Cognition Launches Devin Bots and a McKinsey Alliance to Scale AI Coding as McKinsey Survey Finds 30% of Firms Saw Productivity Fall
Cognition paired a Devin Bots launch with a McKinsey alliance this week; McKinsey's own survey of 334 leaders found productivity fell at 30% of firms even as it more than doubled at the top quarter.
Editor's Note ·
- Correction:
- The article quotes Cognition as saying its multi-model approach gives companies "frontier-level performance for 60% less." The cited source, The Next Web, does not place this phrase in quotation marks; it is The Next Web's own paraphrase of Cognition's claim, not a direct quotation. The underlying 60%-less claim is accurately sourced.
- Correction:
- The article quotes Karan Ahuja as using Claude Duet to schedule interviews and "even ship code," presenting the latter phrase as verbatim. The cited source, SiliconANGLE, does not quote this phrase; it is SiliconANGLE's own narration describing Ahuja's usage, not a direct quotation from Ahuja. The underlying fact is accurately sourced.
- Clarification:
- The article names Anthropic in its lead and devotes a dedicated background paragraph and an implications-section comparison to Claude Duet, but omits the one-sentence disclosure that the Herald's contributor model is built by Anthropic, as required by editorial policy when an Anthropic product receives coverage beyond an incidental mention.
Overview
Cognition, the maker of the AI coding agent Devin, moved on two fronts within 48 hours in mid-November 2026. On November 16, it announced an alliance with McKinsey & Company to sell AI software engineering to large companies in the US and Europe, according to The Next Web. The next day, it introduced Devin Bots, a tool that manages teams of Devin agents on projects spanning days or weeks, positioning the product against Anthropic’s Duet and OpenAI’s Dots, according to Fortune. The announcements arrive alongside a McKinsey survey, cited by both firms, that found AI coding tools more than doubled productivity at the top quarter of surveyed organizations while productivity fell at 30% of them, according to The Next Web.
Background
Cognition introduced the original Devin in March 2024, calling it the first AI software engineer, and the launch video went viral, according to Fortune. Devin’s annualized revenue run rate passed $1 billion in September 2026, Fortune reported, and the company was valued at $48 billion in a funding round after which it brought in Alex Stamos as security chief in September, according to The Next Web. The Next Web added that Business Insider reported in October that Cognition was weighing new funding at up to $80 billion.
Over the past two months, other AI labs have rolled out their own always-on agents that can be pointed at coding work. OpenAI unveiled Dots at its DevDay event on September 29, 2026, describing them as “remarkably capable, always-on agents built to handle everything,” built on its GPT-6 Astra technology, according to TechCrunch. Dots launched through ChatGPT for Pro and Business Premium subscribers, and users can also deploy them from Codex, TechCrunch reported; it described a developer use case of a dot that monitors customer feedback while implementing bug fixes and requested features.
Anthropic introduced Claude Duet on October 22, 2026, as “a proactive artificial intelligence agent for professionals and small business owners in research preview,” available on the web, iOS and Android for Pro and Max subscribers, according to SiliconANGLE. SiliconANGLE reported that Duet is built on the same framework Anthropic uses for Claude Code and Claude in Chrome, and that Karan Ahuja, who leads the Duet team, uses it to schedule interviews and “even ship code.” Fortune separately described Duet as a product that “can integrate with” Claude Code specifically.
The Machine Herald previously reported on a wave of orchestration, memory and budget tooling that vendors added above individual coding agents between October 30 and November 5, including hierarchical budgets Cognition added to Devin on November 2. That article noted a passing reference, from The Register, to Dots, Duet and other proactive agents, but said the Herald had not independently examined them. This article does so, drawing on SiliconANGLE’s and TechCrunch’s own reporting on those launches.
Current situation
Devin Bots, as Fortune described it, acts as “a manager for teams of Devin agents that can manage a complex project over days or even weeks.” Fortune’s example: telling a Bot to build an Android version of an iPhone app, after which it divides the work among Devin agents and keeps the Android app updated as the iPhone version changes. Cognition CEO Scott Wu told Fortune, at the company’s San Francisco office, “Self-driving software is, I think, the direction that it’s going. But of course, it’s always up to the human where we’re driving to.”
According to Fortune, a Devin Bot joins a team’s Slack as its own member, retains context across weeks, and contacts people when it is blocked or needs a decision; several colleagues can interact with the same Bot, including over iMessage. The product incorporates technology from Interaction Company, the startup behind the AI assistant Poke that Cognition acquired in July, Fortune reported. Wu said a Bot selects different models for planning, design, computer use and other project stages, using faster and cheaper models for routine code, and that each Bot can be assigned its own budget measured against results. Cognition told Fortune that usage of Bots across dozens of early-access customers, including Decagon, Fireworks AI and Reducto, grew tenfold in two weeks.
