Cognition and McKinsey to Sell Devin AI Coder to Corporations: Claimed 60% Cheaper
Cognition, the maker of the Devin AI programmer, has teamed up with McKinsey. Consultants will pick projects and measure cost, speed and quality, while Devin takes on tasks and returns them for review. The promised 60% savings are the company's own claim: Devin routes simple work to cheaper models and complex work to expensive ones. In a survey of 334 executives, a quarter of leaders more than doubled productivity, but 30% saw it fall, so companies should record baseline metrics before starting a pilot.
AI-processed from TNW; edited by Hamidun News
Buying an AI programmer is easy. Getting it to deliver value across a company with thousands of developers is much harder. That gap is what the new alliance targets: Cognition, the maker of the Devin coding agent (a program that takes engineering tasks on its own and returns finished work for review), has joined forces with the consulting firm McKinsey & Company.
Together they will sell AI-driven development to large companies in the US and Europe. Two numbers from the announcement need unpacking right away. First: Cognition promises "frontier-model performance at 60% lower cost."
This means Devin does not run every task through the most expensive neural network. It sends simple work to cheaper models, and the final bill comes out 60% lower. Second: in a McKinsey survey, a quarter of the best companies more than doubled productivity with AI, while 30% saw it fall.
Who does what
The roles are simple. McKinsey handles the organizational side. Consultants help the client choose engineering projects that really matter and record a starting point: how much development costs now, how fast it moves, and how good it is. Without that baseline, there is no way to prove that AI improved anything. McKinsey also helps restructure teams and workflows around AI.
Cognition provides the tool itself. Devin takes on engineering tasks and returns the result to humans for review. Cognition co-founder and President Russell Kaplan framed the goal as helping leaders direct additional engineering capacity toward measurable business results and return on investment.
McKinsey senior partner Lari Hämäläinen explained why the alliance emerged now. According to him, until the end of 2025 many AI coding tools were suitable only for narrow tasks. New models handle longer and more autonomous work. But, as he noted, scaling that to 2,000 or 20,000 developers is a very different story. McKinsey says it has already helped more than 50 companies restructure development around AI.
Survey numbers: AI does not help everyone
Both companies cite a McKinsey survey conducted in the second quarter of 2026 among 334 leaders of product and engineering units. The picture is mixed. In the best-performing quarter of organizations, AI more than doubled productivity. But in 30% of organizations, productivity actually fell.
For practitioners, this is the main argument for the alliance itself. If the same technology doubles output at some companies and reduces it at others, a lot depends on implementation: which tasks go to the agent, how review is organized, and who is accountable for the result. That is the layer the companies are offering to sell consulting for.
Cheap models for simple work
Cognition also speaks about economics. According to the company, Devin works with several AI models at once and matches each to a particular type of task. This, the company claims, delivers frontier-model performance at a cost 60% lower. Vice President of Global Partnerships Art Levy explained the logic to Business Insider: expensive models can be kept for complex work. "You don't need to go to the dealership in a Ferrari," he said. These 60% are the company's own claim; the material contains no independent verification.
Cognition's enterprise base is already substantial. The company says it works with six of the eight global systemically important banks and with three of the six largest medical insurers. The European software company Comarch uses Devin to rebuild its 20-year-old XL ERP system over 18 months, and together with McKinsey it wants to extend this approach to other teams. In October, Business Insider reported that Cognition was considering a new funding round.
What follows from this
The market for AI coders is shifting from selling subscriptions to individual developers toward large deployments, where processes, metrics and change management matter. If you are responsible for engineering at a large company, before a pilot you should do what McKinsey proposes: record current cost, speed and quality, and choose specific projects. You should also decide in advance who will review the agent's work and how. The idea of distributing tasks across models of different prices is useful even without Devin: on simple tasks, an expensive model often just burns through the budget.
There are caveats too. The survey results show spread but do not explain its causes. The savings figures and client base so far come from the companies themselves. The alliance should therefore be read as a sign that the market is maturing, not as proof that any pilot with Devin will double output.
- Nov 21, 2026Skein Agent Erases Finished Stages From Context: 97% Success Rate, But More Expensive Than OpenCode
- Nov 21, 2026Instead of Cursor: Six Open-Source Coding Agents, OpenCode Has 228K Stars on GitHub
- Nov 21, 2026ClawDOS Writes and Runs Code on an IBM PC 5160: Primes Under 100 in 64 Seconds
- Nov 21, 2026Apple Is Building an AI Agent and Making It Harder for Outsiders to Get Access: Permission Now Takes 14 Steps, Not 4
Key Questions Answered
Who does what?
The roles are simple. McKinsey handles the organizational side. Consultants help the client choose engineering projects that really matter and record a starting point: how much development costs now, how fast it moves, and how good it is. Without that baseline, there is no way to prove that AI improved anything. McKinsey also helps restructure teams and workflows around AI. Cognition provides the tool itself. Devin takes on engineering tasks and returns the result to humans for review.
What are the key details on Survey numbers: AI does not help everyone?
Both companies cite a McKinsey survey conducted in the second quarter of 2026 among 334 leaders of product and engineering units. The picture is mixed. In the best-performing quarter of organizations, AI more than doubled productivity. But in 30% of organizations, productivity actually fell. For practitioners, this is the main argument for the alliance itself.
What are the key details on Cheap models for simple work?
Cognition also speaks about economics. According to the company, Devin works with several AI models at once and matches each to a particular type of task. This, the company claims, delivers frontier-model performance at a cost 60% lower. Vice President of Global Partnerships Art Levy explained the logic to Business Insider: expensive models can be kept for complex work. "You don't need to go to the dealership in a Ferrari," he said.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.