Films
DevBench, in motion.
Short films made in code, and the real app recorded against a live server. Download any of them and share it.
The film
Thirty seconds on why DevBench exists, with sound.
Agents write the code now: code streams into windows. Coordinating it is the mess: questions pile up (which branch, did the tests run, who merged that). One session, the whole team in it: a person asks the agent for a rate limit in the workspace chat, the agent answers, and four checks land: tests, a preview, a video and a merge request. What you get: a live preview for every branch, a video of it working, and a merge that stays human. A person presses Merge. The DevBench mark builds, with the line Try it in a preview. Merge what you keep.
Showreel
The app taken apart and cut to music: direct, track, build, see, merge.
To a 120 bpm beat, one part of DevBench per bar: Direct, the workspace chat, whose messages break out across the frame; Track, the board, where a task crosses to Done and the cards fan out; Build, the agent's terminal running tests, pushing, recording and opening the merge request; See, the preview being recorded while a rate-limit message appears; Merge, the diff and its four checks, and a person pressing Merge. It ends on the DevBench wordmark.
Why the merge button is human-only
An agent can do almost everything. Except one thing.
An agent writes the code, runs the tests, puts up a preview, records a video and opens the merge request. Then its cursor reaches the Merge button and is refused: merging is for a person. A diagram shows the two refusals: the agent can push its own branch, but a push to main is stopped by the box's git, and a push that gets past it is refused again by the server. Every commit is authored as the person who asked, co-authored by DevBench, and merged by a person. An agent does the work; a person stays in charge.
Try it in a preview
One idea from New workspace to merge, then hundreds like it. Ninety seconds; press Sound on to hear it.
You have an idea: Ava names a workspace, Settings as tabs, says what she wants and picks three repositories. It gets a sandbox of its own: its own container and branch, the three repositories in it. An agent reads it and comes back with a plan: the screens, how the web app, the API, the database and the worker fit together, and the steps; nothing is built yet. Ava opens the plan, then approves in the chat. You approve, it builds: tabs in the web app, endpoints in the API, and the tests, which pass. It runs, live, before anything is merged: a seeded database, the API, the web app and the worker, at one URL; Ava clicks through Profile, Security and Billing. You like it, you merge it: only a person can press Merge, and she does. Every idea gets a sandbox like this, isolated from the others and from main: the window shrinks into its box and the camera pulls back to ninety boxes. Hundreds of agents, thousands of experiments, all at once, on your own servers. Only what you merge ships: everything else is dropped, and main never sees it. The merged boxes drop onto a bench as the blocks of the DevBench mark: devbench.sh. Try it in a preview. Merge what you keep. Music and interface sounds play under it.
Twenty loud seconds from inside the box. Made for a phone.
Point of view: you are the agent in a DevBench box. Requests pile up; the tests fail, then pass; it records itself, puts up a preview and opens merge request 412. It reaches for Merge and the button goes grey: not your button. A person merges instead. Best job I've had.
Made for a phone.
Twenty seconds from inside the agent's box, upright, for Reels, Shorts and TikTok. The explainer on the merge button also comes square, for a feed.
The real app, recorded.
Filmed against a live DevBench server, on the workspace that built this site. Nothing in them is drawn.
A workspace and its finish line
A workspace's overview: the finish line the agent works to, the plan, the repository.
The overview of a real DevBench workspace. At the top is its finish line, the condition the agent keeps working to until it is true, then the requirements and the repository with its branch. The pointer moves to the Tasks tab, which opens the workspace's board.
The workspace chat
The team's requests and the agent's reports, in one thread.
The chat of a real DevBench workspace, scrolled through: a person's request, the agent's progress notes, a report of what shipped with a list of changes, and the finish line pinned above the composer.
The board
Every request as a task, done with a note on how.
The Tasks tab of a real workspace: 127 tasks, 118 done, each with its group, its notes and its review state. A group is opened to show the tasks the agent made for one request, every one ticked.
Merge requests
What the agent opened, and each one merged by a person.
The Changes tab of a real workspace: its branch, how far it is behind main, and its merge requests, each named for the release it carried and marked merged.
Previews
The branch deployed as live containers, step by step.
The Previews tab of a real workspace: a deployment of the branch's backend and web app, each with its steps (clone, install, build, start, health check) and their timings, then the environments the workspace's work flows through.
How these were made
Each film began as a prompt on prompt-motion.com, credited under it. It was written as code, so every frame is rendered the same way each time, and its sound is synthesised from the same timeline. A separate reviewer agent then watched each film and asked for changes: pacing, legibility on a phone, beat timing, colour and loudness. Those changes were made before anything was published. The recordings are the real web app, filmed through the browser against a DevBench server. Email and IP addresses are blanked as they render, and other workspaces' names are hidden.
Invite only
We set your organization up with you.
Tell us about your team and the repositories you would bring; we reply within a working day.