TechCrunch ran a headline this week that I keep thinking about: “Ford rehires gray beard engineers after AI falls short.” Apparently Ford leaned hard into AI to handle work that veteran engineers used to do, the kind of deep mechanical and manufacturing knowledge that takes decades to build, and then quietly went back and rehired the people they’d moved away from. The phrase “gray beard” is doing a lot of work there. It’s affectionate and a little embarrassed at the same time.

And it’s the most honest thing I’ve read about AI all month.

The thing AI couldn’t fake

A gray beard engineer at Ford isn’t valuable because they can recite torque specs. They’re valuable because they’ve watched a particular assembly line behave badly for fifteen years and they know, in their gut, that the weird vibration in bay three usually means a fixture is a half-millimeter out of true, and they know which supplier’s bolts strip if you look at them funny, and they know who to call when the official process says one thing and physical reality says another.

That’s context. Lived, specific, un-documented context.

You can’t prompt your way to it. Ford seems to have learned this the expensive way, which is the way most companies learn things.

Autonomous magic versus quiet help

Here’s where I land after watching a year of this: AI gets oversold every single time it’s pitched as a replacement, and it quietly overdelivers every time it’s pitched as an assistant. The framing is the whole ballgame. When a vendor walks into a boardroom and promises that their system will replace the senior engineers, the messy human knowledge, the institutional memory, the entire expensive department, they are selling a fantasy that demos beautifully and collapses on contact with a real factory floor where the edge cases live and breed and never appear in the training data.

But take the same model. Same weights, same capabilities. Now point it at one engineer and say “help this person move faster.” Suddenly it works. The model isn’t doing the impossible thing anymore. It’s doing the achievable thing, which is handling the tedious 70% while a human who actually understands the stakes stays in the loop for the 30% that matters.

Ford didn’t have an AI problem. They had a framing problem. Somebody sold them replacement when the product was augmentation.

Why this rhymes with browser agents

I write for a company that builds a browser agent, so take the bias as given. But I think Ford’s walk-back maps almost perfectly onto the fight happening in AI tooling right now.

Most AI agents are sold as autonomous workers. You give them a task, they go off to some cloud server, spin up a fresh headless browser that has never met you, and try to do your job from scratch with zero context about who you are or what you’re logged into. They hit a login wall on the third step. They guess wrong on the form your company uses. They have no gray beard memory of how your tools actually behave. It’s Ford’s mistake in miniature, replayed a thousand times a day across a thousand startups.

A browser agent does the opposite. It sits in your side panel, inside the tab you already have open, looking at the same authenticated session you’re looking at. Dassi doesn’t replace you logging into your CRM. It works inside the CRM you’re already logged into. The human context that Ford paid fortunes to rehire? Your browser already holds a version of it, in your sessions, your tabs, your open accounts. We wrote more about that idea in your browser already has the human context AI agents are missing, and it’s the same lesson Ford just paid for.

The 70/30 split is the actual product

Nobody at Ford was hired back to do everything. They were hired back because the AI couldn’t do the last unautomatable slice, and that slice turned out to be load-bearing.

The same is true for browser work. I don’t want an agent that pretends I don’t exist. I want one that drafts the reply while I decide whether to send it, that fills the form while I catch the field it got wrong, that pulls the data off the page so I can do the judgment part. That’s not a smaller vision. It’s the vision that actually ships value instead of press releases. We’ve argued before that keeping the human in the loop is a feature and not a limitation, over in AI is de-skilling workers and browser agents are the fix.

If you want to feel the difference, the Dassi Chrome extension is free and lives in your existing browser, using your existing logins. No fresh headless box pretending to be you.

Ford spent a fortune to relearn something simple. The gray beards were never the problem. Selling them as replaceable was. And every AI company about to make the same pitch should probably read that TechCrunch headline twice before they print the slides.