The flood is real

You’ve seen the headlines: every couple of weeks, somebody ships the “most powerful AI ever.” This month it’s a pile-up. Anthropic put out a new flagship. Google’s next Gemini and OpenAI’s next GPT are reportedly days away. Each one lands with a benchmark chart and a wave of “you’re behind if you’re not on this” energy.

Here’s what nobody selling a course will tell you: for a store owner, almost none of it changes your week.

Why the leaderboard doesn’t matter to you

The jobs you’d actually point AI at — answering “where’s my order,” writing a product description, sending an abandoned-cart nudge — aren’t hard problems for a model. They were solved two or three generations ago. A cheap, older, “boring” model does them perfectly. Paying for the newest frontier model to write a cart-reminder email is like renting a race car to go get groceries.

And the trend quietly works in your favor. As the top models sprint ahead, the good-enough ones keep getting cheaper. The price to do a fixed task — summarize a ticket, draft an email — drops every few months whether or not you ever upgrade. You don’t have to chase anything to pocket the savings.

When is a newer model worth it? When the task is genuinely hard — dense reasoning, real code, long messy documents. A cart email is not that. Match the tool to the job.

The number that actually matters

Forget benchmark scores. Here’s the only figure worth holding onto: about two cents. That’s what it costs to have AI write one personal, on-brand abandoned-cart email — on a normal, cheap model, not a frontier one. Set that against a recovery app that takes a percentage of every sale it wins back. The cheap model isn’t the compromise; it’s the entire advantage.

So the real question is never “which model is best.” It’s “what’s the cheapest one that clears the bar for this specific job.” Answer that, wire it in once, and stop reading launch posts.

Your one move this week

Pick one task you’re paying for right now — a tool subscription, a freelancer, or just your own hour — and test it on a cheap model, once. A product description. A support reply. A week of captions. Don’t automate it yet; just check whether “good enough and two cents” is true for your store. It usually is.

And Thursday, we build the proof. The next AnvilHQ workflow — Abandoned Cart Recovery — catches a left-behind cart, reads what the shopper left, and writes them a warm, personal nudge back. About two cents an email. No code. The free prompt ships with it.

See you Thursday.

— The AnvilHQ Team

P.S. On Shopify Plus and still running Scripts? They stop executing today (June 30) — rebuild that logic as Functions or an app before you notice something break. Everyone else, ignore this and carry on.

P.P.S. Reply with the one task you’re going to test on a cheap model — I’ll send back the quickest way to try it.

Keep reading