Devin for MongoDB Modernizations: AI Agents Take On Migrations

What was announced

MongoDB and Cognition, the company behind the AI software engineer Devin, have launched a joint product called Devin for MongoDB Modernizations. MongoDB announced it on September 29, 2026. 1 The offering connects Devin directly to MongoDB's Application Modernization Platform (AMP). Its purpose is to help enterprises move off legacy infrastructure and onto MongoDB Atlas, the company's multi-cloud data platform. 12

The work is divided between the two systems. Devin plans the migration and rewrites business logic and data access layers. AMP's tooling moves the data into Atlas and validates it. 23 The companies say the two components coordinate the whole process, which cuts down on the manual handoffs that usually slow migrations. 23 Investing.com's write-up notes that AMP covers four phases of a legacy migration, beginning with understanding the existing application. 3

The claims, and how far they go

The main performance figure comes from the companies' own early joint testing. Tasks that used to take five to six hours were finished in just over an hour. 2 MongoDB CEO Dev Ittycheria framed the benefit around scale. In his account, building Devin into AMP lets customers modernize more of their application estate at once, so legacy apps can move to MongoDB "in months rather than years." 2 Cognition co-founder and CEO Scott Wu described Devin as rewriting code across an entire codebase. 2

The one customer voice in the announcement is Kosta Krauth, CTO at Bilt. He said his team used Devin to rebuild search and personalization features on MongoDB, and that during the final push engineers averaged dozens of merged pull requests per day. 3

These figures deserve some caution. The time savings come from vendor-run testing, and the single customer example describes one rebuild rather than a large legacy migration. The coverage at StreetInsider and Investing.com relies on the press release and offers no independent benchmarks. 23 A drop from roughly six hours to one on discrete tasks is plausible for an agent rewriting data-access code. Whether that holds across a full enterprise migration, with its edge cases, undocumented business rules and validation work, has not been shown yet.

Part of a bigger MongoDB push

The partnership was one of several announcements at MongoDB.local NYC, the company's developer conference. MSSP Alert, drawing on CRN's reporting, lists the others: 5

• MongoDB 9.0, a core engine upgrade promising up to 35% faster query performance plus security enhancements.
• MongoDB Atlas Infinite, a deployment option that separates compute from storage for more elasticity, aimed especially at AI workloads.
• Atlas Agent Engine, a unified layer for running AI agents in production.

Atlas Agent Engine targets a familiar enterprise problem. Agent prototypes are quick to build, but putting them into production means stitching together separate tools for retrieval, memory, identity, audit trails and policy controls. 4 The engine combines execution, memory and governance on top of Atlas. Customers can use the memory and governance pieces on their own and keep their existing models and frameworks. 4 It is in public preview, with consumption-based pricing for its runtime and memory components. 4

Why it matters

Taken together, the announcements show MongoDB working both ends of the AI pipeline. On the inbound side, it is trying to cut the cost of leaving legacy systems. On the outbound side, it is positioning Atlas as where production AI agents run. The press release states the link directly: it calls modernization "a necessary first step" for enterprises building AI applications and describes legacy infrastructure as expensive to maintain and hard to scale. 1 By that reasoning, every legacy app Devin moves onto Atlas becomes a potential future customer for Agent Engine and Atlas Infinite.

For Cognition, the deal places Devin in a clearly defined, high-value job. Migration work is repetitive, follows recognizable patterns, and is unpopular with engineers. That makes it a better fit for an autonomous coding agent than open-ended feature development. Pairing the agent with a platform that handles data movement and validation also addresses a weakness of code-only tools, which can produce plausible rewrites but cannot confirm that the data arrived intact.

The takeaway

In my reading, this is less a story about one partnership than a sign of where AI coding agents are finding commercial footing first. They are being attached to narrowly scoped, verifiable workflows that a platform vendor has a direct financial interest in speeding up. The benchmarks so far come from the vendors. The real test will be whether customers report migrations finishing in months on large, messy estates. For now, the strategic logic is clearer than the evidence: MongoDB wants migration to Atlas to feel nearly automatic, and it has hired an AI engineer to help make that case.