Ford replaced its engineers with AI. Then rehired 350.

Hey friend, Happy 4th of July and 250th 🎆

This week Ford handed us the clearest proof I've seen of something I say all the time: your experience is what makes AI work. I couldn't not share it.

Here's what I wrote about it this week:

Ford spent three years replacing veteran engineers with AI. This month it quietly rehired 350 of them. Then it hit number one in quality for the first time in 16 years.

Without the people who knew the work, the AI was just guessing well.

What makes AI powerful is the augmentation it adds to your own, or your company's, deep and proprietary knowledge. Take that away and all it can deliver is generic output.

Ford pulled its experienced people out of the building before their judgment ever made it into the systems meant to replace them. The tools amplified weak inputs instead of catching the flaws.

Quality decreased and (ironically) their solution turned out to be the senior people they had asked to leave.

In my own career, there were many times when knowledge was held in basic docs but the harder lessons were only in the heads of team members. Capturing that knowledge in a formal way was always difficult, since people needed the discipline to write everything down, and even then it was a poor copy of the real processes and decision making. AI makes it easier to capture data from meetings and artifacts, but it still has the issue of knowing what is valuable in that trove of data.

I remember being a young consultant at Gemini. We had a fairly formal process for capturing project learnings, but you always wanted the veteran consultants on the engagement to lend their real client experience.

I once ran a diagnostic that on paper made sense for a client's workflow, only to find the data didn't give the insight I expected. Later, in a conversation with the analytics lead, she recommended a better option that delivered every time.

Your decades of hard won judgment are the edge that makes AI work, so bring it with you.

Capture the thing you know that isn't written down anywhere then use it for AI context.

The pattern is bigger than one automaker. These tools are only as good as the judgment you feed them and that judgment is the thing you have spent a career building so bring it with you.

Meta's research team released a system that decodes freely typed sentences from a wearable brain scanner, hitting an average of 61% word accuracy, a big jump over anything non invasive before it. They frame it as a path to tools for people who can't speak or type.

A real accessibility breakthrough for anyone who has lost speech or motor control. The question it bring up is who ends up owning your brain data? Read more

On July 1, Claude Fable 5 from Anthropic returned worldwide after a June export control directive had pulled it and Mythos 5 offline for weeks over a security finding. The administration lifted the order and the model is live again across the apps.

This is a good argument for not wiring your whole routine to a single tool you don't control. Read more

The UN launched an AI commission called “AI for Good” of company CEOs and heads of state, ahead of its Global Dialogue on AI Governance in Geneva starting July 6. The goal is to shape international rules for the technology. Read more

The administration is treating the most powerful AI models as instruments of American influence, restricting access even for longtime allies like Europe while cutting separate deals to lock up chips and critical minerals. Being a US ally no longer guarantees a seat at the frontier AI table.

It is a reminder that the AI tools are now shaped by trade and foreign policy, not just by the companies building them. Read more