You’ve probably felt it. That creeping dread when you see a new AI demo and wonder if your job is next. But you’re worrying about the wrong thing.
The real threat isn’t a machine that thinks like a human. It’s a human who knows how to spend $100,000 a year on tokens — and turn that into a one-person team.
That number comes from Clay Bavor, co-founder of Sierra, a company that builds AI agents for enterprises. In a recent interview, he dropped a bombshell: top engineers at his company are burning through AI token budgets that high. And it’s not waste. It’s the new economics of production.
The faster you execute the wrong direction, the larger the waste. Speed without judgment is just organized chaos.
Here’s what’s actually happening: AI is shifting from a nice-to-have chat tool to core enterprise infrastructure. And that shift forces companies to rethink everything — budgets, permissions, team structures, even the definition of a job.
1. The Model Portfolio (Not Just One Model)
You’ve probably assumed that cheaper, smaller models will eventually replace expensive frontier ones. Wrong. The opposite is happening.
Low-risk tasks like customer Q&A, data extraction, and routine workflows can (and should) use cheap models. But high-value tasks — architecture decisions, product innovation, complex analysis — demand the best. Companies will use a mix. The real skill is in routing work to the right model at the right price.
It’s like cloud computing. You don’t run everything on the most expensive server. You also don’t run a critical system on the cheapest. The same logic applies to AI: Your model selection should be driven by business risk, not benchmark scores.
2. Token Budgets as a New Resource
Imagine a world where your employee’s cost includes not just salary and laptop, but also a token allowance. That’s the future. Some engineers at Sierra already consume $100k worth of tokens annually. If that lets one person do the work of a small team, it’s a bargain.
But here’s the trap: companies that track token usage as a cost-control metric will drown in waste. The right approach is to tie token spend to workflow outcomes. How much did it shorten the dev cycle? How many support tickets did it deflect? How much reusable code did it generate?
Token budgets aren’t a perk. They’re a production input — and the companies that treat them as overhead will lose to the ones that treat them as capital.
3. Smaller Teams, Bigger Scope
When AI automates the grunt work, you don’t need a dozen people to crank out code. You need a few sharp people who can direct the AI, validate its output, and handle the edge cases.
This doesn’t lower the bar. It raises it. The ability to decide what to do becomes more valuable than the ability to do it. A product manager’s job shifts from designing human-software interactions to designing human-AI-business collaborations.
One engineer with a $100k token budget can outproduce a team of ten who don’t know how to orchestrate AI. The competitive moat isn’t access to frontier models. It’s the ability to build automated governance, permission boundaries, and accountability systems around AI.
4. The Real Infrastructure: Governance, Not Chatbots
Most companies start AI adoption by giving everyone a ChatGPT account. That’s like giving everyone a hammer and calling it construction. The real work is in building the scaffolding: identity, permissions, audit logs, human-in-the-loop approvals, and fallback procedures.
At Sierra, AI agents can’t cross permission boundaries. If an employee doesn’t have access to a file, the AI can’t read it either. The model’s power is amplified, but its access boundaries are enforced. They also built an internal agent called Pinecone that screens interview notes — but the hiring decision stays with a human.
This is the pattern: AI analyzes, suggests, and flags. Humans decide and take responsibility. Without that infrastructure, every AI capability is a potential data leak or decision disaster.
AI can amplify your organization’s capabilities. It can also amplify your organization’s chaos. Which one depends entirely on how you govern it.
The Bottom Line
That $100k token budget isn’t a benchmark to follow blindly. It’s a symbol of a deeper shift. AI is moving from a tool you use occasionally to a resource you invest in deliberately.
Three phases: first, employees use AI for writing and research. Second, AI connects to your business systems. Third, you redesign your entire operating model around AI — roles, budgets, processes, and accountability.
Companies that reach phase three won’t just have better AI. They’ll have better organizations. They’ll know how to allocate tasks across humans and models, how to measure output instead of activity, and how to keep the human in the loop where it matters.
So stop worrying about being replaced by AI. Start worrying about the engineer who knows how to spend $100k on tokens — and who works for a company that knows how to spend it wisely. That’s the real competition.
FAQ
Q: Is $100k per engineer a realistic budget for most companies?
A: No, it's a signal, not a standard. It applies to elite engineers doing high-value work. For most roles, the budget will be lower, but the principle stands: AI is a production input that needs to be measured and allocated like any other resource.
Q: What's the practical takeaway for a product manager?
A: Stop evaluating AI models by benchmarks alone. Start mapping model capabilities to specific business tasks, risk levels, and workflow steps. Your job is to design the handoff between AI and human judgment, and to ensure that every AI-generated output is traceable and accountable.
Q: Doesn't this contradict the idea that AI will make jobs easier and cheaper?
A: It does. The contrarian truth is that AI lowers execution costs but raises the value of strategic direction. The easier it is to do things, the more costly it is to do the wrong thing. So the real skill shift is from 'doing' to 'deciding' — and that puts a premium on human judgment, not a discount.