Your Newest Producer Doesn't Sleep: AI Agents Are Now First-Class Users in OnAirFlow

Every daily show has a shift nobody staffs. The hours between the post-show wrap and the morning editorial meeting, when the wires keep moving and the board sits still.

As of today, you can staff it.

AI agents are now first-class users in OnAirFlow. Not a bolted-on integration, not a Zapier hop — an agent gets its own account in your organization, its own name in the show event log, and its own revocable credentials. It works the same board your producers do, through an API scoped to the rundown and nothing else. It's live on every tier, and it never counts toward your seats.

What "first-class" actually means

When we say an agent is a user, we mean your show's activity log reads like this the morning after:

• Wire desk (agent) filed "Port strike enters day three — freight backlog builds" to Ideas
• Wire desk (agent) attached a two-page briefing doc to the story
• Wire desk (agent) moved "Council budget vote" from Ideas to Working

Every write is attributed. When your EP asks "who moved that story," the answer is never a shrug — it's a name, even when the name belongs to software.

Here's what that looks like in practice for a daily show:

Overnight wire triage. An agent reads the wires while the desk sleeps and files candidate stories into your Ideas column — typed, titled, summarized. The morning meeting starts with a stocked board instead of a blank one.

Asset prep. Scripts, images, and briefing docs land directly on stories, up to 50MB a file. Nobody hunts a shared drive at air time.

Backlog grooming. Stale pitches killed, placeholders retitled, the Ideas column kept tight. The board your producers open is already clean.

Rundown restructuring. Columns created and reordered, stories moved between blocks. What a producer does on the board, an agent can do over the API.

None of this is aspirational. It's the shipping API surface, and it's been running production rundowns for weeks.

Two ways to connect

One click, if you use claude.ai or ChatGPT

Add our MCP server as a custom connector:

https://api.onairflow.com/mcp

That's the whole setup. The client discovers OnAirFlow's OAuth endpoints, registers itself, and sends you to a consent screen where any team member approves the connection and its scopes — read-only is an option, and the connection inherits your own permissions. No API keys to copy, nothing to paste, nothing to renew; tokens rotate automatically behind the scenes.

Each connection shows up as its own named agent under Settings → AI → Assistants, and you (or an admin) can revoke it there any time. Then you just talk to your rundown: "what's on tomorrow's show?", "file these three stories into Ideas", "move the weather block up."

Scoped keys, if you run your own agents

If you run an agent fleet — Claude Code, Cursor, a cron job, anything that speaks HTTPS — mint a key instead. Go to Settings → AI → Agents & keys, create an agent, and mint a key with the scopes you choose: read-only, or read + write. Keys are shown exactly once and stored hashed, like a password.

From there, it's plain REST with a bearer token, or the same hosted MCP server with the key in a header. The settings page generates ready-to-paste config for both when you mint. There's also a downloadable agent skill that teaches Claude the endpoint shapes, field rules, and safety rails before its first call.

Eleven MCP tools cover the whole surface: shows, stories, columns, and attachments — including base64 file upload straight onto a story.

The part your engineers will ask about

We built this like it's going to touch a live broadcast, because it is.

Access is default-deny. Agent credentials reach the rundown surface — shows, stories, columns, attachments — and nothing else. Billing, user management, org settings, and exports are off-limits by construction. Out-of-scope requests get a flat 403.

Every write is on the record. Agent edits land in the show event log under the agent's own name, same as any producer.

Revocation is instant. Keys can be revoked or deactivated from the settings page the moment something feels off; OAuth connections die from the Connections list. No deploy, no support ticket.

Concurrent edits don't clobber. The API uses the same per-field conflict detection our human editors get. If an agent and a producer touch the same field, the second writer gets a conflict response and re-reads — nobody silently overwrites the control room.

Uploads land in your storage. Agent file uploads go to your organization's connected Google Drive — your storage, your retention rules, never our metered buckets.

Why we're doing this

The newsroom stack is about to be full of agents. Wire triage, show prep, archive research — the tedious middle of production work is exactly what this generation of AI is good at. The question is whether your rundown software treats those agents as trusted, auditable coworkers or as scripts wearing a trench coat.

We think the tools that win the next decade of live production will be the ones AI can operate as easily as people can. So we built the boring parts first: real accounts, real scopes, real audit trails, real revocation. The exciting parts — what your agents actually do with a seat at the rundown — are yours to write.

Available today

Agent access ships on every tier, Creator through Newsroom, at no extra cost. Agents never count toward your seat limit. Run one or run a dozen.

• Meet the agents — what they do and the full trust model
• Connect your first agent — two-minute setup, both paths
• REST API reference — every endpoint, every error, no surprises

Put an agent on the overnight shift. Start your 7-day trial — no credit card, agents included.