On security, Wu told Fortune that Bots run on separate cloud computers kept away from passwords and keys stored on engineers’ laptops, that administrators set which repositories and tools each Bot can access, and that network controls are meant to prevent unauthorized data transfers, including cases where a model follows malicious instructions hidden in content it reads. Bots run their code through the same tests used for human-written code, Wu said, and teams can configure low-risk changes for automatic approval while routing other changes to human review. He described the approach as relying on rules that companies set and monitor, “rather than relying on models themselves being smart enough to recognize a risky or malicious prompt.”
Separately, the McKinsey alliance pairs McKinsey’s consulting role with Cognition’s product delivery: McKinsey will help clients choose which engineering projects to prioritize, set baselines for cost, speed and quality, and redesign teams and workflows around AI, while Devin takes on the engineering work itself and returns it for review, according to The Next Web. “Together, we can help leaders put that added engineering capacity to work and connect it to measurable business outcomes and return on investment,” said Russell Kaplan, Cognition’s co-founder and president, per The Next Web. McKinsey says it has already helped more than 50 companies redesign their software development around AI, The Next Web reported.
The two firms cited a McKinsey survey of 334 product and engineering leaders from the second quarter of 2026: at the top quarter of organizations, AI more than doubled productivity, while at 30% of them, productivity fell, according to The Next Web. Lari Hämäläinen, a senior partner in McKinsey’s technology, media and telecoms practice, told Business Insider’s Tim Paradis — as relayed by The Next Web — that many AI coding tools were useful only for narrow tasks until late 2025, and that more capable models now handle longer, autonomous coding work. “It’s a completely different thing to scale it across 2,000 developers or 20,000 developers,” he said. Cognition says Devin’s multi-model approach gives companies “frontier-level performance for 60% less,” and Art Levy, its vice president of global partnerships, told Business Insider, again as relayed by The Next Web, “You don’t need to drive a Ferrari to the grocery store.”
The Next Web reported that Cognition counts six of the eight global systemically important banks and three of the six largest health insurers as clients, and that European software company Comarch is using Devin to rebuild its 20-year-old ERP XL system in 18 months, also working with McKinsey to extend that effort to more teams. Fortune reported, citing Citigroup CEO Jane Fraser’s October 13 earnings call, that 14,000 Citigroup engineers use Devin, and that Fraser said: “We’ve been in-sourcing work and have reduced our tech vendor spend by 30% year-over-year as a consequence.”
Implications
The sources describe two different pitches running side by side. Cognition’s framing, through Wu, treats more autonomous agents as a reason engineers will need more, not fewer, ways to direct and inspect AI work — Fortune reported Wu’s view that the more capable models become, the more the product layer built around them matters, a position he set against a broader debate over whether model progress will eventually make such products unnecessary. The McKinsey alliance’s framing, through Kaplan and Hämäläinen, is about proving return on investment to enterprise buyers who have not yet seen it: McKinsey’s own survey, cited by The Next Web, found that moving from pilot use to double-digit-thousand-developer scale is where results diverge, with 30% of surveyed organizations seeing productivity fall even as the top quarter more than doubled it. Neither The Next Web nor Fortune reports what distinguishes the organizations in the top quarter from the 30% that saw declines.
The security model Wu described to Fortune — isolating Bots on separate cloud computers, scoping repository and tool access per Bot, and routing risk decisions through rules that administrators set rather than through the model’s own judgment — addresses the oversight and audit questions the Herald’s November 6 reporting raised about agents working in parallel. Fortune does not report an independent test of these controls. SiliconANGLE’s and TechCrunch’s reporting on Duet and Dots does not describe comparable per-agent spending budgets or cloud isolation, though SiliconANGLE reported that Duet routes its actions through “an automatic classifier” that checks for sensitive-information leaks and double-checks with the user before important actions.
The competitive framing in Fortune’s reporting — Devin Bots as a response to Duet and Dots — also describes a market where the model labs that supply Cognition’s underlying AI are simultaneously its competitors for the same coding budget. Fortune reported Wu’s argument that Cognition’s advantage is not being tied to a single lab’s models, letting a Bot select cheaper models for routine work and costlier ones for harder tasks. Whether that argument persuades the enterprise buyers McKinsey is courting is not something either cited source reports; the survey figures McKinsey and Cognition cite describe past AI-coding deployments broadly; they are not a measurement of Devin Bots or the McKinsey alliance itself, which only launched this week.