When people ask questions like “are companies getting ROI from their AI implementations,” they’re stuck in an old conversation — it’s not the right measurement.
Companies know they can’t fall behind on automation, regardless of whether they can measure the impact. The question now is: What does it take to implement AI successfully? Is it plug and play? Does it require custom models or more fine-tuning after it’s created?
“There’s probably — I’m just going to pick a number — 3,000 to 5,000 people in the world, frankly, who can implement critical workflows that deliver value,” Scale AI CEO Jason Droege said at a recent Semafor event.
Scale, of course, employs a good handful of them, he said. Droege said the best of them have done work post-training AI models and have developed a sense for the strengths and weaknesses of specific models. Scale sends those employees directly into companies to understand the specific needs of customers, and then build systems tailored to their security and business needs.
If that sounds reminiscent of Palantir’s “forward deployed engineer,” you wouldn’t be too far off base. In fact, Scale is now competing for Defense Department contracts, Palantir’s original feeding ground.
But there’s no easy way of making AI work. It might be simple for software developers to just use AI, but not for companies with complex business processes. Watch the full interview with Droege here.