SIAgentNews.com
← All news

Cognition launches Devin Bots to manage persistent software projects

The startup is moving beyond single-task coding to offer persistent agents that manage complex software development over weeks.

· First reported by Fortune

$1 billionannualized revenue run rate in September

The short version

  • Devin Bots act as managers that assign work to multiple Devin agents to complete complex projects over days or weeks.
  • The tool integrates with Slack and iMessage, allowing multiple human colleagues to collaborate with a single Bot.
  • Bots can select different AI models based on the task, using cheaper ones for routine code and more capable ones for planning.
  • Cognition reported that usage across early-access customers grew tenfold in just two weeks.
  • Citigroup reported a 30% year-over-year reduction in tech vendor spend after using Devin for in-house development.

Why it matters

This marks a shift from AI that simply suggests code to AI that manages entire workflows autonomously. For companies, it means the potential to reduce vendor costs by moving more development in-house using persistent, multi-step agents.

From code snippets to project managers

Cognition is moving its AI coding tool, Devin, into a more persistent role with the launch of Devin Bots. While the original Devin was marketed as the first AI software engineer, the new Bots are designed to act as managers for teams of agents. These bots can oversee complex projects that span several days or even weeks of work.

This move comes as major players like OpenAI and Anthropic release their own proactive agents, such as Dots and Duet. However, Cognition is distinguishing its product by building it specifically for software development rather than as a general-purpose tool. The goal is to create a system where a human can give a high-level command and the Bot handles the execution.

For example, a user could tell a Bot to build an Android version of an existing iPhone app. The Bot would then divide the work, assign tasks to various Devin agents, and ensure the Android version stays updated as the original iPhone version changes. CEO Scott Wu describes this shift as the direction toward "self-driving software."

How the Bots coordinate work

Devin Bots are designed to function like human colleagues within a company's existing communication infrastructure. They can join a team's Slack channel as members and remember context across weeks of conversation. They are also capable of contacting human team members via iMessage if they encounter a block or require a specific decision.

The system incorporates technology from the Interaction Company, a startup Cognition acquired in July. This allows the Bots to be proactive rather than just reactive. Instead of waiting for a prompt for every single line of code, the Bot works through a project board, looks at diagrams, and provides updates to the human team.

To manage costs, the Bots are not tied to a single AI model from one laboratory. They can select different models based on the specific needs of a task, such as planning, design, or routine coding. This flexibility allows the system to use faster, cheaper models for simple tasks while reserving more powerful ones for complex engineering problems.

Security and human oversight

Because these Bots have the ability to perform autonomous work, Cognition has implemented specific safeguards to protect company data. The Bots run on separate cloud computers, which keeps them isolated from the passwords and keys stored on an engineer's local laptop. Administrators can set strict rules on which code repositories and tools each specific Bot is allowed to access.

The company uses network controls to prevent unauthorized data transfers, which is especially important if a model encounters malicious instructions in the content it reads. Furthermore, the Bots run their code through the same testing protocols used for human-written code to ensure stability and safety before deployment.

Teams can also configure which types of changes require a human to sign off on them. For instance, low-risk edits like changing a button's color can be set for automatic approval and merging. This allows the Bot to handle the

A new way to communicate with AI

Cognition's approach represents a shift in how humans interact with software tools. In the past, tools like GitHub Copilot functioned by suggesting specific lines of code that a programmer could choose to accept or ignore. The new Devin Bots, however, require a different style of interaction because they handle much more complex, multi-step tasks.

CEO Scott Wu believes that as AI becomes more capable, humans will actually need more ways to communicate with it rather than fewer. He compares the interaction to how colleagues work together in an office, using whiteboards and scratch paper to discuss requirements and look at diagrams. The goal is to create a collaborative environment where engineers can direct the Bot on how to use project boards or when to send specific updates.

Business models and future growth

To make these tools viable for large-scale enterprise use, Cognition is introducing specific financial controls for the Bots. Each individual Bot can be assigned its own dedicated budget, which is measured against the actual results it delivers for a project. This allows companies to track the cost-effectiveness of autonomous work as it scales across different departments.

This strategy is designed to ensure that Cognition benefits from the progress of the broader AI industry. Wu wants the company to be in a position where they are happy when frontier capabilities continue to improve, rather than feeling threatened by them. By building a product layer that manages these models, Cognition aims to remain a necessary partner as the underlying technology becomes more powerful.

What to watch

  • How other model providers like OpenAI and Anthropic respond to Cognition's specialized software engineering focus.
  • Whether the 30% reduction in vendor spend reported by Citigroup becomes a standard metric for AI adoption in banking.

Try it

Check if your company's engineering team has a strategy for using 'agentic' workflows to manage long-term projects rather than just one-off code generation.

Sources and links

First reported by Fortune. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news