The $280 Overnight Surprise

A friend pinged me last week, half-laughing, half-panicking 😅 . He’d been running a couple of AI coding agents overnight, just background refactoring and test generation. He woke up to a $280 bill. One night. One developer.

The New Math: Tokens vs Headcount

Welcome to the new math of software engineering. We spent the last decade arguing over whether to hire senior or junior engineers. Now engineering managers are about to add a third line item to that decision: how many tokens does this task deserve?

The Structural Shift

Think about what this actually means structurally. A senior engineer running an agentic loop to architect a new service might burn $200 in an afternoon. A junior engineer, handed the same task without AI, takes three days and costs roughly the same. The math works, provided the output is commensurate. But organizations haven’t built the muscle to measure that yet.

What to Expect in the Next 18 Months

Here’s what I expect to see in the next 12 to 18 months: → Token budgets by role. Just as cloud teams get AWS spending limits, engineering orgs will start allocating token budgets. Senior devs get higher limits. Juniors get constrained, which creates a strange dynamic where AI access becomes a proxy for seniority. → A new class of “prompt frugality.” The best engineers won’t just write great code or great prompts; they’ll get the most output from the fewest tokens. Efficiency as craft. → Finance teams suddenly caring about context windows. Model selection gets pulled out of engineering into procurement. Why use the most powerful model when a smaller one handles 80% of the task at 10% of the cost? → The ROI conversation gets uncomfortable. Teams are in the honeymoon phase right now: shipping faster, feeling the magic. But when the monthly AI bill rivals headcount costs, boards ask hard questions. That’s not a bad thing. Scrutiny breeds intentionality.

Efficiency as Personal Craft

The organizations that win won’t be the ones who spend the most on tokens. They’ll be the ones who figure out fastest which problems deserve an overnight agent, and which ones deserve a single, well-crafted prompt. Interesting times, indeed.

Cursor

IDE AI Native

The AI-first code editor that helps you build software faster.

Try Cursor

Aider

CLI Agentic

Command line tool that lets you pair program with LLMs, to edit code in your local git repo.

Explore Aider

LLM Price Check

Tools FinOps

Compare pricing across different LLM providers to optimize your token budget.

Compare Prices