Code generation is the easy part now. The hard part is everything that comes after it: getting software from a working change to running in production. That’s the problem we put to the community in the Transcend Hackathon: “Life after code, the path to production at the speed of imagination”.
The hackathon, which ran during October, challenged participants to build agents and flows on GitLab Duo Agent Platform that can take over the work after the code is written. They also had to decide how much control to hand the agents, from approving every step to making them completely autonomous. More than 3,300 developers registered and 783 projects were submitted.
The winning projects show how far agents and flows can go after code generation. A pharmacy app bug reported simply by shaking a phone was fixed and shipped in 18 minutes with no human in the middle. A teacher's email became a 3D-printable math graph a blind student could read by touch. A flow applied to a benefits software app helped detect payment miscalculations based on a won appeal. What these projects share: Agents do the reading, writing, and routing, while automated checks and people make the final calls.
The winners
The prize pool was $45,000 across nine categories, with each project eligible for one path prize and one special prize. You can find the full field in the project gallery.
Path A: Start Fresh
New projects built from scratch on GitLab
Best Assisted Agent ($4,000): Aftermath
Aftermath by Ahadi Cyizere answers the question scanners can't: When a hijacked npm package is disclosed, had it already run in last week's CI? A custom GitLab Duo flow with three subagents traces past CI jobs and the credentials each job could reach. A person signs off on the scope of the investigation, then on the containment plan, and finally closes the incident once CI has rotated only the exposed keys and deleted the affected images, and proof is attached.
Best Supervised Agent ($4,000): Lookback
Lookback by Ujwal Suresh Vanjare and Arpita Madhukar Kalburgi starts from a won appeal. When benefits software gets the rules wrong for one person, a custom GitLab Duo flow triages the decision, reproduces the error with a failing test, and opens a fix merge request. After a person merges it, plain code replays past decisions to find every other household the bug affected and works out what each is owed. In the demo run, the replay found 413 underpaid households, and the amounts owed were published as a batch for the agency to pay.
Best Hands-Off Agent ($5,000): ShakeFix
ShakeFix by Venkatesh Devale turns a phone shake into a production fix. A user of a pharmacy app shakes her phone and says what went wrong, and a GitLab Duo flow with four agents writes a failing test, fixes the bug, and checks the fix against production behavior side by side before it merges and deploys. In the team's live run, that process took just 18 minutes and 34 seconds with no human touches.
Path B: Bring Your Own
Existing open source projects with agent-driven automation added and deployed on GitLab
Best Assisted Agent ($4,000): Unstuck
Unstuck by Aryan Shah and Riken Shah finishes the dependency upgrades that bots start but can't complete. A bot bumps a version number, but it can't read the release notes, fix what breaks, or decide what happens when behavior changes. Two custom GitLab Duo flows plan an upgrade campaign and carry it out one merge request at a time, quoting the release note behind every code change and stopping to ask a person when behavior would change. A daily monitor opens the next campaign when a new advisory appears. On doccano, an open source annotation tool stuck on an unsupported Django version with 80 known advisories, the campaign took the project to zero advisories and Django 5.2 in six reviewed merge requests.
Best Supervised Agent ($4,000): CodeTinder
CodeTinder by Jason Yu is built on a simple idea: review less code, teach agents more. On each merge request, a GitLab Duo agent picks one function and writes variants that pass the tests, and users swipe through the variants on their phone. The swipes update a versioned principles doc in the repo, which every agent reads, and an agent rewrites the rest of the merge request to match the updated doc.
Best Hands-off Agent ($5,000): Locus
Locus by Ahammad Shawki, Nourin Richi, and Mohammad Ashfaqul Hasan, built on the open source JSCAD library, lets a math teacher email a function in plain words and get back a tactile graph for a blind student. GitLab Duo agents design a raised-line graph with braille labels, check it against research-based touch rules for how fingertips read raised lines and braille, correct their own mistakes, publish the result on Google Cloud, send the print files to the teacher, and order the print graph. The agents can't mark their own work as ready to print; a graph is released only when it passes the touch rules.
Special prizes
The Most Stages Covered prize goes to the projects that touch the most of the nine DevSecOps lifecycle stages: Plan, Create, Verify, Package, Secure, Release, Configure, Monitor, and Govern. There is one winner from each path.
Most Stages Covered, Start Fresh ($5,000): Stale
Stale by Diptarka Nandi, Harshit Anand, Abhinav Kumar, and Ashutosh Pandey re-checks "safe to ship" verdicts when the world changes, not just when code does. A scheduled pipeline watches live feeds, such as a root certificate being distrusted, and opens an issue for a GitLab Duo flow that writes the fix. CI proves the fix, and a canary release probes it before traffic moves. It maps every lifecycle stage to a real pipeline job.
Most Stages Covered, Bring Your Own ($5,000): Earned Autonomy
Earned Autonomy by Shivansh Kalra gives every merge request on the FastAPI full-stack template its own level of autonomy, from hands-off to blocked, based on evidence. Five GitLab Duo agents handle planning, triage, infrastructure review, routing, and rollback, and can only ever raise the risk score. In one run, a vulnerable dependency went from advisory to a fixed production deploy in 35 minutes with no human step. In another, a bad release was rolled back in 3.3 seconds.
Most Environmentally Impactful ($5,000): Memory Lane
Memory Lane by Oliver McBride helps care-home coordinators build reminiscence sessions for residents with dementia. A coordinator files a plain-English issue, and three GitLab Duo flows carry it through safety review, CI/CD, and Cloud Run to production. The team measured the carbon cost of each shipped request using the Software Carbon Intensity spec, then re-engineered the pipeline to build once and skip needless runs. That cut CI-per-shipped requests by 48%.
Most Creative ($4,000): CodeTinder
CodeTinder wins in this category, too. Swiping to teach agents your taste in code was the highest-scoring idea across both paths.
What we learned
The winners kept agents on a short leash and still got a lot out of them. Agents handled the reading, writing, and routing: working out what an upgrade would break, tracing a bug to the line of code behind it, and handing the work to the next step. Automated checks like tests and CI decide whether each step passes, and, in most projects, a person signs off before anything risky ships. The winners also aimed their agents at problems outside typical software work, including pharmacy apps, benefits systems, care homes, and blind students’ classrooms.
Keep building
To everyone who entered, thank you. Thank you also to our judges: Bill Staples, Missy Davies, Mattias Michaux, Lee Tickett, Rajesh Agadi, Luca Chun Lun Lit, Alan Zhang, Jackson Dolphin, and Arianna Haradon.
If you'd like to experiment with GitLab Duo Agent Platform and contribute to GitLab itself, sign up at contributors.gitlab.com to get a personalized onboarding issue and an issue finder to help you pick your first contribution.




