AI spending can run negative. Qodo’s CEO built an ROI equation to fix it.
The New Stack Alex Wilhelm
Qodo caps employee AI use at $10,000 in tokens a month. Its CEO says the real goal is to measure AI ROI before the bills outrun the gains.
Based on reporting by The New Stack, Alex Wilhelm — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Qodo just raised a $70 million Series B, but it isn’t treating internal AI like free candy. CEO Itamar Friedman says engineers get up to $10,000 in tokens each month, and most never come close. The cap, he says, is there to force visibility, efficiency, and a hard question: which automation path is actually worth the money?
That discipline matters because Qodo’s own AI infrastructure bill is heading the other way. The company says the cost of running its product for customers is growing at roughly 5x year over year, helped by more adoption and by agents taking on longer tasks. At the same time, Qodo says it is pushing costs down by routing work more efficiently and tightening inference.
Inside the company, AI is woven through the workflow. Qodo runs its own pull requests through Qodo, says the product powers the full software development life cycle, and uses tools like Slack, Notion, Google Workspace, and models from several providers, including Google. Claude Code is still the internal favorite, but OpenAI’s Codex has been gaining ground fast.
Friedman’s bigger point is that speed alone is a trap. Qodo says it sees roughly double the number of PRs every couple of months, while bugs and incidents are falling. That is the kind of outcome he wants people to translate into a simple equation: benefits over costs. If AI helps write more code but slows review, approval, or governance, the company can spend a fortune and end up nowhere.
That logic already changed one Qodo experiment. The team went hard on email automation, then backed off when the system couldn’t match tone or pull tasks from messages reliably. Friedman’s lesson is blunt: start with automation, then hand the messy bits back to human judgment. Imperfect measurement beats vibes, which is a polite way of saying most AI budgets are still being managed like someone found a credit card in the sofa.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI everyone keeps skipping: the bill arrives whether the demo was pretty or not. Qodo’s trick is embarrassingly simple, which is why it matters — put a number on the upside, put a number on the drag, and stop pretending “more AI” is a strategy. Plenty of companies are still buying tokens like they’re snacks at a conference booth.
Read more about this at: The New Stack
