I had a short-form video that I had previously paid someone to edit. This time I gave the edit to Claude Opus 5.5 instead, and it did it. A job I had paid a person for, done by a model.

This one is about life outside work, my own videos and experiments. Nothing in here is about my day job.

What interests me is the kind of job it was, one many of us still assume models cannot do yet.

Video matters to me because I post to around 28K people on LinkedIn, and a video that never gets edited never gets posted. So the question I ask now is “which other jobs in my week am I still treating as too real for a model?”

I know it has been a while since I last posted a video. That edit is part of how I am getting back to it, so expect more soon.

Here is a pattern I see a lot. People sort AI into two boxes. Serious work goes in the box marked "I do this myself, maybe with a little help".

Everything else goes in the box marked "toy": you try a prompt, get something impressive, and never use it again.

That edit is why I think the boxes are wrong. It told me models are capable enough now to take a real job, one that has to be finished and used.

Here is where I have landed. I use different models for different jobs.

Creating. I use Claude Code to make my LinkedIn videos.

Experimenting. My experiments live in Claude Projects, so each one has a home I can come back to.

Learning. I use Codex and ChatGPT as a learning coach to help me upskill. A good coach keeps asking what you are avoiding, and a model is patient enough to keep asking.

Running. Grok Bot helps me run this newsletter. I come up with the ideas and the outline; it drafts, checks and schedules, and I still make the final call. It also gets second opinions on my research from other models.

I treat each of these as a companion. None of them replaces judgement. Check every output the way you would check a colleague's work, before it goes anywhere.

Choose something already on your list, a job that somebody, possibly you, was going to do anyway.

Pick a job that recurs. Something you do every week and quietly resent. Recurring jobs are where a companion pays back, because the brief gets better every time.

Brief it like a colleague. Say what the job is, what done looks like and what it must not do. In my experience, most disappointing AI output comes from a disappointing brief.

Keep the judgement. Check the output as if a new colleague handed it to you: are the facts real, does it sound like you, did it do everything you asked? Note what it did well and where you had to step in. After two weeks you will know whether that job is now shared, or still yours. The card in the P.S. has all five checks.

This week's ad is from TLDR AI, a daily email on new AI tools and research. If you are handing more real jobs to models, it helps to know what changed this week.

From ML engineers to founders, 1.1M+ readers use TLDR AI to spot new tools and research before they go mainstream. One free email, every morning.

Subscribe for free

💬 Think about the last AI output you had to fix. What check would have caught it before it went anywhere?

Reply with COACH and the check that caught something for you. I will tell you how I would build it into your next brief. I read every reply.

Know someone still treating AI as a toy? Share your referral link. One friend subscribes, you get a free LinkedIn Profile Optimisation Checklist.

{{rp_refer_url}}

Shola

P.S. If you want the five checks on one page before you trust any AI output, here is the AI Output Review Card (one-page PDF), direct link, no signup. Next Tuesday: the exam that forced me to build instead of memorise, and how to study for every cert that way. 6PM UK.