by AI, we now have the same procurement team that can look at even the long tail”

The core thesis - build the data layer before the AI layer - is a defensible but increasingly common enterprise software argument, not a contrarian one. The long-tail spend point has some freshness, but most observations recycle well-worn enterprise AI caution without adding a genuinely novel frame or counter-intuitive claim.

“when everyone was rushing to our direction, we had to uh, first rebuild everything. And that was a tough choice to make”

“owning this data, ah, and honing the system of record means that we will be uh, the one underlying tool so that AI works well”

Marc-Antoine Lacroix is a legitimate operator who served as both CTO and CPO at Qonto, a real European fintech unicorn, and witnessed its scale from 40 to 1,500 people and €50M to €5B+ valuation before founding Pivot - he has genuine practitioner credibility. He does not have the depth of a category-defining executive and is still in early company-building mode, but he is clearly not a career podcast guest.

“when I joined Konto, um, uh, who is now the uh, multi, uh, uh, billion valuated fintech in Europe. It was a very small company, like 40 people. It was early 2018 and I saw the scale of Konto from 40 to 1500 people and from 50 million to 5, uh, more than 5 billion valuation”

“at some point in this company I had to implement an incubant procurement tool. And uh, that was quite uh, traumatic experience”

The episode offers a handful of concrete data points from the guest's Qonto years (headcount, valuation, timeline) but Pivot's own traction is entirely absent - no customer names, no ARR, no adoption metrics, no specific ROI figures. Claims like '10x your productivity' and 'tremendous amount of value' go completely unsupported.

“I saw the scale of Konto from 40 to 1500 people and from 50 million to 5, uh, more than 5 billion valuation”

“we managed to put everything under control. And it did bring value. The level of work that we had to put into that was crazy”

The host asks a reasonable opening question and lands one productive 'Why?' follow-up, but then derails the most substantive part of the episode - the AI trust and hallucination discussion - with a multi-minute Tesla/Fast and Furious anecdote that allows the guest no space to answer. Closing questions ('one word for the future') are generic podcast filler and the host's own lengthy self-promotion further squeezes the guest's airtime.

“And I'm like, dude, you don't want to do that. You know, I don't know what connections to operating system of your car, but you might really upset it and something happens”

“if you could solve one business problem with AI tomorrow, what would it be? And, um, what's a word that describes, uh. And let's talk about the future of procurement”