What happened
Cognition announced Devin Apps on November 13, 2026. The company describes them as live, private web applications hosted directly on Devin’s cloud computer. A user opens Apps in the Devin sidebar, describes what the application should do, and selects “Build with Devin.” They can also begin from prompt templates for a metrics dashboard, project status page or review queue. Devin starts a session, asks any questions in its first message, builds the app and puts it online. An app can also be requested during another session; if Devin has already produced a browser preview, the user can ask it to publish that preview as an App.
Each app appears on the Apps page with a stable URL. Everyone who opens one must sign in to Devin; apps are never public. The owner can use Share to choose who can open it: named people, the organization or the whole enterprise. Users can pin apps, and organization administrators can pin one for everyone. Cognition says Devin Apps can use the team’s existing MCPs, integrations and codebase, and can be built with the team’s own internal stack and components or a built-in template.
To change an app, a user opens it, chooses Edit in session and tells Devin what to change. Devin builds and publishes a new version without changing the URL. Versions appear under Deployments. If a new version fails to start, the previous version keeps serving. The company also describes automation for refreshing data on a schedule or redeploying an app when a branch changes.
Apps run on Devin machines built on memory snapshots, which let Devin hibernate between tasks. The machine sleeps when nobody is using the app and wakes when someone opens it; Cognition says the first load after sleep can take a few seconds. Billing is based on how long the machine is awake. Devin says an app can use any stack that serves HTTP on a port, use a database such as SQLite or Postgres on the same machine, and call the Devin API. Each app runs on one machine. Cognition positions the feature for internal tools and dashboards, not high-traffic production services. Examples include admin panels, request forms, approval queues, dashboards connected to team data sources, and beta or staging copies with frontend, backend, database and workers.
What the archive adds
New Runtime’s earlier coverage described Devin integrations for turning support tickets into coding sessions, an AI SRE workflow for incident response, and spend controls tied to estimated output. The Pylon integration (October 18, 2026) connected customer support work to engineering tasks; Devin Oncall (October 21, 2026) applied the coding-agent workflow to incidents; and efficiency-based auto-approvals (October 22, 2026) addressed spend governance. Devin Apps is a new event in that broader product trajectory: Devin now builds and hosts small team-facing interfaces as well as helping with operational work.
Trend signal
This supports the shift toward agent-built interfaces, with an important constraint: Devin Apps are private, stable-URL applications assembled from a team’s stack, but Cognition explicitly limits the fit to internal tools rather than high-traffic production services. The announcement is relevant to the assumption that stable hand-built applications and forms will remain necessary in regulated, accessibility-critical and audit-heavy workflows; generated interfaces may complement those systems without replacing them. New Runtime’s assumption on generated interfaces
Why now
Cognition’s announcement does not explain the timing. It presents Apps as a newly available workflow and describes common team needs—small internal tools, dashboards and beta or staging copies—without connecting the launch to a specific market event or technical milestone.
Why it matters
For teams already using Devin, the change is a shorter path from a request to a working interface: the agent can use existing integrations and code, build an app, host it, and keep its URL stable as it changes. That can make a small request form, approval queue or project dashboard practical without creating a separate deployment pipeline for every internal tool.
The deployment boundary matters as much as the convenience. A single sleeping machine may suit low-traffic team utilities, and version retention provides a basic fallback when an update fails. It does not establish that these apps meet production expectations for availability, scaling, auditability or regulated data. Teams should assess access, data handling and operational requirements before moving important workflows onto the hosted machine, and retain their existing production platform for workloads that need high traffic or stronger guarantees.
Limits and uncertainty
Cognition’s announcement gives no pricing figures, machine specifications, availability guarantees or quantified performance results. It says billing depends on awake time and that a sleeping machine may take a few seconds to serve its first load, but does not quantify costs or latency across workloads. The company does not describe backup, disaster recovery, data residency, compliance certifications or the details of app-level logging and audit controls. The feature’s stated single-machine limit also leaves capacity and concurrent-user limits unclear. These are open deployment questions, not evidence that the service lacks such controls.
Skill to build now
Pick one low-risk spreadsheet-driven workflow, such as a request form or approval queue, and write down its data sources, who should have access, and what a successful update looks like. Use that specification to test an Devin App with non-sensitive data. Check the stable URL, sign-in and sharing settings, behavior after the machine sleeps, and whether the previous version remains available after a failed update before considering a real team workflow.
Reader takeaway
Devin Apps turns natural-language requests into hosted internal web tools with stable URLs and controlled sharing. It is a concrete option for lightweight team interfaces, but Cognition’s own boundary is clear: one machine per app, suited to internal use rather than high-traffic production services.