Yesterday I let Claude Fable 5.1 rebuild a client's website in one prompt. Today I gave it a stranger job. I told it to direct the video about that video.
Not write a script. Direct. Storyboard, hook, chapter names, which b-roll goes where, where the cuts land, and the thumbnail. Then it cut the thing.
I filmed the whole session, waiting included, because the waiting is the part nobody shows you.
What directing actually meant
The setup was simple. Claude Code was running Fable 5.1 with the video skills I built for my own shop. I pointed it at the raw Descript recording of the last video and told it to make a plan.
What came back was a real director's plan. It laid out where people leave a video and what holds them. It picked the camera moves. It named the lower thirds. It wrote the hook itself:
You still build websites by hand? One prompt, seventy one pages, a real client website rebuilt while I went downstairs.
That is a better hook than the one I would have typed, and I did not have to record it. My voice clone read it on the open, which is the one place I use a clone. My face is never generated.
The part that made it usable
One line in that plan is the whole reason this works instead of producing garbage.
Every moving thing mapped to something that already existed.
It did not invent a graphics package. It went and found the motion kit I already had, the b-roll bank I already shot, the thumbnails I already made, and it built the plan out of those pieces. When it wanted a screen tour it pointed at footage that was already on the drive.
That is the difference between an AI that helps you and an AI that hands you a wish list. If your assets, your brand, and your past work are somewhere the model can actually read, it plans against reality. If they are not, it makes something up and you spend your afternoon arguing with it.
It took thirty one minutes and I sat there
Let me kill the fantasy the same way I did last time. This was not instant.
It ran for about thirty one minutes. Video work is heavy. It renders on your machine, it checks its own output, it pulls from your systems, and it eats your computer while it does it. I was recording a screen capture on top of running a video editor on top of running Claude, and I was genuinely worried about overloading the machine.
At one point it threw out the storyboard it had just built and started over. I have no idea why. It did not like something. I let it cook.
If you sit and stare at the progress bar you will talk yourself into thinking it broke. Go do something else. That is the actual skill.
Three projects at once, and none of them are me typing
Here is what I really wanted to say in that video, and it has nothing to do with which model you use.
While that was running, two other projects were running next to it. Three at once. I was not doing any of them, and I was doing all of them, because I was the one feeding in the input and checking the material coming out.
That job has a name. Operator. You are not the person typing every line anymore, you are the person who knows what good looks like and points the machines at it. A model can produce a hundred versions of something. It cannot tell you which one your customer will care about.
It knows things, but it does not know what is cool. It does not understand culture the way you and I do. It has never sat across from your customer. That gap is not closing this year, and it is where the work moved.
So no, none of us are getting written off tomorrow. But the skill set moved, and it moved fast. Mine went up in twelve months more than it did in the five years before that, and I did not get smarter. I just stopped doing the parts a machine does better.
It is fine to copy, if you copy the right people
I said something in that video I want to repeat here, because a lot of people starting out need to hear it.
I watch Nate Herk and Jack Roberts. I like how they build. I want to make things that feel like theirs. In your first stretch in business, it is okay to be a copycat as long as you are copying the right cat.
Not word for word. Copy the concepts. The methods. The thing that made it work. Then do a little more than you did yesterday, whether that is more calls, more follow ups, more posts in the groups where your customers actually hang out. With AI on top of that you are running at ten times the pace you used to.
What AI content automation looks like for a Bay Area business
Most local businesses I talk to around San Jose have the same content problem. Not ideas. Output. Somebody films something in February and it sits on a phone until June.
What changed is that the gap between a raw recording and a finished, captioned, branded, published piece is now a process a machine can run while you are on a job site. Not the taste part. The grind part. The cutting, the captions, the thumbnails, the clips, the blog version, the posts that point back at it.
Two things have to be true first, and neither one is about the model. Your brand assets have to live somewhere a system can read them, and somebody still has to look at the output before it goes out. Every single thing in my pipeline stops at an approval step. Nothing publishes on its own. That is on purpose, and it is the thing most people skip right before they post something embarrassing.
The businesses getting real leverage out of AI video production in San Jose are not the ones with the best model. They are the ones who did the boring setup work so the model has something true to work from.
Watch the whole thing
The full session is on YouTube, thirty one minute wait and all, including the part where I rant about being an operator and the part where it scrapped its own storyboard.
Watch: I Let Claude Fable 5.1 Direct My YouTube Video
The prompts and the skills behind this run are free in the Academy. No pitch attached, they are just sitting there.
If you would rather somebody build the pipeline for you, that is the job. On Point Tech Solutions builds websites, funnels, and content systems for businesses across San Jose and the Bay Area. Tell us what you are sitting on and we will tell you honestly whether it is worth automating yet.