Visual Studio Code, Microsoft's widely used code editor, has gained a feature in version 1.137 that sounds unspectacular at first glance yet changes quite a lot in day-to-day project work: tasks handled by an AI agent in the editor can now be started on a fixed schedule. According to the announcement, hourly or daily runs are possible, for example.
Until now, the same principle applied to such agents as to any other interaction with an assistance system: someone has to give the impulse. A prompt, a click, a command. With scheduled Agent Tasks, this impulse is no longer needed. The task is stored in advance and runs when the time is reached.
What an Agent Task is
In this context, an agent is more than an autocomplete. It works in multiple steps: it reads files, executes commands, checks results and makes changes. An Agent Task bundles such a multi-step task in reusable form — comparable to a stored work order that always runs the same way.
The new feature consists of linking this order to a schedule. This brings the editor closer to the model known from the automation of build and test processes: something happens because a time is reached, not because someone remembered it.
What this is suitable for
The announcement does not name a fixed list of use cases. However, the obvious candidates are precisely those activities that regularly arise in projects and regularly get left undone:
- an initial review of open changes before a human takes over the review
- checking dependencies for new versions and known security issues
- recurring test runs and summarising error messages
- detecting inconsistencies in documentation, configuration or translation files
What these tasks have in common is that they require little creativity and a lot of regularity. They are rarely urgent until they suddenly are. That is exactly why they disappear in day-to-day business — and exactly why they benefit from a schedule.
Why this becomes visible in the maintenance budget
Maintenance contracts contain a considerable share of routine. Someone checks whether packages are outdated. Someone reads a changelog. Someone starts a test run and interprets the result. This work is necessary, but it is rarely the reason why a company commissions a digital project.
If part of this preparatory work happens outside working hours and is available in the morning as a prepared finding, the ratio shifts. The team does not begin with the search but with the decision. For clients, this means shorter maintenance cycles and more room for features that actually work on the business model — in the best case.
Realistically, however, one thing remains: an agent delivers suggestions and findings, not certainty. The professional assessment of whether an update may be deployed, whether a test failure describes a real problem or whether a proposed change fits the architecture remains with the team.
What must be clarified beforehand
Automation that starts without a human trigger needs clear boundaries. Three points must be clarified before introduction.
- Permissions and scope. What may the agent read, what may it execute, what may it change? A run that collects findings is something different from a run that submits changes.
- Traceability. Every run should be documented: what was checked, what was suggested, what did the human make of it? Without a log, automation emerges that no one can explain anymore.
- Responsibility for the results. A schedule reliably produces output. If no one reads this output, a second queue emerges instead of relief.
The last point is the most common stumbling block. Automated reports that run into nothing are no progress compared with tasks that no one completes — they merely create the additional feeling that something has happened.
Starting sensibly
A narrow start is advisable: a single recurring task, a clearly defined time, pure collection of findings without write access. After a few weeks, it can be assessed whether the results are reliable and whether they are really used in the team. Only then is it worth expanding the scope or taking on further tasks.
For web and software projects, the new feature is less a technical sensation than an organisational invitation. It shifts routine work to where it can be done without attention, and thus makes visible how much maintenance budget has so far flowed into searching and checking instead of further development. However, only those who define beforehand who reads the results, decides and takes responsibility will draw benefit from it — the technology schedules appointments, not responsibilities.
Sources
- Visual Studio Code 1.137: Entwickler können Agent Tasks automatisieren — heise developer News