This is an archive episode, first broadcast in 2024 under the show's previous name. It carries a short synthetic announcer introduction; the conversation underneath it is untouched human audio.
Everybody is hunting for AI use cases. What if "use case" is the wrong question to be asking?
Jack Nugent works at an unusual junction. His consultancy, Tradavex, advises ship owners, brokers and inspection companies who understand commodities but not machine learning, and technology companies who understand machine learning but have never chartered a ship. He spends his time translating between the two.
Recorded in July 2024, when the show carried a different name.
What this episode covers
- Prediction is not judgment. Nugent separates what these systems actually do, which is predict, from what people assume they do, which is decide. Most of the disappointment in corporate AI lives in that gap, and so does most of the risk. He is also clear that the internet comparison holds, which cuts against calling any of it a fad.
- The data that does not exist yet. His most useful point: the valuable dataset for your business is often one nobody is collecting, because until recently there was no reason to. Water pumps and satellite imagery are his examples.
- Why low-hanging fruit is self-defeating. Picking the easiest wins first feels prudent, gets signed off quickly, and quietly guarantees the programme never reaches anything that matters. He contrasts top-down and bottom-up adoption and makes a case for running both rather than choosing.
- The doorman fallacy. Automate the task, lose the thing the person was actually doing.
- Two hundred employees. If AI makes each of them more productive, do you employ a hundred, or do you arm two hundred and take more of the market? He is clear which he thinks wins.
- Tacit knowledge as an asset. Transcripts, context windows, and the practical business of turning what your workforce already knows into something proprietary that a competitor cannot rent. This is where his argument gets genuinely actionable: the model is a commodity, your data is not.
- Business strategy before AI strategy. The ordering he insists on, and the reason a company without a clear answer to "how do we make money" cannot possibly answer "where should we use AI".
- Curious engineers, humble experts. The two dispositions he thinks a team needs, and why the combination is rarer than either.
- Why models trained on the past cannot adapt. The limitation underneath a lot of failed deployments, and the reason he keeps returning to judgment as the human contribution.
- If it is easy, it is not worth much. The line that ties the whole argument together.
About the guest
Jack Nugent runs Tradavex, a consultancy bridging technology and the commodity markets. Its clients fall into two groups. On one side, traditional operators - ship owners, ship brokers, inspection companies - who have the domain expertise but neither the knowledge nor the headcount to assess machine learning properly. On the other, technology companies with strong products and no in-house trading or chartering experience, who need to understand how cargoes are actually bought, sold and chartered before they can price or position anything.
Resources mentioned
- Tradavex - Nugent's consultancy
- The doorman fallacy - Rory Sutherland's framing, applied here to automation
- Moneyball - his reference point for risk appetite and where human judgment still earns its place
Quotes
"AI is a tool and it's one of many tools, and it might not be the right tool for your business. But it's the most powerful tool we have today."
Jack Nugent
"Use cases is a lazy term."
Jack Nugent
Books - the Human-First AI Series
- Hello! I'm New Here
- An Absolute Beginner's Guide to AI
- Use AI, Stay Human
- The AI Pilot Handbook
- When AI Becomes Your Responsibility
Connect with Alastair on LinkedIn.