The unit of work in software has changed fundamentally. For sixty years, developers wrote instructions that machines executed deterministically — one function at a time, one Jira ticket at a time. That era is ending. The new unit of work is the token: not an instruction, but a unit of purchased intelligence.
In early 2026, StrongDM's CTO disclosed that his three-person team targets $1,000 a day in token consumption — with no handwritten code. Similar numbers are arriving from across the industry. Anthropic itself spent $2.66 billion on AWS through September 2025 against an estimated $2.55 billion in cumulative revenue over the same period. More than 100% of its top-line revenue went to AWS alone, before accounting for Google Cloud spending.
Perplexity spent well over 164% of its entire 2024 revenue across AWS, Anthropic, and OpenAI combined.
These companies are not trying to find a more expensive way to do things. They are operating in a fundamentally different paradigm where intelligence is now a purchasable input — a commodity with a price curve and a consumption curve.
The price curve alone is remarkable. Inference costs have been falling at rates that make Moore's Law look gentle: somewhere between 10x and 200x a year, depending on the benchmark. GPT-4 equivalent performance cost $20 per million tokens in late 2022. It costs about $3 today. Claude Sonnet runs at $3 per million input tokens, and the raw commodity is deflating faster than any computing resource in history.
When a resource gets cheaper, people don't use less of it — they use an enormous amount more. This is called Jevons' Paradox. As AI gets more efficient and accessible, its use skyrockets. The average organization now spends $85,000 a month on AI, up 36% year-over-year. And the share planning to spend more than $100,000 monthly has doubled from 20% to 45%.
OpenAI is reportedly planning multiple agent pricing tiers: from $2,000 a month for knowledge worker agents, up to $10,000 a month for specialized coding agents, and up to $20,000 for AI research agents. The point is not whether they will launch at exactly that price point. The point is that this is now a conceivable price point when organizations are reframing their entire budget around compute measured in tokens.
Enterprise buyers are doing the math and concluding that even at $20,000 a month, the price is cheap relative to the cost of the human professionals they'd otherwise employ. At many companies, it looks like they'll keep their PhD researchers — but they'll give each researcher a mini-me, expanding their footprint by 2-3x.
The New Developer Tracks
The role of developers is now differentiating into at least three distinct tracks, each with different skill requirements, compensation dynamics, and career trajectories.
Track one: Orchestrators. These developers do not write code but specify outcomes and manage the intelligence that produces those outcomes. Their core skills are system design, specification writing, quality evaluation, and token economics. They think in terms of agent architectures, context windows, eval frameworks, cost per outcome. They're effectively factory managers with intelligence. Their value scales with the volume of intelligence they can direct, meaning their compensation will likely correlate with token budgets rather than lines of code.
Track two: Systems Builders. These developers build the infrastructure that orchestrators use — agent frameworks, evaluation pipelines, context management systems, routing layers that send the right task to the right model at the right cost. This is deep technical work closer to traditional systems engineering but with an entirely new stack. They need to understand model behavior at a mechanical level: how context windows affect output quality, how different architectures handle different task types, how to build reliable systems on top of probabilistic components.
Track three: Obsolete. The standard narrative has been binary — either AI replaces developers or it doesn't. That framing misses what is actually happening. The role is differentiating rapidly, and one category of developer is about to become obsolete.
Token Economics as Core Competency
In the old paradigm, the scarce resource was time — specifically, developer time. You hired engineers, gave them tools, and the constraint on output was how many hours of skilled labor you could deploy. The management challenge was headcount planning, recruiting, retention — all the machinery that goes with human capital management.
In the new paradigm, the scarce resource has changed. Raw intelligence is abundant and getting massively cheaper. What's scarce now is knowing how to aim those tokens, how to structure context, how to route tasks to the right model at the right cost, how to build agent loops that sustain quality over time, and then measuring whether the intelligence you're purchasing actually produces the outcomes you need.
This creates an entirely new organizational capability: token management, or intelligence operations, or context engineering. What matters is that it's a real skill — measurable — and organizations that build it are starting to pull away from everyone else.
"Intelligence is now purchasable, and it turns out we have a huge appetite to buy it."
A16Z's Enterprise AI survey found average enterprise LLM spend hit $7 million in 2025 — up from $4.5 million just two years prior. Projections show it's going to reach eight figures and $11 million plus in 2026. That spending has shifted from an innovation budget into a centralized IT and business unit budget.
The enterprises that have figured this out are actively building internal platforms that route work to the right model at the right price point: Haiku for cheap stuff, Opus for hard stuff, Sonnet for the middle ground. They're treating token spend not as a cost to minimize but as a lever to maximize ROI and value. They'll negotiate custom API agreements with hyperscalers, commit to consumption floors in exchange for dedicated capacity and volume pricing.
But token management can go catastrophically wrong — and it matters more when you're spending more on it. Cursor, which became a billion-dollar revenue AI coding editor really fast, found itself in a structural trap: it sends essentially all its revenue to Anthropic in API costs. When Anthropic introduced priority service tiers and raised pricing, Cursor's costs exploded overnight, forcing it to gut out its unlimited $20/month plan and introduce a $200/month tier. The subreddit turned into a complaint forum.
The lesson is not that tokens are expensive or hyperscalers play games of gotcha. The lesson is that token economics is now a core business competency, and companies that don't figure out how to master it are just one supplier pricing change away from being in a crisis.
Critics Might Note
Some argue this framing oversimplifies the complexity of AI adoption across different industries and company sizes. Others might point out that the price curves described here could reverse — or plateau — making some of these projections speculative rather than predictive. And the three developer tracks, while useful, may not capture the full diversity of roles emerging in the AI era.
Bottom Line
The strongest part of this argument is that computing has fundamentally changed form — from instructions to tokens — and that changes everything about how developers work, what they get paid, and which skills matter. The biggest vulnerability is strategic: organizations are spending enormous sums on intelligence without yet mastering how to control those costs, and one pricing change could flip profitable operations into crisis overnight. Watch the token economics of your own industry carefully — this shift is already happening, and it's moving fast.