May Contain Jab Ep 1: What the AI is Going On?

About twelve minutes into our first episode, Jab put it better than any slide deck could: if you're not using AI, you're mowing the lawn with scissors while everyone else is sat on a ride-on mower.

May's response was fair. The scissors are more precise. But you've got acres.

That's the show. One of us builds automations for a living. One of us has spent years teaching owners why their marketing doesn't convert, and admits he's been running his own operation like it's 2019. Between us we're trying to work out what an owner should actually do about all this.

Two things from episode one are worth writing down properly. The rest you'll want to hear.

An AI agent is not an automation, and the difference is a risk decision

These words get used interchangeably. They shouldn't be.

An automation is deterministic. A to B, B to C, C to D. Same path every time. An invoice arrives, an OCR tool reads it, the data lands in a sheet, your credits and debits track themselves. Boring, reliable, and exactly right for that job.

An AI agent has a decision in the middle. From point A it can choose B, or skip to C, or go somewhere else entirely depending on what it's looking at. That's how you get email triage that categorises, drafts against your actual internal policy, and routes the technical stuff to someone technical without writing a rule for every scenario.

May ran exactly that demo for the European Union Drug Agency. Email lands, agent categorises it, drafts a response, and inside a minute there's something sitting there for a human to review. Watching it run live changed what the room thought was possible.

But the flexibility that makes an agent useful is the same flexibility that makes it a risk. Which is why the honest answer to “should I use an agent” is another question entirely: which half of this job am I looking at?

Take bookkeeping and cashflow, the two jobs Jab raised because every owner hates them.

Reading invoices and tracking inflows and outflows is mechanical. Automate it. But a 13-week projection is iterative. It runs on assumptions that only exist in your head, and it changes as you argue with it. May doesn't build that as a workflow at all. She works through it in conversation, numbers and assumptions on the table, until the plan makes sense.

Get that split right and most of the hard part is done. Get it wrong and you've either automated something that needed judgement, or you're doing something by hand that a machine should have finished on Tuesday.

We go further into where the line sits on the episode, including the bit where an AI without guardrails discounted an entire product range by 80% overnight, and what that owner should have done differently.

Ask for the recipe, not the cake

This was Jab's moment, and it's the argument he makes to every room he trains.

Most people open an LLM and type write me an advert. That's backwards.

When you look at a great ad, a great landing page, a great email sequence, you only ever see the output. You never see the ingredients: the structure, the mechanisms, the order things go in. And that order isn't decorative. If you're baking and you add the wrong thing at the wrong point, it doesn't rise. Leave it in too long, it burns.

So don't ask AI to bake the cake. Ask it for the recipe.

What are the ingredients of a high-converting advert?

Explain SOSTAC and apply it to my business.

What order do the elements go in on a page that actually converts?

The difference is that the first approach gives you an output you can't evaluate. The second teaches you to think in a structure, which means the next time you're handed something, you'll know instinctively whether it's right.

Jab has watched someone produce a 30-page brochure this way, emoji-strewn and em-dashed to death, and call it game-changing. It wasn't. Nobody had learned the underlying structure, so nobody could tell.

The genuinely interesting part is that these frameworks used to sit behind tuition fees and paywalls. Jab paid for his the hard way through years of bootstrapping, then walking into a room full of marketers throwing acronyms at him like he was meant to know them. Now anyone can ask for the framework, have it explained, and apply it to their own business for nothing.

That should worry some people and excite the rest of you.

Also on the episode

• Why May left ChatGPT after two years, and the one thing that would win her back.
• Where customer support is genuinely ripe for automation and the part she'd never hand over.
• The reason Jab thinks AI video is the most overrated trend going, despite being a marketer.
• The one thing neither of us would automate under any circumstances.
• And the AI phone call that told Jab it was AI, and nearly closed him anyway.

Where to start this week

Pick one process you do every week that you could describe out loud, start to finish, without pausing. That's your first automation candidate, and if you can't describe it, that's useful information too.

Then pick one framework in your own field and ask an LLM to explain it and apply it to your business. Not to do the work. To show you the ingredients.

May Contain Jab: The Owner's Edge. New episodes weekly.