HeadsUpAI

Cognition Deploys Devin to Automate Regulated Software Engineering at AstraZeneca

· Updated

Cognition, an applied AI lab building autonomous software agents, deployed Devin at Evinova to automate regulatory documentation. The agent produced GxP-compliant (Good Practice regulations) documentation 8x faster than human teams, cutting a 40-hour process to under five hours. This follows Devin's deployment at Mercedes-Benz to automate global engineering.

This deployment shifts agentic AI from experimental help to core enterprise maintenance. At Nubank, engineers used fine-tuning (adapting a model to specific data) to double Devin's completion scores during a monolith refactor. These results suggest that Devin's longer autonomous engineering sessions are stable enough for migrations that previously required thousands of manual hours.

You can now implement similar "playbooks" for recurring tasks like bug triage. Evinova's "AutoFixer" playbook resolved 66% of bugs autonomously within 10 days, while test automation timelines were compressed from quarters to weeks. These capabilities are available through Devin Enterprise, which provides the audit trails required for regulated industries.

View the full update on devin.ai
Cognition
Cognition
@cognition
X

Evinova, a separate health tech company within the AstraZeneca group, builds software that modern clinical trials run on. They put Devin to work on regulatory documentation, bug triage, tech stack migrations, and test automation. One striking result: Devin produces documentation for regulators ~8x faster than the 35-40 hours it used to take across teams.

3retweets34likes
View on X

Still wondering? A few quick answers below.

Devin is an autonomous AI software engineer developed by Cognition. Unlike standard coding assistants that offer autocomplete suggestions, Devin functions as a teammate capable of planning and executing complex engineering tasks end-to-end. It can navigate codebases, manage Jira tickets, write documentation, and perform multi-step migrations independently while keeping a human in the loop for final review.

At Evinova, Devin was used to generate User Requirement Specifications and Disaster Recovery Plans directly from source code and Jira tickets. It produced structured drafts with roughly 90% accuracy, reducing the time required for senior engineers and QA teams from 40 hours down to under five hours. This 8x efficiency gain allowed senior staff to focus on high-level regulatory judgment.

AutoFixer is a custom Devin playbook designed for automated bug triage and resolution. It runs multiple times per day to read open Jira tickets, explore the codebase, implement fixes, and run tests. At Evinova, this system attempted 79 bugs in 22 days, successfully merging review-ready fixes for half of them and saving several days of manual engineering effort.

Devin is designed to meet the rigorous standards of GxP-regulated environments, such as those at AstraZeneca's Evinova. The agent maintains a full audit trail for every line of code and change it makes, ensuring that all software releases remain defensible to regulators. This traceability is critical for companies operating under 21 CFR Part 11 and other life sciences requirements.

Nubank used examples of manual migrations to fine-tune Devin for a massive ETL monolith refactor. This process doubled Devin's task completion scores and improved execution speed by 4x, reducing sub-task time from 40 minutes to just 10. By teaching the agent specific architectural patterns, Nubank achieved a 12x efficiency improvement and over 20x cost savings compared to manual engineering.

Share this update