Twice a year an email lands telling me the enrollment window for the employee stock purchase plan is open. I always enroll. Buying company stock at a discount is one of the better deals a regular employee gets offered, and I’ve worked plenty of places that had nothing like it, so I know what it’s worth.

Then there’s the part the money people keep saying.

Ask most financial advisors about loading up on your employer’s stock and they’ll tell you to be careful. I’m not an advisor, but the reasoning is easy to follow. Your salary comes from that company. Your health coverage comes from that company. If a big chunk of your savings sits in its stock too, a bad quarter doesn’t just dent your balance. It shows up the same month as a hiring freeze and a thinner bonus pool. Everything you depend on moves with one ticker symbol, which is the opposite of diversification.

Lately I keep seeing that same shape somewhere else.

Most of us now get our AI through work. The company picked the vendor, signed the contract, and handed out seats. It’s convenient, it’s paid for, and it’s probably the model you use all day.

So, what happens when the seat changes?

I don’t mean that as a doomsday question. Companies adjust this kind of thing constantly, for boring reasons. Usage costs more than anyone forecast, so everybody gets a token budget. A model gets swapped for a cheaper one that’s almost as good. A security review decides which tools can touch the codebase and which can’t. Any of that can land on a Tuesday, and none of it requires a villain. It’s budget.

Then there’s the bigger version. If you leave, or your role gets cut, your access walks out with your laptop. Everything you had learned to do with those tools, you suddenly can’t do, on exactly the day you need to show people what you can do.

AI is becoming part of how work gets done whether any of us voted for it. If your only real access runs through your employer, you’ve stacked two things in one place: how you earn and how you work.

So, months ago, I diversified. I bought a Mac Studio and started running open weight models on it.

I want to be honest about the trade-offs, because this isn’t a pitch. It was expensive. Not an impulse buy, not a gadget, a real decision I thought about for a while. And the open weight models I run are not as capable as the frontier models. For some tasks the gap is obvious within one prompt.

But I found a sweet spot. For a lot of what I actually do (drafting, summarizing, reading unfamiliar code, thinking out loud before I write) the local models are more than good enough. Nobody throttles me at the end of the quarter. Nobody deprecates the one I had gotten used to. There’s no renewal date and no seat to lose. The capability ceiling is lower, and the floor is mine.

The privacy turned out to matter as much as the access. Nothing leaves the machine. When I’m working through something half-formed, or something I wouldn’t paste into a tool owned by somebody else, that’s worth real money to me.

None of this means you should go buy the same hardware. That’s the expensive end of the range, and I went there because this is also how I tinker. A paid subscription you expense to nobody but yourself counts. A laptop running a small model counts. A cheap machine in a closet counts. The point isn’t the gear, it’s owning one path to AI that doesn’t route through your badge.

Here’s the exercise I’d actually do. If your work AI access vanished tomorrow morning, write down what you couldn’t do anymore. Then find the cheapest way to cover the top two things on that list yourself.

I still enroll in the stock plan every window. I just don’t want it to be the only thing I own.