Factory's Droid Now Runs Engineering Workflows Autonomously Without Human Input
Factory's Droid agent can now run recurring engineering workflows on triggers and schedules, with user-chosen models, machines, and connectors.
- Factory made Custom Automations generally available, letting Droid run recurring workflows on schedules or event triggers.
- Each automation combines a trigger, instructions, a chosen model and reasoning level, and a target machine.
- Templates cover PR babysitting, security audits, Sentry alert response, ticket-to-PR, and Slack morning briefs.
- 30% of code review automation runs on the platform currently use an open-weight model.
- Enterprise automation session volume grew 51% month over month during the private rollout period.
- Available now via Factory; see the docs to get started.
Factory’s Droid can now run engineering workflows on its own
Factory has moved Custom Automations from preview to general availability, allowing its Droid coding agent to run recurring engineering workflows without someone starting each session. Developers can describe a workflow in natural language, connect it to a schedule or event trigger, and let Droid execute it end to end. Factory’s announcement says the feature is available to all users.
Recurring tasks such as following up on pull requests, investigating CI failures, checking security findings, and updating documentation consume engineering time across a repository’s lifecycle. Custom Automations keeps Droid assigned to those loops, turning an interactive coding agent into a background worker.
Four controls shape every run
Each automation combines four independently configured components:
| Component | Options | Purpose |
|---|---|---|
| Trigger | Schedule, Slack, GitHub, or webhook | Determines when the workflow starts. |
| Instructions | Prompt or reusable template | Defines the task, constraints, and expected output. |
| Model | Supported model and reasoning level | Sets the model and the amount of reasoning effort assigned to the run. |
| Computer | A user-controlled machine or Factory-managed Droid Computer | Provides the environment where Droid executes commands and accesses tools. |
Runs can use an individual identity or a service account. Visibility controls and execution history give teams a record of what ran, when it ran, and which identity performed the work. Credential scope and repository permissions determine what each automation can access.
Developers can select a template from the Automations Marketplace or choose Create with Droid and describe the workflow in plain English. Droid requests missing details before creating the automation. The same command works during an active session, so an ad hoc investigation can become a scheduled or event-driven job.
From failing CI to finished pull requests
Factory’s initial templates cover several common repository and incident-response workflows:
- PR Babysitter: Checks open pull requests for failing CI and attempts fixes. An optional CI Failure Triage step classifies failures as flaky tests, infrastructure problems, or regressions.
- Security Audit: Runs weekly through GitHub Actions, checks potential security issues against the codebase, and reports its findings.
- Alert Responder: Starts from a Sentry webhook, investigates the likely root cause, comments on the issue, and opens a pull request with a proposed fix.
- Ticket to PR: Collects ready tickets from Linear or Jira, implements the requested changes, and opens pull requests for review.
- Morning Brief: Sends a private Slack digest containing threads, pull requests, and tickets that require attention.
Model choice becomes a routing decision
Factory lets teams choose a model for each automation instead of applying one model across every workflow. A lower-cost model can prepare daily digests, while a more capable model handles security reviews or complex debugging. Teams can also bring their own provider keys through BYOK or use Factory Router, which selects models within the organization’s configured policies.
Factory reports that open-weight models, whose weights can be hosted or managed outside a closed provider’s service, handle 30% of automated code-review runs among current Automations users. That usage suggests cost, deployment control, and data-residency requirements influence model selection for high-volume review work. The BYOK guide explains how to connect an external provider.
The private rollout produced early usage signals
Factory shared the following company-reported figures from enterprise customers that used Automations during the previous month:
| Metric | Reported result |
|---|---|
| Customers using automated code review | 52% |
| Customers using security review | 48% |
| Customers using AutoWiki | 35% |
| Automation session growth | 51% over the preceding 30-day period |
Factory also reported internal operational results. Droid acknowledged incident threads in a median of 20 seconds and reached resolution in a median of six minutes. It reviewed 99.5% of mergeable pull requests, and engineers resolved 81% of its review comments.
The comment-resolution rate measures whether engineers acted on or closed Droid’s suggestions, making it more useful than review coverage alone. These figures do not establish review accuracy or time saved, and their interpretation depends on workload size, severity, and the criteria Factory used to define resolution.
Background agents need operational controls
GitHub Copilot Workspace, Cognition’s Devin, and Cursor’s background agents reflect a broader shift from assistants driven by editor interactions to agents that accept delegated tasks. Factory distinguishes its approach through per-workflow control over models, compute environments, connectors, identities, and execution records.
Each automation includes an Events feed and a rolling seven-day success rate. Repeated failures appear in digest emails, giving teams a way to detect unreliable workflows and inspect prior runs. Those controls help developers monitor agent behavior before granting broader repository access or allowing automated changes near protected branches.
Availability and setup
Custom Automations are available to all users through the Factory app. Factory provides configuration instructions, trigger details, and model options in its setup guide.
Comments
