GM just laid off roughly 600 salaried IT workers, more than 10 percent of its IT department, and is replacing them with AI-native engineers. People who build AI systems, not just people who use them.

The coverage reads like modernization. I want to push back on that framing.

Not because the move is wrong. Because the story being told about it is incomplete – and if you’re drawing lessons from it, the incomplete version can get you in trouble.

Gartner looked at organizations deploying autonomous technology and found that about 80 percent reported workforce reductions. That’s the number that gets cited. Here’s the one that doesn’t: those cuts created budget room but did not consistently correlate with ROI. The organizations that actually outperformed were the ones pairing deployment with reskilling, workflow redesign, governance, and human oversight.

The swap path looks like transformation. The data says it’s only half a strategy.

I’ve sat in enough budget reviews to envision the narrative where the calculus sounds clean: legacy IT roles out, AI-native engineers in. In some contexts, it is clean. Perhaps in automotive, institutional knowledge is largely transferable. Processes are documented. Constraints are visible.

In many verticals like insurance, healthcare, and financial services, it’s not that simple. Institutional knowledge in those environments isn’t overhead – it’s compliance memory. The engineer who knows why your claims adjudication logic is written the way it is, who remembers the workaround a past audit required, who understands which edge cases surface in which state jurisdictions – that knowledge isn’t in your architecture diagrams. It isn’t in any model’s training data. It walks out the door with the engineer.

You can hire a brilliant AI-native engineer. They won’t know what your 20-year SME knows on day one, day 90, or probably day 365. And in a regulated environment, the gap between technically correct and regulatorily defensible is often filled by exactly that kind of institutional depth.

The organizations seeing the best AI outcomes aren’t the ones that moved fastest on talent replacement. They’re the ones building new competence domains alongside their existing teams – the governance work, the judgment work, the human-oversight work that AI still can’t do reliably. That’s slower. It requires more deliberate leadership than posting a requisition.

If you’re looking at the GM headline and thinking the lesson is “move fast, cut legacy, hire new”, the data says you only have half the story.

The institutional knowledge in your existing workforce isn’t legacy overhead. It’s foundation.