Skip to content
Skip to content
Blog

Why We Build an AV AI Agent that is not an Autopilot

Almost every AV AI agent launched in the last few years gets the same pitch. Hand the technical work to a machine, and stop thinking about it. That’s the autopilot pitch, and it’s a good one, because it’s simple and it sounds like relief.

We don’t build our AV AI agent that way. We think the difference is worth explaining, not just stating.

The Pitch We Didn’t Make

An autopilot, by definition, wants you out of the loop. Its whole value proposition is that you don’t have to pay attention, because the system is handling it. That’s a reasonable design for plenty of problems. For a live show, we think it’s the wrong one.

A live broadcast or event isn’t a single decision repeated identically every time. It’s hundreds of small judgment calls made under time pressure. Most of those calls depend on context a system can’t fully see. That includes the mood of the room and a director’s instinct that something’s about to go wrong. It includes the difference between “technically correct” and “actually right for this moment.” Removing the human from that loop doesn’t remove risk. It just removes the person best positioned to catch the risk the system can’t see coming.

So instead of an autopilot, we built our AV AI agent around what we call a Second Chair. The term comes from the actual role on a lot of crews. The second chair backs up the lead. They handle what doesn’t need the lead’s judgment, so that judgment stays free for what does.

What “Human-in-the-Loop” Means for an AV AI Agent

For us, human-in-the-loop design isn’t a tagline on the About page. It shows up as specific decisions in how Showcaller AI, our AV AI agent for live shows, actually behaves:

  • It asks before it assumes. Where a command is ambiguous, the agent surfaces that instead of guessing and hoping.
  • Access needs your permission. You decide which categories of devices and commands your AV AI agent may touch on a given show. You set that before it ever hears a cue. Trust is something you extend deliberately, not something the system takes by default.
  • Tally checks actions against the device where it reports its state. It exists because “the AI sent a command” and “the AI did the thing” are different claims. Only one of them should get to call itself done. If something doesn’t take, you hear about it immediately.
  • You can review everything after the fact. Air Check keeps a record of everything that happened during a show. We don’t expect you to need it often. But a system you can’t audit is a system you have to take on faith.

None of these are flashy features. They’re closer to constraints: limits we’ve deliberately put on what our AV AI agent may do. The alternative is a system that’s more impressive in a demo and less trustworthy in an actual show.

Where an AV AI Agent Actually Earns Its Keep

On a real show, an AV AI agent isn’t there to replace a TD’s ear or a director’s eye. It’s there to handle the layer underneath. That means cutting a vMix input or repositioning a PTZ camera. It means muting a channel on an X32, or checking that a wireless pack is still transmitting. None of that needs a human’s judgment. All of it needs a human’s sign-off on whether it actually happened. That’s exactly what an audit trail like Air Check is for. That layer spans lighting consoles and wireless mic packs, as well as video switchers and audio desks. So an AV AI agent has to speak each device’s native protocol. It can’t settle for a lowest common denominator.

The Real Job: Cutting Technical Drudgery, Not Judgment

The thing worth eliminating from a Technical Director’s day isn’t decision-making. It’s the technical friction around it. You manually rig a stack of devices before you can do anything else. Then you manually check whether a command actually landed. After something goes wrong, you manually reconstruct what happened. None of that requires a TD’s expertise. All of it currently eats the attention that expertise needs.

That’s the gap we built our AV AI agent to close. The point isn’t “stop paying attention.” It’s “stop spending your attention on the part of the job that was never about judgment in the first place.”

Why This Is a Bet, Not Just a Values Statement

We could build a version of this product that does more without asking. It would skip confirmation and quietly handle edge cases instead of surfacing them. It would probably demo better, too. We think it would fail the first time it actually mattered. That’s the first time a show is on the line and something needs a human to notice it.

A Second Chair doesn’t run the show. It makes sure the Technical Director never has to choose between two jobs. One is running the show. The other is watching the technical details that make it possible. That’s the bet we’re making with Showcaller AI. It’s an AV AI agent you can audit, not one you have to take on faith. In this industry, the cost of being wrong isn’t a bad recommendation. It’s a show that went off course, in front of people watching it happen.

If that tradeoff sounds right for your show, the easiest way to see it is to watch it work. See how Air Check works. Or book a live demo to watch our AV AI agent confirm a cue against real hardware in real time.

Run your next show with Showcaller AI

Connect your gear in minutes and run your show by voice or chat.

Blog

More from the blog

October 1, 2026

Elevating Multi-Room Event Management: Showcaller Skybox Makes Its Real-World Debut at Gosec 2026

We are thrilled to share a massive milestone for the team here at Showcaller.ai. At Gosec 2026, we ran the first live, real-world test of our Skybox feature. We also tested a set of new hardware and software integrations. The live events industry is fast-paced, and the margin for error is razor-thin. We built Showcaller […]

Read more →

September 10, 2026

Showcaller Now Talks to Allen & Heath SQ

Allen & Heath SQ-5, SQ-6, and SQ-7 consoles can now be run by voice through Showcaller's native MIDI integration — scene recall and channel control without leaving your position.

Read more →

September 9, 2026

QLab Support Is Here – Voice-Trigger Your Cue Lists

QLab cue lists can now be fired, jumped, and panic-stopped by voice. Most QLab changes are read back from QLab.

Read more →