01What the Devin AI agent update is and how it works
The Devin AI agent update turns an AI coding tool into something that can test its own work, check for security holes, and send you video proof — all before a human looks at it. It doesn't just write code anymore. It understands how your whole app works.
Devin traces how a real person would move through your project and asks: 'Could someone break in here?' Then it runs tests itself and films the results.
In this episode, Julian Goldie shows how Devin caught a dangerous fake package inside a popular code tool just 45 minutes after it went live on March 31st, 2026. Most security teams hadn't even noticed it yet. Devin spotted the bad dependency, flagged it, and told teams what was wrong.
This speed and care can now protect your own business systems, not just giant companies.
02Main features of the Devin AI agent update
The update has three major changes. Each one makes Devin more useful for real projects. Here is what is new:
- Security scanning: Devin reads your whole project and understands how it works. It does not just look for known bad patterns like a spell checker. It finds broken logic, like a password reset page that anyone could call without logging in.
- Self-testing: After Devin writes code, it creates a test plan based on the actual code, not guesses. You approve the plan. Then Devin runs every test by itself — opening the app, clicking buttons, filling in forms like a real user.
- Video proof: Devin sends you an edited recording. You watch the buttons get pressed, the pages load, and what passed and failed is marked on screen. You see the proof with your own eyes in a couple of minutes.
03How to get started with the Devin AI agent update
Getting Devin set up to test your own work is straightforward. Julian Goldie describes the process inside the AI Profit Boardroom. Here are the general steps:
- Give Devin access to your code repository or project files.
- Tell Devin what you want it to check — for example, a sign-up page or an order form.
- Devin reads your whole project and understands how the app works.
- Devin runs a security check, tracing how a real person would move through the app.
- Devin writes a test plan based on the actual code and shows it to you for approval.
- You approve the plan. Devin then runs every single test by itself — clicking buttons, filling forms, checking results.
- Devin sends you an edited video showing what passed and what failed, with clear markers on screen.
04How to use the Devin AI agent update with real examples
Julian Goldie shows a real example from Devin's own team. It is a page where people change their password. A normal scanner checks basic things like input safety and protection from common attacks.
It says 'All clear.' But Devin dug deeper. It found something the scanner missed completely. The password reset page could be called by anyone without logging in.
That means a stranger could reset another person's password and take over their whole account.
Here is how you could use this for your own work. Say you built a simple sign-up page for your email list. You could ask Devin: 'Can anyone get into accounts they shouldn't?' Devin would trace the logic, find weak spots, and show you in a video.
That one check could save you from a security nightmare. Julian says the same speed and care Devin uses for a giant company can protect your own business systems.
05Results and tests shown in the episode
Julian Goldie shares two important results from using the Devin AI agent update. First, Devin caught a fake package hiding inside a popular code tool just 45 minutes after it was released. It spotted the bad dependency and flagged it before most of the world knew it existed.
Second, Devin found a memory problem in an app and fixed the real issue. But Devin also fixed a second thing that was not actually broken. It made a new problem trying to solve one that didn't exist.
The Cognition team watches their people run 10 to 20 Devins at once, each working on something different and each testing itself. Julian points out that one engineer can now do the output of a small team. You simply cannot do that on one laptop with one human.
The gap between people using these tools and people who aren't is getting wider every month.
06Tips, limits and mistakes to avoid with Devin AI agent
Julian Goldie is clear: Devin is not perfect. In one real test, it fixed a memory problem correctly but then also fixed a second issue that wasn't broken. It created a new bug while trying to solve one that did not exist.
The team behind Devin says the same thing: a human should still check the work before it goes live.
Treat Devin like a smart junior helper, not a replacement for your judgment. Before the update, Devin used to wander off during testing. It would test random stuff that didn't matter or get lost before reaching the thing it supposed to check.
The fix was to make Devin write out its plan first, based on actual code not guesses. This keeps it focused. You should always inspect the test plan before approving it.
Don't skip that review step.
07Who the Devin AI agent update is for and why it matters
This update matters if you build things, even simple things. You can now let an AI agent test your work and show you proof. It also matters if you run a business.
You can hand off boring, repeating jobs to run in the background. Julian Goldie says the bottleneck used to be: can AI write code? That is solved.
The new question is: can humans check all this AI work fast enough? Devin helps with the checking, not just the building.
If you manage people, start thinking about how your team's job changes from 'do the work' to 'check the work and decide what matters.' The people who learn this now will be miles ahead of people who wait. One person can run 10 to 20 Devins at once. That is the shift: one person, way more output.
Devin is not trying to be a fancier autocomplete. It is trying to own a whole job from start to finish with proof at the end.
