GitLab teaches AI agents to work on a schedule: from every 15 minutes to once a month
Until now, a GitLab agent started only when a person launched it or an event occurred, such as a new merge request. In the GitLab 19.6 beta, these agent workflows (flows) can run on a schedule, from every 15 minutes to once a month. For example, an agent can check CI incidents once an hour and write up the probable cause. Each run uses its own service account and appears in the audit log.
AI-processed from GitLab Blog; edited by Hamidun News
Every team has work that repeats every day or week: triaging fresh tickets, figuring out why CI failed, finding forgotten tasks. GitLab's AI agents could already do this, but they started only after a person acted or an event fired, such as the creation of a merge request (a request to merge code). If the person responsible is on vacation, the routine stalls.
If no event occurs, it never starts at all. The GitLab 19.6 beta changes that: agent workflows (flows) in Duo Agent Platform can now run on a schedule, from every 15 minutes to once a month.
In practice, an agent can check open CI incidents every hour and leave a probable cause in the ticket before anyone has even opened it. According to a GitLab survey, 43% of DevSecOps professionals who use AI say that automating repetitive tasks is the biggest source of time savings.
How scheduling works
Duo Agent Platform already had event-driven automations, which fired on a merge request or a mention in a comment, for example. The new scheduled automations feature starts work on a timer, with no event required. You describe the task in plain language and pick the frequency once, and the automation runs on its own from then on. Reports show the number of runs and the success rate.
The interval is configurable from 15 minutes to a month. You can set the time of day and the time zone so that runs fit the team's working rhythm. Each automation operates at the project level, because its output affects the whole project team. The simplest way to think of it is as an alarm clock for the agent: before, you had to wake it manually or wait for something to happen; now it gets up on its own at the right hour.
What to assign to an agent first
GitLab suggests five starter scenarios:
- CI incident triage. Once an hour, the agent checks open incidents and comments with the probable cause, so the investigation is under way before anyone opens the ticket.
- Issue sorting. In the morning, the agent reviews new issues, determines whether each is a bug or a feature, and applies a label.
- Abandoned issues report. Once a week, the agent creates an issue listing tasks with no activity in the last 30 days.
- Duplicate detection. Every day, the agent comments on new issues if it finds similar open requests, so duplicates get linked before triage.
- Friday digest. The agent summarizes in plain language what was closed and merged during the week and publishes it as a separate issue for people who don't read MR titles.
For a product manager, the last scenario may prove the most useful: a status update for stakeholders gets prepared without involving engineers.
Control and transparency
When agents work on their own, you need to be able to answer three questions: who acted, what permissions they had, and what exactly they did. Otherwise, any review or audit begins with reconstructing events. That is why each automation runs under its own service account with limited permissions, and the audit log makes it easy to tell an agent's work from a human's. Changes to the automations themselves are also recorded, so you can see who changed a task and when.
The Triggers page has been redesigned and renamed Automations. It lists schedules and event-driven automations along with the number of runs, the success rate, and session details. If a run fails, the page shows the cause, and you can open the session to see what the agent did.
GitLab positions the feature within its governed software factory concept: a path from idea to production that follows the organization's rules, where every change is backed by a chain of evidence. For example, a weekly vulnerability scan can create work items that the rest of the system then picks up.
Limitations and how to try it
This is a beta, and the article was published by the vendor itself, so it contains no independent data on actual time savings. The scenarios consume GitLab Credits, like the platform's other flows, so running a heavy task frequently should be factored into the budget. A poorly worded task running every hour can also clutter comment threads.
To try it, enable experimental and beta features in the Duo Agent Platform settings. A Premium or Ultimate subscription, which includes credits, will work, as will a free trial.
Practical advice: start with a harmless scenario, such as the abandoned issues report or duplicate detection. It breaks nothing, and the agent's mistakes are easy to spot on the Automations page. If the success rate is acceptable and the comments are useful, you can move on to more critical tasks such as CI incident triage.
- Nov 19, 2026AWS AgentCore Gets Lambda Hooks and Uses 34% Fewer Tokens Than Alternatives
- Nov 19, 2026Open-Source Rust Video Editor FilmCraft Hits 18.5K GitHub Stars and Lets AI Agents Drive the Engine
- Nov 19, 2026Gemini API Can Now Load Tool Schemas on Demand Instead of in Every Prompt
- Nov 19, 2026OpenCode Gets a Mobile App: A Remote Control for the Agent on Your Computer
Key Questions Answered
How scheduling works?
Duo Agent Platform already had event-driven automations, which fired on a merge request or a mention in a comment, for example. The new scheduled automations feature starts work on a timer, with no event required. You describe the task in plain language and pick the frequency once, and the automation runs on its own from then on. Reports show the number of runs and the success rate. The interval is configurable from 15 minutes to a month.
What to assign to an agent first?
GitLab suggests five starter scenarios: CI incident triage. Once an hour, the agent checks open incidents and comments with the probable cause, so the investigation is under way before anyone opens the ticket. Issue sorting. In the morning, the agent reviews new issues, determines whether each is a bug or a feature, and applies a label. Abandoned issues report. Once a week, the agent creates an issue listing tasks with no activity in the last 30 days. Duplicate detection.
What are the key details on Control and transparency?
When agents work on their own, you need to be able to answer three questions: who acted, what permissions they had, and what exactly they did. Otherwise, any review or audit begins with reconstructing events. That is why each automation runs under its own service account with limited permissions, and the audit log makes it easy to tell an agent's work from a human's. Changes to the automations themselves are also recorded, so you can see who changed a task and when.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.