This website uses cookies

Read our Privacy policy and Terms of use for more information.

A photorealistic, futuristic illustration of a friendly white humanoid robot working at a glowing blue holographic keyboard inside a high-tech software development workspace. The robot has a smooth, rounded helmet, a glossy black faceplate with two curved, bright blue eyes resembling a cheerful expression, and white metallic body panels with illuminated blue accents. Its articulated robotic fingers hover over the transparent keyboard as multiple floating digital screens surround it. On the left, holographic displays show colorful programming code and a software workflow diagram with interconnected tasks. On the right, an “Active Projects” dashboard lists software development activities, including building a new feature, fixing bugs, updating documentation, running tests, and deploying to production. Beneath it, an “AI Agents” panel displays four agents with individual progress bars, illustrating multiple AI systems working on separate assignments. Electric-blue lighting, transparent interfaces, and reflections across the dark workspace create a realistic, high-tech atmosphere. The image conceptually represents Cognition's Devin Bots coordinating AI coding agents to manage ongoing software engineering projects.

An AI-generated illustration depicts a robotic software engineer coordinating multiple coding agents and development tasks through holographic displays, representing the autonomous engineering capabilities of Cognition's Devin Bots. AI-generated image via ChatGPT (OpenAI)

Key Takeaways: Cognition's Devin Bots and the Race to Build AI Engineering Teammates

Devin Bots are long-running AI agents that take on an ongoing engineering goal, split it into tasks, and manage teams of Devin coding agents to get it done over days or weeks.

  • Cognition, valued at $48 billion, launched Devin Bots on Tuesday as it competes with OpenAI's Dots and Anthropic's Duet.

  • Cognition has raised more than $4.5 billion to date. Founders Fund led its first round and also invested in its latest, a Series E led by Andreessen Horowitz and Accel.

  • Usage of Devin Bots grew tenfold in two weeks among early-access customers, Fortune reported.

  • Cognition's annualized revenue run rate passed $1 billion in September, and 14,000 Citigroup engineers use Devin.

When Cognition introduced Devin in March 2024, the pitch was an AI that could take a coding assignment and finish it on its own. Two and a half years later, the company wants its AI to stop waiting for assignments.

On Tuesday, Cognition launched Devin Bots, AI agents that act less like a tool and more like a member of an engineering team. Instead of handling one ticket at a time, a Bot is handed a standing responsibility, such as keeping an Android app in sync with its iPhone version, and keeps working on it for days or weeks.

The launch was first reported by Fortune.

How Devin Bots Work

Think of a Devin Bot as a team lead made of software. A person gives it a goal. The Bot breaks that goal into smaller jobs, hands each one to a separate Devin agent running on its own cloud computer, and checks the results before deciding what to tackle next.

Bots keep their memory across sessions, so they don't start from scratch each morning. They join a team's Slack like any coworker, and people can also text or call them. They can run on a schedule, for example, checking every night whether a company's code tests are passing, and they only interrupt a human when they hit a decision that needs one.

Cognition offers ready-made Bots for common roles, including a QA engineer, an Android engineer, a latency optimizer, and a "chief of staff" that coordinates the other Bots.

CEO Scott Wu told Fortune he sees the industry heading toward software that largely builds itself, with people still choosing the destination. "Self-driving software is, I think, the direction that it's going," he said.

What Devin Bots Have Built So Far

Cognition tested Bots on its own products first. With no Android engineers on staff, the company assigned a single Bot to create an Android version of the Devin app. Over 44 days, it merged more than 1,100 code changes and closed more than 1,200 tickets, while the engineer overseeing it sent fewer than 200 messages, according to Cognition. The app is now in open testing on Google Play.

In another test, an employee ran a mobile game studio as its only human worker, supported by about 20 Bots. Weekly code changes jumped from roughly 50 to about 1,900, the company said.

Outside Cognition, early users include Decagon, Fireworks AI, and Reducto. Payments company Bilt told Cognition its Bots merged about 1,600 code changes in a single week.

Keeping AI Agents Under Control

Handing an AI agent a project for weeks raises an obvious question: what stops it from doing something it shouldn't?

