What we mean by Agentic TV Operations
Agentic engineering changed how software gets written. An engineer describes the outcome, an agent does the assembly, and the engineer stays accountable for the result. The same shift is reaching broadcast and streaming operations.
Our version of it is deliberately narrow. We are not describing an autonomous newsroom. We are describing a video chain, camera to viewer, assembled from open-source components that an agent can provision and call, with engineers in the loop at every step.
The building blocks sit on the Open Source Cloud: 180+ open-source services covering ingest, transcoding, packaging, playout, delivery and analytics. The platform is MCP-native, so a general-purpose coding agent can provision them directly rather than through glue code we maintain for you.
What an agent can do today, and what it cannot
There are two verbs here, and conflating them is how this subject gets oversold.
Standing up means an agent provisions a service and wires it into a pipeline. Most of the chain is there today.
Steering means the running service exposes an API an agent can call to change what it is doing: start a job, retag an asset, poll a status, react to a failure. That set is much smaller. Encoding and media asset management are steerable now. Packaging, playout origination, distribution and player analytics are, for the most part, deploy-only.
We are close to an agent-operable chain. We do not have one. A human still runs most of a broadcast operation, and should.
The gap is the engagement
The steering gap does not close by adding more open-source projects to a catalog.
The services that expose an agent-callable API are, almost without exception, the ones we built ourselves. Upstream open source is typically built for humans and for scripts: it ships a config file and a CLI, not a contract an agent can read. Closing the gap means someone builds a control surface over open source that does not have one.
That is an engineering project rather than a procurement decision, and it is our day job. We build in the open with broadcasters: Open Intercom replaced proprietary intercom hardware with a browser-based system built together with SVT, YLE, NRK and TV2 Norway, and Open Live put a vision mixer, audio engine and multiviewer into software.
What a control surface looks like: Open Videocore
Open Videocore is the worked example. At its core it is media asset management with no user interface at all: ingest, transcode, package, search and deliver, exposed as one REST API with a committed OpenAPI contract. 100 endpoints across 18 resource groups, and no SDK to install.
The contract is the point. A general-purpose coding agent reads
openapi.json and calls every endpoint directly. Ask it to
upload a large file, wait for it to process, package it for streaming
and hand back a playback link, and it makes the multipart upload
calls, polls the asset status, runs the pipeline and fetches the
delivery URL, with no bespoke integration in between. When the agent
is connected to the Open Source Cloud over MCP, the platform
authenticates the calls, so there is no instance URL or bearer token
to hand it at all.
Open Videocore is Apache-2.0 and runs on the Open Source Cloud.
How we work with you
Build the control surface. We wrap the open-source component you depend on in an API that your agents and your engineers can both call.
Build the chain. We design, build and operate the pipeline on the Open Source Cloud, camera to viewer, and hand it over running.
Embed a Builder. A senior streaming engineer inside your team, with the platform and the tooling behind them.
Frequently Asked Questions
What is Agentic TV Operations?
A video chain, camera to viewer, assembled from open-source components that an AI agent can provision and call, with engineers accountable for the result. It covers ingest, transcoding, packaging, playout, delivery and analytics running as services on the Open Source Cloud.
Can an AI agent run our channel today?
No. An agent can provision most of the chain and it can steer some of it, notably encoding and media asset management. Packaging, playout origination, distribution and player analytics are mostly deploy-only right now. A human still runs a broadcast operation.
What is the difference between provisioning and steering?
Provisioning stands a service up and wires it into a pipeline. Steering changes what a running service is doing: starting a job, retagging an asset, polling a status, reacting to a failure. Steering needs an API the agent can call, and that is what most open-source media software does not ship.
Do we have to use the Open Source Cloud?
The components are open source and you can run them where you want, including your own cloud. The Open Source Cloud is the fastest path to a running chain, and it is what makes the agentic part work today, because an agent can provision and authenticate against it over MCP.
Tell us which part of your chain you want an agent to drive
Bring us the component that has no API, the pipeline that is stitched together by hand, or the workflow your team runs on a checklist. We will tell you what an agent can take over today and what it cannot.
Start a Conversation →