Track · AI in Product3 min read
McKinsey and Cognition Team Up to Prove AI Coding Pays Off
McKinsey and Cognition are pairing consulting with the Devin AI coding agent to prove ROI. Survey data shows AI doubled productivity at top firms — and cut it at nearly 30%.

Context
McKinsey and Cognition announced a partnership on Monday combining consulting with the Devin AI coding agent.
In a McKinsey survey of 300+ product and engineering leaders, the top quarter of organizations saw AI more than double productivity, while nearly 30% saw it drop.
One Cognition client compressed a five-year ERP modernization into 18 months using Devin.
Devin draws on multiple AI models, reserving pricier frontier models for harder work.
Goldman Sachs, Cisco, and JPMorgan have become more selective about AI tool access and usage.
One quarter of organizations saw AI more than double their software productivity, while nearly 30% saw it drop, according to a McKinsey survey of more than 300 product and engineering leaders. That spread is the business case behind a new partnership announced Monday: McKinsey & Company and the AI lab Cognition are joining forces to prove that AI coding is worth its cost.
Under the deal, McKinsey will advise clients on which engineering projects to target and how to redesign teams and workflows around AI. Cognition will supply Devin, its AI coding agent, which the company says can understand, plan, write, and test code. The pitch is a full-stack one: strategy and change management from the consultancy, execution capacity from the agent.
Why now? AI coding tools crossed a capability threshold
Devin's launch in early 2024 unsettled software engineers who feared AI would upend their workflows and their livelihoods. Since then, developer work has shifted from writing code to guiding AI output. But the biggest changes came in the past year, said Lari Hämäläinen, a senior partner in McKinsey's technology, media, and telecommunications practice.
Until the final months of 2025, most AI coding tools handled only narrow tasks: looking up technical guidance, checking syntax details. "That was about it," Hämäläinen said. Since then, more capable models can take on longer, more autonomous coding work.
The capability curve now outruns most organizations' ability to absorb it. "This is really a step-change phenomenon" that "requires a complete rethink" of software- and product-development processes, Hämäläinen said.
What does the productivity split actually mean?
The McKinsey survey result cuts both ways, and product leaders should read it as a warning, not a mandate. AI adoption is not automatically productivity-positive; nearly a third of surveyed organizations regressed.
One common pitfall: a small team experimenting with an AI tool is easy. "It's a completely different thing to scale it across 2,000 developers or 20,000 developers," Hämäläinen said. The failure mode is process redesign lagging the technology, not the technology itself.
That hunt for measurable returns has already changed enterprise behavior. Large companies including Goldman Sachs, Cisco, and JPMorgan have become more selective about which employees get access to which AI tools and how those tools are used.
How does model routing control the cost side?
Cost discipline is the second half of the value equation. Art Levy, VP of global partnerships at Cognition, said Devin can draw on multiple AI models: pricier frontier tools for harder engineering work, cheaper models for routine tasks.
"You don't need to drive a Ferrari to the grocery store," Levy said. AI routing tools, which direct tasks to models based on capability and cost, have become a popular way for companies to manage AI spending.
Levy pointed to a concrete payoff case: one client's enterprise resource planning modernization, expected to take five years, is being compressed into 18 months. That is the kind of deferred, large-scale software modernization work where Cognition argues Devin delivers.
"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?'"
What should product managers take from this?
The partnership signals where enterprise AI buying is heading. Vendors can no longer sell capability alone; they must sell cost-efficient outcomes and help buyers restructure teams around agents that operate autonomously for long stretches.
For practitioners, the Monday-morning lesson is sequencing: pick the right projects first, route work to the right-priced models, and budget as much effort for workflow redesign as for tool rollout. The organizations that doubled productivity did something different from the 30% that regressed, and the difference is far more likely organizational than technological. Expect every major consultancy to package a version of this play as agentic coding moves from pilot line item to core engineering budget.
via feeds.businessinsider.com (Source)
More from Marcus Bennett
Show full bio
Market editor covering regulation, policy and their impact on product management at Roadmap File.
56 articles