Your Marketing Team Doesn't Need a Developer. It Needs an Agent.
One person with no engineering background closed the gap between "the video is edited" and "the video is published, captioned, and reviewed" in about a day. Here is what he built, and why it only worked because the AI agent had real infrastructure to call, not just code to write.
Most stories about AI and video production stop at generation: an agent writes a script, cuts a rough edit, maybe drafts a caption. This one starts after the edit is already done, which is where the actual busywork lives. The person who runs social media at Eyevinn had a workflow problem. Every time he exported a finished clip, everything that made it useful disappeared with the export: what he said, how he said it, what b-roll he used, what hook he opened on. Then he'd write captions by hand for YouTube, TikTok, and Instagram, and hand the marketing team's analytics agents a title and a thumbnail to work with, nothing richer. He's not a developer, and says so directly in the video below. Learning transcription, frame analysis, and multi-platform publishing from scratch would have cost him more time than the tool could ever save, so he didn't learn it. He connected Claude to Open Source Cloud and built the pipeline himself, in about a day.
What he actually built
Drop a finished video into a watch folder on a Mac, and a chain of OSC services does the rest. VideoCore converts the file and serves a review link that plays instantly over HLS, even on a bad connection. Whisper transcribes the audio to a timecoded SRT, used both for the analysis and for auto-generated subtitles. Extracted frames get analyzed alongside the audio, so the agent can describe the hook, the pacing, the structure, the tone, and the delivery, the kind of notes a producer would give. That analysis becomes separate draft copy for each platform, and if a line reads too AI, he tells it what he actually meant and it rewrites. A structured version of the same data lands in a GitHub repo, so the marketing team's own analytics agents finally have more than a title to reason about. Reviewers get a link with timestamped comments, click one and jump straight to that moment in the clip. One button publishes to every connected platform once it's approved. None of that is a video tool. It's a chain of unmodified open source services, wired together by an AI coding agent that never had to build a transcoding pipeline or a transcription service from scratch.
Watch the pipeline in action
The loop that was missing
We've written about this exact gap before, from a different angle. A recent post on this blog described an agent preparing a review copy for an editor, Magnus, using the same open-videocore service: upload, package, deliver an HLS link, no manual console work. That was an editorial review. This is a marketing approval flow. Same service, same mechanism, different department.
The same primitive, a different job
That's the pattern behind My Agent Tasks too: an agent that keeps working against your real repo and your real services after the first build is done. In this case, the clip analysis itself runs as a scheduled agent task with access to the repo and a clear brief on what to look for. It's the same primitive our own engineering agents use for production work. There's no separate consumer tier of this technology and enterprise tier, it's the same tool, pointed at a different job.
Why this only works with unmodified infrastructure
An AI coding agent can write an application fairly easily now. Getting that application to actually run, on real transcoding, real transcription, real storage, without a platform engineer standing in the middle, is where most agentic projects stall. That's the gap Open Source Cloud exists to close: close to 200 unmodified open source services an AI agent can discover and call directly through MCP. Connecting a coding agent to OSC takes one line, claude mcp add --transport http osc https://mcp.osaas.io/mcp. Authenticate once, and your agent can provision and call the same services this pipeline runs on: VideoCore, Whisper, agent tasks, and everything else in the catalog, from databases to live streaming to uptime monitoring.
This isn't a broadcast story
It's tempting to file this under a cool thing for media companies, and that undersells it. The problem here wasn't specific to video. It was specific to any workflow where useful context gets thrown away at a handoff, and where writing the connective tissue between two systems has never been worth an engineer's time. Every team has one of these: an ops team stitching data between a CRM and a spreadsheet, a support team summarizing tickets by hand, a finance team re-keying numbers between two tools that don't talk to each other. The pattern is the same. Someone who understands the problem, usually not an engineer, sits down with an AI agent and infrastructure that stays out of the way, and the gap closes in a day instead of a quarter. He said he's running every video he makes through this pipeline from now on. That's the actual proof point, not that it worked once, but that it became how he works. A companion piece on Eyevinn Technology's Medium publication, "When the Marketing Guy Builds the Pipeline: A Case Study in Agentic TV Operations," covers the same story from the production side.
Our own engineering team runs on this too
This is one proof point among several. Our own engineering team runs an AI dev team that builds and operates OSC itself.
One person, one day
A separate six-agent team built Streaming Tech TV+, a full streaming platform, in 36 hours across 14 services. This pipeline is the same story at a much smaller scale: one person, one day, no engineering background required.
Try it yourself
Connect your agent to OSC and see what a day gets you.
Frequently Asked Questions
Do I need to work in media or broadcast for this to apply to me?
No. The example here happens to be video, because that's the workflow that was broken. The underlying pattern, an AI agent calling real infrastructure instead of a human doing the handoff by hand, applies to any repetitive workflow with a gap in it.
What did OSC actually provide that Claude alone couldn't?
The production-grade pieces: transcoding, transcription, HLS packaging, scheduled agent execution, and about 200 other services, all unmodified open source, reachable over MCP. Claude wrote the application logic. OSC supplied everything underneath it that an AI agent should not have to reinvent.
Does this lock the pipeline into OSC?
No. Every service underneath it is unmodified open source software, speaking the same APIs it would if you ran it yourself. If you moved it elsewhere, nothing about it is OSC-specific.
Can I use a different coding agent instead of Claude?
Yes. OSC's MCP server works with any MCP-capable agent, Claude included. The video above uses Claude because that's the presenter's preference, not a requirement.
Related Posts
Your Agent Doesn't Just Deploy the Backend. It Can Run It.
Most agent-infrastructure stories stop at "the agent deployed it." A real run this week went further: an agent uploaded, packaged, and delivered a review copy, then reported an operational mistake nobody asked it to check. Here is why that only works when the agent is calling real, unmodified open-source APIs.
Give your AI agent a backend it can deploy, and you can walk away with
Your AI coding agent can build the app. OSC gives it a backend to deploy through one MCP connection, built from unmodified open-source services you can export and run yourself, on any cloud, any time.
Add an MCP Server to Your Stack in One Line
Connect your AI coding agent to real infrastructure. Add the Open Source Cloud MCP server in one line and let your agent provision and run open-source services.