The same AI models are accessible to basically everyone. So why will some organisations win while others won't? (podcast recap)

In my conversation with Sabine Gromer earlier this year, we picked up on the question of what the currently level playing field* for raw AI capability means for competition: in a world where everyone can get access to the most recent GPT or Claude, where does the competitive advantage actually come from?

It comes down to two things: do you use it smarter than the other organisation; and can you access and validate a new model faster than them? Both of those are organisational capabilities, not just technical ones.

To build this capability, your test cases - the scenarios that reflect how your organisation actually operates, what decisions you make, what good output looks like - are becoming important intellectual property. They are what let you evaluate any new tool quickly and accurately against your own reality.

Years back, I worked on a legal tech project where it took months to identify the right test scenarios for a specific type of brand protection work. Once we had them, evaluating new tools went from "weeks of uncertainty" to "days of clarity." That library became more valuable than any single tool we evaluated with it.

Organisations that block experimentation - no sandbox, all tools locked down to Copilot - aren't just moving slowly. They're also not building the evaluation muscle. When the next model drops and it matters, or any other new relevant technology arrives, they'll be starting from zero. The organisations that have been iterating will already know how to plug it in and test it overnight.

One important consideration in this process will be, where exactly you build up this intellectual property and who you give access to it. It is what makes your organisation unique and reflects knowledge that no artificial intelligence system will publicly find for training. In a market where commodification is a threat, it will make sense to think twice and select the relevant technology platforms based on effective data ownership considerations, and not just on promises made in a compliance statement.

Listen to the episode:
Spotify: https://lnkd.in/dEermDcB
Apple Podcasts: https://lnkd.in/ddX5xD8M
Buzzsprout: https://lnkd.in/d6_CTeu2

* please don't rely on this staying the same. There is a potential for large partners (like Infosys in India for Anthropic) to become gatekeepers; there has been a first instance of geopolitical doctrine limiting export of a cutting edge models outside the US; and with all platform companies, pricing models are highly likely to be diversified once a certain degree of lock-in has been established.