EN
Back to the archive

The archive · Developer & Business Tools · Product decision · 2024–2026

Mosaic bets video editing becomes agent workflows — HN 148 points, $3.8M seed

Two ex-Tesla engineers rebuilt video editing as a node canvas of multimodal AI agents after a chat-first copilot failed; YC W25.

Mosaic

The betThat video editing moves from manual timelines to composed agent workflows on a node canvas, where reusable multimodal agents deliver an 80-90% draft to polish.Live

What the business is

Mosaic is a browser-based agentic video editor where multimodal AI agents run on a node-based canvas — users compose workflows for rough cuts, montages, captions and reframing that process footage and export to its timeline or to Premiere, DaVinci Resolve and Final Cut.

Starting capital$3.8M seed announced 2026-04-06 (round date per PitchBook: 2026-04-07); YC W25 standard investment before it.

How it started

Adish Jain and Kyle Wade were engineers at Tesla when a side project to film Cybertrucks in Palo Alto left them with hours of raw footage to scrub in DaVinci Resolve and Premiere Pro. Frustrated by menus and manual edits, they set out in 2024 to build what they called 'Cursor for Video Editing'. Accepted into YC's Winter 2025 batch, they spent the first month building a video editor with a multimodal chat copilot, then scrapped the chat UX after user testing showed it was a bottleneck for real editing workflows.

What happened

Month two of YC was spent rebuilding around the new paradigm: a node-based canvas where each tile is a configurable video-editing operation, letting users create, branch and run their own multimodal agents in parallel and A/B test variants. The Launch HN post on 2025-11-19 drew 148 points and 134 comments, with the strongest reception from creators juggling lots of raw footage and commenters arguing agents need interfaces, not just APIs. On 2026-04-06 the company announced a $3.8M seed round, saying agencies, platforms and news networks were using its AI video-editing workflows to scale content production.

No ending yet — it is still running.

Background

Mosaic bets that video editing will stop being manual timeline work and become the composition of agent workflows. Its product is a browser canvas where users connect tiles that represent editing operations — rough cuts, montage, captions, reframing — into multimodal AI agents that process footage and return an 80-90% draft, which the user then refines in a timeline editor.

The company began with a side project: two Tesla engineers filmed Cybertrucks in Palo Alto and found the editing took far longer than the shooting. Convinced multimodal AI could do better, Adish Jain and Kyle Wade built 'Cursor for Video Editing', spent their first YC month on a chat-based copilot, then pivoted in month two after users said sequential chat prompts were a bottleneck for content teams with established workflows.

The node-canvas redesign let agents branch, run in parallel and produce multiple variants from the same footage, with exports to Premiere, DaVinci Resolve and Final Cut. When Mosaic launched on Hacker News on 2025-11-19, the post drew 148 points and 134 comments, largely from creators and engineers weighing whether agentic editing can replace scrubbing hours of footage.

On 2026-04-06 the company announced a $3.8M seed round, reporting that global agencies, platforms and news networks were using its workflows to scale production. YC lists Mosaic as an active Winter 2025 company with a five-person San Francisco team.

What has to be true

  • Both founders hit the pain themselves: filming Cybertrucks was fun, scrubbing the footage was not, so the product solved their own workflow first.
  • The chat interface lost to the medium: video editors have parallel, branching workflows, and a sequential prompt stream capped what agents could do.
  • A node canvas made the agents user-programmable — teams could save, branch and reuse editing logic instead of re-prompting every time.
  • Exporting to existing tools removed the switching cost: users got agent speed without abandoning Premiere, DaVinci or Final Cut.

What can be applied

Don't bolt AI onto the old UI: Mosaic threw out its chat copilot after a month of testing because sequential prompts don't fit video — the interaction paradigm had to match the medium.

Aftermath

As of 2026-09-05 YC lists Mosaic as an active Winter 2025 company with a five-person San Francisco team, and the product is live at mosaic.so with free and paid usage. The company's 2026-04-06 announcement of a $3.8M seed round says global agencies, platforms and news networks rely on its AI video-editing workflows, but no revenue, valuation or customer names were disclosed in the sources below.

Sources

spotted an error? The archive wants to know.

Your turn

You just read one. Describe what you are building, and see who is betting on the same thing.

Free account · 3 free questions · no card

Related cases