Why I’m Not Building RISE to Replace Operators

A lot of AI messaging in media still leans on the same lazy fantasy. Fewer people. Less human involvement. Automated outputs. The machine takes over the boring stuff, then quietly starts taking over the meaningful stuff as well.

I understand why that message gets pushed. It is simple, aggressive, and easy to sell on a slide.

I also think it misses the point.

I am not building RISE to replace operators because I do not think operators are the bottleneck. In a lot of live environments, they are the reason the whole thing still holds together. When a workflow is messy, the experienced human in the chain is often the one compensating for the gaps. They are the one catching the miss, correcting the timing, finding the clip that should have been easier to retrieve, or making the editorial call that no system should be making alone.

That matters, because there is a difference between removing judgement and removing friction.

I am interested in removing friction.

That means helping people get to the right material faster. It means reducing dead time spent hunting through feeds, re-checking moments, rebuilding context, or pushing content through steps that should already be lighter by now. It means using systems to narrow the field, surface likely events, attach useful metadata, and cut down the first-pass labour that burns time without adding much value.

That is a very different ambition from replacement.

And to be honest, replacement talk usually comes from too much distance. It often comes from people looking at live sport as a generic automation problem instead of a pressure environment built around timing, judgement, trust, and context. Live production is not a neat lab. It is messy, fast, and full of edge cases. The real issue is not whether a model can produce an answer. It is whether that answer is relevant, timely, stable, and usable under pressure.

That is why I do not find the “AI will replace the operator” story very interesting. It sounds bold, but most of the time it is shallow.

A healthier product philosophy is simpler. Let the system do what the system is actually good at. Process large volumes quickly. Keep track of things consistently. Recognise recurring patterns. Flag likely moments. Surface clues. Reduce the amount of blind searching. Help move an event from happening to being reviewable faster.

Then let the human do what the human is good at. Judge editorial value. Understand nuance. Notice context. Decide what matters. Decide what is safe. Decide what is ready. Decide what deserves more attention and what should be ignored.

That split is not a compromise. It is the stronger model.

It leads to better products because the interface gets built around assistance rather than fantasy. It leads to better adoption because people trust tools that respect how they actually work. And it leads to better output because the goal stops being “prove the AI can do everything” and becomes “make the whole workflow sharper.”

That is the real standard for me.

Not whether a tool can generate a flashy demo. Not whether it can produce a clever label on a clean clip. Not whether it sounds futuristic in a funding conversation. The real question is whether it helps the people inside the workflow move with more clarity and less drag.

If it cannot do that, it is noise.

This is also why I think too much AI product messaging gets the emotional tone wrong. It talks as if human involvement is the cost to be reduced. In my view, the better way to see it is the opposite. Human attention is the most expensive and most valuable thing in the chain. That is exactly why it should not be wasted on repetitive search, clumsy retrieval, first-pass sorting, or admin-heavy mechanical work that a better system should have made lighter years ago.

That is the angle behind RISE.

Not human replacement. Human amplification.

Not “how do we get rid of the operator?”
More “how do we stop wasting the operator’s time?”

Because if a system can help surface likely moments faster, keep metadata cleaner, reduce hunting, support review, and speed up the route from event to usable output, then the person in the seat gets to spend more time on actual judgement. More time on timing. More time on what deserves to go to air, to archive, to digital, or nowhere at all.

That is a better future than the usual automation fantasy.

And it is more realistic.

The industry is already moving toward AI-assisted workflows in areas like event detection, clipping, reframing, captioning, and review acceleration, but the practical implementations being discussed by broadcasters and infrastructure providers are still about support, scale, and efficiency, not some clean disappearance of human production roles. AWS’s recent FOX Sports case, for example, describes automatically detecting key moments, extracting clips, and surfacing them in a review portal within seconds. That is a useful direction because it speeds up the path to decision. It does not eliminate the need for decision. TV 2 Norway’s use of AWS to support massive live event scale points in the same direction: automation and cloud orchestration expand capability, but they do not erase the operational importance of people.  

That is closer to how I think about the problem.

Build systems that reduce drag. Build tools that behave well under pressure. Build interfaces that respect how people really work. Build around trust, not theatre.

I am not building RISE because I think operators are obsolete.

I am building it because I think their time is too valuable to keep wasting on the wrong tasks.