I once watched a team lead argue against buying a $20/month Git tool for engineers.
The tool saved maybe 15 minutes a day per developer. Nothing dramatic. Just a small quality-of-life improvement that made rebasing, cherry-picking, and conflict resolution less painful. A handful of engineers asked if the company would cover it.
After all, we already covered IDEs and CI/CD integrations.
The answer? “We pay you well enough. If you want it, buy it yourself.”
That stuck with me. It was a reasonable-sounding position from someone who genuinely believed they were being fair. But it revealed something about how that company thought about its engineers. Tools were a personal preference, not a business investment.
Cloud Budget Debates
This debate comes back every few years wearing different clothes.
I’ve seen it play out with hardware budgets. With conference attendance. With professional development stipends. With cloud credits.
At one company that used AWS heavily, a group of engineers proposed getting individual AWS credits to experiment and learn on their own. They desperately wanted to learn AWS. But it’s expensive.
The organization’s fear was predictable: someone would spin up a massive instance, forget about it, and rack up a five-figure bill over a weekend.
Our solution was a compromise. We Terraformed a shared “sandbox” account, wiped on a regular schedule. Engineers could experiment freely without risk to production or to the company’s wallet. Nobody owned the sandbox. Everybody benefited from it. The total cost was a rounding error compared to the production AWS bill. Meanwhile the team learned things they never would have otherwise.
That sandbox made us better engineers. It was an investment that paid for itself many times over.
AI Tooling – The Great Token Debate
Now the debate is about tokens.
Jensen Huang told the All-In Podcast that if a $500,000 engineer didn’t consume at least $250,000 worth of tokens in a year, he’d be “deeply alarmed.” NVIDIA is reportedly trying to spend $2 billion on tokens for its engineering team. Huang framed token budgets as a recruiting tool — something engineers will soon be asking about in interviews alongside salary, equity, and bonuses.
That’s a bold position from someone who sells the infrastructure that processes those tokens. Take it with appropriate skepticism.
But he’s not entirely wrong.
Engineers with access to AI tools are more productive. That’s not hype — I’ve experienced it firsthand. The question is how to give engineers access.
Matthew Levy wrote a thoughtful LinkedIn post pushing back on the stipend model. His argument boils down to something I think is important: a fixed token budget sends the wrong message. It tells engineers “use as much AI as possible and stop when the tokens run out.” That doesn’t encourage productive use of AI. It encourages maximum use, which isn’t the same thing.
He’s right. Token spend varies wildly between tasks. It varies within a single project — lighter during planning, heavier during execution. The real bottleneck was never the tokens allowance, it’s the amount of attentions humans can afford to spare. No one can meaningfully supervise nine AI-assisted work streams at once. At some point you’re just burning compute without actually thinking about the output.
I’ve managed multiple teams and projects simultaneously. I can tell you from experience that my value decreased in direct correlation to the number of things I was tracking. More context switches, less depth. More activity, less quality. I don’t want that for my engineers either.
Finding balance
So where does that leave us?
I think the answer looks a lot like that AWS sandbox. Not a fixed stipend. Not “buy your own tools.” Something in between — structured access with enough freedom to learn and experiment, but with guardrails that keep the focus on productive work rather than raw consumption.
The companies I’ve seen get this right share a few things in common. They treat tool access as a team investment, not a perk. They create shared environments where experimentation is encouraged. They measure outcomes rather than spend. And they trust their engineers enough to give them the resources without micromanaging every dollar.
The companies that get it wrong? They fall into one of two traps. Either they refuse to invest — “we pay you enough, figure it out” — and watch their best engineers leave for places that take their growth seriously. Or they throw money at the problem without structure — here’s $250k in tokens, go be productive — and wonder why the results are underwhelming.
Neither extreme works. They never have.
We’ve been having this argument for as long as I’ve been in tech. The tools change. The answer doesn’t. The companies that figure out how to give engineers meaningful access to AI tools will be the ones whose teams build the best things over the next decade.
Invest in your people. It’s more than worth it.