Cognition and McKinsey announced a partnership this week to bring the Devin AI coding agent into large enterprises across the United States and Europe. Under the arrangement, McKinsey will help companies select projects, set baselines for cost, speed, and quality, and redesign engineering workflows, while Cognition supplies Devin for supervised software-development work. The deal marks one of the clearest signals yet that AI coding agents are moving beyond individual developer productivity tools and into structured, consulting-led enterprise transformation programs.
The announcement lands at a pivotal moment for AI-assisted software development, as the technology shifts from code-completion and chat-based assistance toward autonomous systems capable of modifying large, mission-critical codebases. European ERP vendor Comarch is already using Devin to rebuild its roughly 20-year-old XL system over an 18-month engagement, a project that illustrates how coding agents are being positioned for legacy-system modernization rather than greenfield development alone. With Google, Microsoft, and OpenAI all racing to expand their own agentic coding offerings, the McKinsey-Cognition tie-up suggests the next phase of competition will be fought not just on model capability, but on who can credibly sell transformation at enterprise scale.
From Autocomplete to Enterprise Transformation
For much of the past two years, AI coding tools were largely framed as productivity boosters for individual engineers: autocomplete suggestions, chat-based debugging, or small function generation. The Cognition-McKinsey partnership represents a deliberate break from that framing. Rather than selling Devin as a tool developers opt into, the two companies are packaging it as an enterprise transformation offering, complete with consulting-style project selection, performance baselines, and workflow redesign aimed at chief technology officers and engineering leadership rather than individual coders.
This positioning mirrors how McKinsey has historically sold large-scale technology and operations overhauls, applying the same playbook used for ERP rollouts or cloud migrations to agentic AI adoption. By establishing cost, speed, and quality baselines before deployment, McKinsey is attempting to solve one of the thorniest problems in enterprise AI adoption: proving return on investment before committing to full-scale rollout. For Cognition, the arrangement provides access to McKinsey's existing relationships with Fortune 500 companies, a distribution advantage that pure-play AI startups often lack.
Comarch's ERP Rebuild as a Proof Point
The most concrete evidence of this strategy in action is Comarch's use of Devin to rebuild its ERP XL system, a platform that has been in continuous development for roughly two decades. The 18-month project is emblematic of the kind of work enterprises have historically avoided automating: large, deeply entrenched codebases with accumulated technical debt, undocumented business logic, and significant operational risk if something breaks. Legacy modernization has traditionally required armies of contractors and multi-year timelines, making it an expensive and politically fraught undertaking for most organizations.
If Devin can meaningfully accelerate even a portion of that work under supervision, it would validate one of the most commercially significant use cases for agentic AI: not writing new greenfield applications, but untangling decades-old systems that companies are otherwise reluctant to touch. Comarch's willingness to use an AI agent on a system this old and this critical signals a level of confidence that many enterprises have not yet reached, and other vendors will likely watch the outcome closely before committing to similar engagements of their own.
A Crowded and Accelerating Market
Cognition is not operating in isolation. Google has pushed forward with Jules, Microsoft has expanded GitHub's AI agent capabilities, and OpenAI has both built out Codex and reportedly committed roughly three billion dollars to acquire coding startup Windsurf. Each of these companies is racing to establish itself as the default agentic layer for professional software development, and the capital being deployed suggests the market views this as a winner-take-most opportunity rather than a niche feature add-on.
What differentiates the McKinsey-Cognition approach is the explicit focus on enterprise sales infrastructure rather than developer-tool virality. Where Microsoft and Google can lean on existing cloud and productivity-suite relationships, and OpenAI can lean on brand recognition and raw model capability, Cognition is effectively renting McKinsey's enterprise credibility and consulting apparatus. That approach could prove decisive in industries such as banking, insurance, and manufacturing, where large organizations are unlikely to hand critical infrastructure to an AI agent without a trusted intermediary managing the rollout.
Supervised autonomy is the operative phrase. Enterprises don't want a black box rewriting their core systems; they want a partner who can establish baselines, manage risk, and prove the economics before scaling up.
The Economics Behind the Push
The timing of the partnership also reflects broader shifts in how AI is changing the cost structure of software development itself. A recent 600-run benchmark using Claude Code to implement a simplified version of Git across thirteen programming languages found that dynamically typed languages such as Ruby, Python, and JavaScript were both the fastest and cheapest to complete, costing roughly 36 to 39 cents per run, while statically typed languages cost between 1.4 and 2.6 times more under the same test. Those findings suggest that AI-assisted development is not merely changing how code gets written, but is beginning to reshape which languages and architectures make the most economic sense for AI-heavy workflows.
That economic lens matters for enterprise buyers evaluating a platform like Devin, where cost and speed baselines are central to McKinsey's sales pitch. If agentic coding tools systematically favor certain languages or architectural patterns, large organizations undertaking modernization projects may find themselves making technology choices based as much on AI-agent compatibility as on traditional engineering considerations. That dynamic, still in its early stages, could become one of the more consequential and underappreciated side effects of the shift toward autonomous software engineering.
Sources
- https://www.infoworld.com/article/4247952/how-ai-is-warping-both-developers-and-developer-tools.html
- https://www.infoworld.com/programming-languages/
- https://www.infoworld.com/article/4214999/go-1-27-brings-support-for-generic-methods.html
- https://www.infoq.com/Programming-Languages/news/
- https://developers.slashdot.org/story/26/11/15/0710240/rust-leaves-top-10-in-tiobe-index-making-room-for-go
- https://www.economictimes.indiatimes.com/tech/artificial-intelligence/operation-coding-war-thanks-ai/articleshow/121420810.cms
- https://github.blog/news-insights/octoverse/what-the-fastest-growing-tools-reveal-about-how-software-is-being-built/
- https://www.infoworld.com/application-development/page/6/
- https://www.infoworld.com/news/page/3/
- https://adtmag.com/articles/list/news.aspx











Leave a Comment