Each Bot runs on separate cloud machines, away from passwords and keys stored on engineers' laptops, Wu told Fortune. Company administrators decide which code and tools each Bot can access, and network controls are designed to block data from leaving without permission, even if a Bot is tricked by malicious instructions hidden in something it reads. Teams also choose which changes need human sign-off and which small, low-risk edits can go through automatically.

That focus on guardrails matters. When Devin first debuted, WIRED noted that agents that take action, rather than just generating text, raise the stakes: a single mistake can derail the whole job.

Cognition's Funding and Main Backers

Cognition is one of the most valuable private AI companies outside the major model labs.

Together, those company figures put Cognition's total funding at more than $4.5 billion. Founders Fund, Peter Thiel's firm, led Cognition's first funding round in 2024 and returned as an investor in its September Series E.

The business is growing to match. Cognition's annualized revenue run rate passed $1 billion in September, and Citigroup CEO Jane Fraser said 14,000 of the bank's engineers use Devin, Fortune reported.

Teams of AI Agents Are Spreading Beyond Coding

The idea behind Devin Bots, one AI coordinating a group of specialized agents toward a shared goal, is catching on well beyond software.

In science, Google's AI co-scientist splits research into roles. Some agents propose hypotheses, others critique and rank them, and a supervising agent keeps the work on track. Google says some of the system's drug ideas for a form of leukemia showed promise in lab experiments.

In business, Salesforce made multi-agent orchestration generally available in June, letting a lead agent pass customer requests to specialist agents that handle areas like billing or shipping.

And OpenAI tested its always-on Dots agents internally on procurement, invoice processing, and customer support before launching them in September, with plans to let users run several at once.

That overlap is what makes Cognition's competition so direct.

Competing With Its Own Suppliers

As Fortune noted, Cognition's biggest suppliers are also its competitors. It relies on AI models from labs like OpenAI and Anthropic, and those same labs now sell rival coding agents. OpenAI's Dots works with its Codex coding tool, and Anthropic's Duet works with Claude Code, Fortune reported. The stakes are high: in August, SpaceX bought Anysphere, the maker of the AI coding tool Cursor, in a deal that valued it at $60 billion.

Wu's answer is flexibility. Devin Bots aren't tied to a single lab. They choose different models for different parts of a job, using cheaper ones for routine code, and Cognition also trains its own lower-cost models, he told Fortune. Each Bot can be given its own budget and measured against what it delivers.

His bet is that as AI models get smarter, the product wrapped around them matters more, not less. Developers will need better ways to steer AI, check its work, and keep costs in line.

What This Means: Autonomous Software Engineering

Devin Bots represent a further step toward AI systems that can manage ongoing responsibilities rather than wait for individual assignments. For businesses, that could mean assigning AI to maintain software, identify problems, and carry out routine development work while employees focus on priorities and decisions that require human judgment.

The examples Cognition has shared demonstrate how much work these systems can perform, but the number of completed tasks alone does not establish the quality of the results. Businesses considering this approach will still need to evaluate reliability, costs, security, and how much human oversight is necessary.

Devin Bots are available now through Cognition's website. Wu says the goal is to keep making Devin more independent as AI models improve, so that each new advance from the big labs strengthens Cognition's product instead of threatening it. As the underlying AI models continue to get better, Cognition is betting that businesses will increasingly value systems capable of coordinating their work over extended periods. For software teams, that could mean spending less time managing individual coding tasks and more time guiding the development of new products and technologies.

Sources:

About Natalie Bartels

Natalie Bartels is a contributing writer at AiNews.com covering AI startups, venture capital, and emerging technologies. She is also the founder of Monstera Communication, a communications firm for AI and frontier technology companies.

Over the past decade, Natalie has led communications strategy inside top tech agencies and tier-one venture funds, overseeing more than 100 announcements for companies from seed to IPO.

Natalie has worked with founders across physical AI, AI models, science, and deep tech. Her writing explores how AI companies secure funding and how frontier AI technologies move from research labs to commercial markets.

Connect with Natalie on LinkedIn

Email her at [email protected].

Learn more about Monstera Communications