GitLab names Transcend Hackathon winners focused on life after code

GitLab has named the winners of its Transcend Hackathon, where developers built agents and flows to handle the work between writing code and running software in production. More than 3,300 developers registered for the October event, with 783 projects submitted.
- More than 3,300 developers registered and 783 projects were submitted.
- The hackathon focused on moving software from code generation to production.
- ShakeFix completed a pharmacy app fix and deployment in 18 minutes and 34 seconds in a live run.
- Lookback found 413 underpaid households in its demonstration.
- The prize pool was $45,000 across nine categories.
The challenge was what comes after code
GitLab framed the hackathon around the idea that code generation is now the easier part of software development. Its challenge, titled “Life after code, the path to production at the speed of imagination”, asked participants to build on GitLab Duo Agent Platform.
The projects had to address tasks such as testing, security, release and monitoring after code was written. Participants also had to decide how much authority to give their agents, ranging from approval at every step to fully autonomous operation.
According to GitLab, the strongest projects shared a common model: agents handled reading, writing and routing, while automated checks and people made the final decisions. The company said the winning entries showed that these systems could be applied beyond conventional software workflows, including pharmacy apps, benefits systems, care homes and classrooms for blind students.
Three projects took the Start Fresh path
The Start Fresh path covered projects built from scratch on GitLab. Aftermath, by Ahadi Cyizere, won Best Assisted Agent and received $4,000. Its GitLab Duo flow uses three subagents to investigate whether a hijacked npm package had run in previous CI jobs, identify credentials those jobs could access and prepare containment steps. A person approves the investigation scope, the containment plan and the incident closure.
Lookback, by Ujwal Suresh Vanjare and Arpita Madhukar Kalburgi, won Best Supervised Agent and received $4,000. Starting from a successful benefits appeal, its flow reproduces a software error, creates a failing test and opens a fix merge request. After a person merges the fix, the system reviews earlier decisions. In the demonstration described by GitLab, it found 413 underpaid households and published the amounts owed as a payment batch.
ShakeFix, by Venkatesh Devale, won Best Hands-Off Agent and received $5,000. A pharmacy app user reports a problem by shaking a phone. Four agents then write a failing test, fix the bug and compare the fix with production behavior before merging and deploying it. GitLab said the live run took 18 minutes and 34 seconds with no human touches.
Existing open source projects added agent automation
The Bring Your Own path focused on existing open source projects. Unstuck, by Aryan Shah and Riken Shah, won Best Assisted Agent and received $4,000. Its flows manage dependency upgrades by reading release notes, fixing broken behavior and pausing for a person when an upgrade would change behavior. Applied to doccano, the project moved from an unsupported Django version with 80 known advisories to zero advisories and Django 5.2 in six reviewed merge requests.
CodeTinder, by Jason Yu, won Best Supervised Agent and received $4,000. On each merge request, an agent selects one function and creates tested variants. Users swipe through those variants on a phone, and their choices update a versioned principles document that agents use when rewriting the rest of the merge request.
Locus, by Ahammad Shawki, Nourin Richi and Mohammad Ashfaqul Hasan, won Best Hands-off Agent and received $5,000. Built on the open source JSCAD library, it turns a teacher's plain-language description into a tactile graph with braille labels for a blind student. Agents check the graph against research-based touch rules, correct errors, publish the result on Google Cloud and send the print files to the teacher. GitLab said a graph cannot be released for printing until it passes those rules.
Special prizes reward coverage, safety and efficiency
GitLab also awarded prizes for projects covering more of the nine DevSecOps lifecycle stages: Plan, Create, Verify, Package, Secure, Release, Configure, Monitor and Govern.
Stale, by Diptarka Nandi, Harshit Anand, Abhinav Kumar and Ashutosh Pandey, won Most Stages Covered in the Start Fresh path and received $5,000. Its scheduled pipeline monitors live feeds, opens issues when conditions change and uses a flow to prepare fixes. CI and a canary release test the changes before traffic moves.
Earned Autonomy, by Shivansh Kalra, won the corresponding Bring Your Own prize and received $5,000. It assigns each merge request on the FastAPI full-stack template an autonomy level based on evidence. GitLab said one run took a vulnerable dependency from advisory to fixed production deployment in 35 minutes without a human step, while another rolled back a bad release in 3.3 seconds.
Memory Lane, by Oliver McBride, won Most Environmentally Impactful and received $5,000. Its flows help care-home coordinators create reminiscence sessions for residents with dementia, taking requests through safety review, CI/CD and Cloud Run. After measuring carbon costs with the Software Carbon Intensity spec, the team cut CI per shipped request by 48 percent.
CodeTinder also won the $4,000 Most Creative prize. GitLab said the project was the highest-scoring idea across both paths. The company said the results support a supervised approach in which agents perform substantial operational work, but tests, CI and people retain control over risky changes.

