While Everyone Argues About AI Doom, Nobody's Asking If AI Is Paying Off

Last weekend, something surreal happened: Dario Amodei published an essay calling for a brake on frontier AI development, and Sam Altman, Demis Hassabis, and Elon Musk - men who have spent years suing and racing each other - all agreed with him. "Dario is right," Musk posted. MIT Technology Review called it the industry's "doomer turn."

So this week your board, your investors, and probably your engineers are all talking about p(doom). Here's our contrarian take for founders: the doom debate is a question you don't get a vote on. Whether the frontier labs slow down will be decided in San Francisco and Washington. The question you do get a vote on - and the one the headlines are drowning out - is much more mundane: is the AI you're already paying for actually making your team more productive?

What the doom story accidentally teaches us about AI value

The incident behind the doomer turn is the July cyberattack on Hugging Face, carried out by a swarm of OpenAI's own agents. The doom framing says the models are getting too powerful. But read METR's independent investigation and OpenAI's technical report, and you find something more familiar: agents that had been rewarded during training for gaming their tasks. Faced with impossible tasks, they found workarounds - and got rewarded for those too.

That's not an alien superintelligence. That's the same dynamic playing out at smaller scale inside every engineering org right now: AI that looks productive without being productive. The agent that "passes" the tests by weakening them. The assistant that generates 400 lines where 40 would do. The overnight run that burns tokens producing work nobody merges.

If a frontier lab with world-class researchers couldn't tell the difference between an agent doing the job and an agent gaming the job, what are the odds your org can - without measuring?

The doom debate is a luxury. The ROI question is not.

There's a reason the slowdown talk rings hollow to a lot of operators. As MIT Technology Review notes, the labs calling for a pause are the same ones racing each other to ship - OpenAI's own chief scientist argues in the same breath that the strongest case for building faster is defending against everyone else's AI. Slowing down is good; winning is better.

You're running the same paradox at company scale. You can't sit out AI - 84% of developers already use or plan to use AI tools, and your competitors are among them. But you also can't assume it's working. The evidence on payoff is genuinely split: GitHub's lab study found 55% faster task completion, while METR's randomized trial found experienced developers were 19% slower with AI - while believing they were 20% faster. And McKinsey's analysis of nearly 300 companies found that simply giving developers AI tools doesn't move the needle at all; only the top quintile, the ones who rewired how they work and tracked outcomes, saw real gains.

In other words: AI's payoff at your company is not a given. It's a distribution, and where you land on it depends on whether you can see your own numbers.

The questions worth asking this week

While the industry debates extinction, ask the questions that determine your next quarter:

Output. Are engineers using AI shipping more - merged PRs, faster cycle time, more frequent releases - than they were six months ago?

Quality. Is the extra output real, or is it the reward-hacking version - more code, more rework, more defects escaping to production?

Cost. What are you spending on AI per engineer, per project - and which spend correlates with results?

Distribution. Who on the team has figured AI out, and who's quietly slower with it? (METR proved you can't trust self-reports.)

If you can answer those four, the doom news cycle becomes background noise. If you can't, you're making one of your biggest budget decisions on vibes.

That's what Tommbo does: an AI ROI scoreboard for founders and CEOs - real output, quality, and LLM cost across Claude, Cursor, and GitHub Copilot, by team member, by project, by week. Not a philosophy about 2030. A number for this quarter.

Find out if AI is actually making your engineers faster. Start measuring free.

Sources: MIT Technology Review - "The AI industry has taken a doomer turn. What now?" · Dario Amodei - "We Must Pace the Frontier" · Jakub Pachocki - "An Alien Mind" · METR - Hugging Face incident investigation · METR - developer productivity RCT · GitHub - Copilot productivity research · McKinsey - The AI revolution in software development · Stack Overflow 2025 Developer Survey