e27 reported on 10 November 2026 on a small AI video production business. A single AI production worker produced roughly two hours of AI-assisted drama in a week, and the client accepted the finished project. There was no team. Model usage, generation and related tools accounted for roughly 60 per cent of direct production costs, and the worker’s own labour made up the rest.

I could only work from the feed summary, because the full article would not load for me. So I am keeping to what it says.

The excitement is about volume

The easy reading is that two hours of drama in a week shows how much a tool can now generate. One person, no crew, a client who said yes. That is a real result, and I don’t want to wave it away.

But the first question I ask is what actually changed. Generating footage got cheap and fast. That part is true. What did not change is that someone has to decide what the footage should be, and someone has to check that it is right before it goes to a client.

The 60 per cent is the part that gets quoted, and it is the smaller story

Model usage and generation took roughly 60 per cent of direct costs. That sounds like the headline. If you run a business, though, look at the other 40 per cent. That was one person’s labour, and it was the labour of adjusting, directing and deciding what stayed.

Generation is a cost that scales with how much you produce. Directing and checking is a cost that scales with how much you are willing to put your name on. Those are different things. A tool can make you ten times the footage. It cannot make the footage acceptable to your client.

AI makes volume free, or close to it. Volume was never the goal. The client did not accept two hours because it was two hours. They accepted it because it worked as a finished piece. That is where the unit economics sit.

What this means for a real operator

Take a marketing team that wants video ads, or a clinic that wants patient education clips, or a shop that wants product stories. The temptation is to price the work by what the tool costs per minute of output. That number will mislead you, because it leaves out the person who briefs, rejects, fixes the brief and signs off.

You direct, the AI produces, you check and decide. You are responsible for every output that leaves your hands. If you cost only the producing step, your margin looks better than it is. If you cost the directing and checking, you can see which jobs are worth taking and which ones will eat your week.

It also changes what you hire for. A person who can judge whether a scene works, and say clearly what to change, is worth more than someone who knows the current tool. Tools expire. Judgement does not.

Where I see the same pattern

I run content at volume for my own brands. The volume is cheap. What makes it worth anything is that each piece carries a specific frame and opinion, and that I check it before it goes out. When I ran the same approach on my own coffee store, the result was a first Google AI Overview citation in 48 hours, in Singapore, in 2026. The tool did not supply the frame. I did.

What a business should do

Before you take on AI-assisted production work, or buy it, cost it in two lines. One line is the generation and tool spend. The other is the hours of the person who directs and checks. Price the second line honestly, because it is the one that does not shrink as the tools improve.

Write the brief before you generate anything. Decide who signs off and what they are checking for. When the output misses, fix the brief and run it again, rather than regenerating until something looks fine.

Then track how many hours the checking takes per finished minute, not how many minutes the tool can produce. That figure tells you whether the job pays.

Source: e27: The hidden costs of running an AI video production business