Sulat.com / Writing on AITuesday, November 17, 2026
Agents/17 November 2026

McKinsey and Cognition team up to measure the return on Devin’s engineering work

McKinsey will help clients pick projects and redesign teams around Cognition's Devin agent. Its own May 2026 survey found 30% of director-level respondents reporting a fall in team productivity.

Black 3D Devin logo on a glass boardroom table beside a laptop of code, a ledger, a calculator and a stack of coins

At a glance

  • McKinsey and Cognition say their alliance will help enterprises choose engineering projects for Devin, set baselines for cost, speed and quality, and redesign teams around the agent.
  • McKinsey's May 2026 survey of 334 leaders found 25% of director-level respondents reporting meaningful acceleration and 30% reporting that team productivity had fallen.
  • Cognition says it measures Devin's value as equivalent human engineering hours converted to dollars, and backs it with a guarantee of usage credits up to US$10 million.
  • Cognition says Comarch set an 18-month goal to rebuild its 20-year-old ERP XL system with Devin, and Comarch's CTO says McKinsey is helping redesign how its teams work.

McKinsey & Company and Cognition, the company behind the Devin coding agent, have formed an alliance to help enterprises choose engineering projects for Devin, set baselines for cost, speed and quality, and redesign teams around the agent, according to McKinsey’s announcement. Business Insider reported the deal on 16 November, which was 17 November in New Zealand, and Cognition published its own post the same day.

Russell Kaplan, Cognition’s co-founder and president, said regulated enterprise clients want to know “what was done, what it would have cost without AI, and what completing the project makes possible.” Saptarshi Ganguly, a McKinsey senior partner, said doubling software development productivity takes more than deploying AI agents: “It requires fundamentally rethinking how work gets done.”

What each company brings

Cognition describes Devin as an autonomous software engineer that works across several engineering projects at once, connected to the systems and security processes a company already uses, and sends its work back to engineers for review. Engineers set direction and decide what ships. Devin is model-agnostic, meaning it matches different AI models to different kinds of work, and Cognition says that gets enterprises frontier performance for up to 60% less. It says it already works with six of the eight US global systemically important banks, the banks regulators designate as systemically important worldwide, and three of the six top US health insurers.

McKinsey says it has helped more than 50 companies rewire for AI-powered software development, and it will bring training and hands-on coaching for engineers, managers and senior executives. Cognition’s post cites McKinsey’s Rewired research, which found that across 20 leading transformations, companies that focused on one to three high-value domains such as software averaged an EBITDA uplift of about 20% and broke even within one to two years. EBITDA is earnings before interest, tax, depreciation and amortisation.

The survey behind the pitch

McKinsey’s survey of 334 product and engineering leaders in May 2026 found a split. Among the 173 respondents at director level or above, 25% reported meaningful acceleration, which McKinsey counts as organisations where over a quarter of teams at least doubled their productivity, and 30% said team productivity had fallen. Among the engineers surveyed, about 80% reported an average gain of roughly 3%, while the top 20% averaged 55%.

Bar chart: 25% of director-level leaders report 2x teams, 30% report productivity fell, McKinsey survey May 2026
In McKinsey’s survey, 25% of senior leaders report big gains and 30% report a fall.

McKinsey says the difference comes from how work is organised rather than which tools a team has. Respondents whose organisations redesigned their processes before adding AI were more than twice as likely to report productivity gains above 20% as those that layered AI onto existing ways of working.

Side-by-side: layer AI on existing work versus redesign the process first; redesigners more than 2x as likely to gain over 20%
Teams that redesigned the process before adding AI were more than twice as likely to report gains above 20%.

Cost pressure sits alongside that. Business Insider reports that Goldman Sachs, Cisco and JPMorgan have become more selective about who gets access to which AI tools and how they’re used. Cognition’s Art Levy said Devin’s approach suits large software modernisation projects that companies have put off, and pointed to one client whose enterprise resource planning update, expected to take five years, is being compressed into 18 months. “It’s no longer, ‘Can you get the job done?'” Levy said. “It’s, ‘Can you get the job done, and are you cheap at doing it?'”

How Cognition says it counts the value

In an interview McKinsey published on 16 October, Kaplan said tokens, seats and lines of code show how much a tool was used, not what it produced. Cognition’s estimator measures the equivalent human engineering hours for each completed Devin task, validates them against engineers’ own estimates, converts the hours into dollars and compares them with what the customer spent. Kaplan said Cognition calibrated it to undercount. Cognition backs it with an AI Productivity Guarantee: if Devin delivers less engineering value than a customer pays for, Cognition covers the gap in usage credits, up to US$10 million.

Four-step chain: Devin finishes a task, estimator counts human hours, hours become dollars, compared with customer spend
Cognition turns each finished task into dollars and sets them against what the customer paid.

Comarch, a Central European technology company, is the example in both companies’ posts. Kaplan said it set an 18-month goal to rebuild ERP XL, a 20-year-old system of about ten million lines of code, and ran a controlled proof of concept scored on quality, speed and total cost of ownership. Comarch’s chief technology officer, Tomasz Stachlewski, said McKinsey is helping the company redesign how its teams work. Kaplan also said Devin now writes roughly 95% of the code Cognition ships, and that its total code output grew about 7x in the past six months.

Organisations that want to deploy Devin with McKinsey can contact Cognition’s enterprise team, the post says.

Sources: McKinsey alliance announcement, Cognition alliance post, McKinsey, “Beyond the copilot”, McKinsey interview with Russell Kaplan, Business Insider.

Disclosure: I’m an OpenAI, Firecrawl and Devin ambassador.


Discover more from Sulat AI & Tech News

Subscribe to get the latest posts sent to your email.