From AI hype to resilience and environmental practice. A conversation with Gaël Gioux.
- Terraquota
- Jun 25
- 6 min read
Co-founder of NoeSysAI, Gaël Gioux helps companies adopt AI in a practical way, from strategy to governance. He's also an old friend of Henri and of TQ. With his help, we brought AI into our LAB platform, where it now structures our teardown data and speeds up our reports. We wanted his view on AI, and on a parallel we keep noticing: AI today looks a lot like sustainability a few years ago.
When companies come to you, what are they usually worried about, and what question should they ask instead?
Most companies walk in with a defensive question. "Everyone is talking about AI, are we late?" Or "How much could we save?" Behind it is fear: of competitors, of disruption, of missing something everyone else seems to understand. And fear produces a predictable answer, the efficiency reflex. Cut costs, speed things up, do the same with less.
The better question is the opportunity one: what could AI do for us? Then you map where value actually sits in your own processes. Efficiency is part of it, but so is growth, through better offers, better client experience, wider reach. And resilience, like capturing knowledge before it walks out the door. Research shows the growth side dwarfs what cost-cutting can ever deliver, yet most companies invest almost only in efficiency. HBR recently called it the growth blindspot. In my experience, the real opportunity usually sits where the company wasn't looking.
AI changes so quickly that what felt advanced three months ago can already feel outdated. How do you advise companies to deal with that?
You don't keep up with the tools. You build the capacity to absorb change. Tools and models change every month. Your processes, your data and your people's judgment change slowly. Invest there. Knowing how to break down a task, brief an AI and check its output works across every tool generation. Those skills compound. Tools don't.
In practice, that means running small experiments on real work rather than betting big. With one condition: the experiments need a North Star. A long-term vision of what you want AI to make of your company turns each experiment into a brick. Without it, you end up with a pile of disconnected pilots that prove everything and change nothing.
And to be honest, we are surprised every quarter too. That's the strongest argument for method over tool-chasing.
A few years ago, sustainability was everywhere, but not all of it turned into real change. We sometimes wonder if AI is in a similar moment now. Do you see it that way?
The pattern is the same. In sustainability, the companies that lasted put it into their operations and their P&L. The others wrote nice reports. AI has its own version of this: the "Copilot license for everyone" announcement is the new greenwashing.
The "Copilot license for everyone" announcement is the new greenwashing.
The companies that make it last share a few habits. Every AI initiative is tied to a measurable result in a real process. Leadership uses the technology itself instead of delegating it to a task force. Employees are trusted to find use cases rather than receiving them from above. Governance comes early and stays light, enough to make experimenting safe without killing it. And they treat AI as a transformation with a learning curve, not a project with an end date.
Sustainability gave us the test: is it in the process, or is it in the brochure? AI passes or fails the same one.
A lot of AI today depends on a small number of providers. From a resilience perspective, how should companies think about that?
The concern is fair, and you can treat it the way you treat any supply chain: assess the dependency, diversify where it's critical, design for substitution.
Two things make it more manageable than it looks. First, the model layer is commoditizing. The very best models sit with a few providers, but most use cases don't need the very best. There is enormous value between what companies do today and what average models can already deliver, and that average is served by a widening field: open-weight models, European players, smaller specialized models. Second, you can build for portability. If your use cases are designed one layer above the model, switching providers becomes a configuration change, not a rebuild.
And the most resilient asset is skill. A team that knows how to work with AI is portable across any provider. Full independence is not realistic today. Managed dependency is, the same as for cloud or energy.
The very best models sit with a few providers, but most use cases don't need the very best
AI is energy-intensive. Are companies taking that seriously yet?
Honestly, it rarely comes up yet. Clients ask about cost and confidentiality first. But for companies, the energy question will arrive exactly there, in the cost: the energy intensity of AI will end up in the price of using it.
So the answer is the same discipline we just discussed. Right-size the model to the task instead of using a frontier model for everything. Build the skills and the portability that let you switch as prices move. Sustainability taught the method: measure, then optimize. Most companies today measure neither their AI usage nor what it costs them. The ones that build that awareness early will adapt fastest when energy shows up in the bill.
Where do you see AI and environmental work actually starting to meet?
We can imagine a few directions: handling the huge volumes of data behind environmental assessments, widening EPD coverage, chatbots that help people read and use environmental data, or simply taking the manual, repetitive work off people's plates. Which take do you have, that we might miss?
All four directions are real, and your LAB is a live example: AI structuring teardown data and speeding up reports, with value showing up in weeks.
The take I'd add: the biggest impact is not doing environmental analysis faster, it's collapsing its cost so it can sit inside everyday decisions. Today, environmental assessment is an expert bottleneck, done once a year or once per product. When AI makes it cheap and instant, eco-design criteria can appear at the moment an engineer picks a material or a buyer selects a supplier. Environmental data moves from reporting to decision-making. That's a different order of impact.
There's also the unglamorous part: regulatory reporting is repetitive, structured and high-volume, exactly what AI handles well. Automating it frees scarce environmental experts for the judgment work.
For a small company, or even for someone who has barely touched AI yet, where is the smartest place to start?
First, flip the premise: small is an advantage. Short decision lines, visible processes, no committees. A small company can capture AI value faster than a corporate.
Start from a pain, one process that hurts, like quoting, reporting or support emails. Use off-the-shelf tools before building anything. Put your money into people's time to learn rather than into software, it's the highest-return line in the budget. Time-box a pilot to a few weeks with one measurable outcome. And set one rule from day one: be clear about which data never leaves the house.
For someone who hasn't really touched AI yet, use it on your own work, today. Take a real task from yesterday and try doing it with an AI assistant. The one skill to learn is checking the output. Treat AI like a brilliant intern: fast, tireless, and sometimes confidently wrong. The craft is in the briefing and the checking. Start on tasks where mistakes are cheap.
Treat AI like a brilliant intern: fast, tireless, and sometimes confidently wrong.
What's the one question about AI you wish someone would finally ask you, and nobody does?
Not "what can AI do?" but "what do we want it to do for us?" Capability is no longer the constraint. Intent is. The companies that can answer the second question are the ones that move.
If one thing should stay with your readers: don't wait for AI to stabilize, and don't reduce it to a cost-cutting tool. Set a vision, build the capability inside your team, start small and start safe.
Interview conducted by Terraquota
