Field notes
Automation and AI do different jobs, and the difference is worth real money.
Automation follows rules you can write down. A model handles the judgment calls a rule can’t cover, and it bills you every time it runs. Mix the two up and you’ll pay a model to do a spreadsheet’s job.
Automation follows rules you can write down. Models are for the judgment a rule can’t cover, and they bill you on every call. Run the deterministic work on rules and keep models for real judgment, or you’ll pay a model to do a spreadsheet’s job.
On this page
There's a lot of "AI automation" being sold right now that is neither. The line between the two is easy enough to draw once you know where it sits.
Automation follows rules you can write down
If a deal passes €50k, notify the VP. If a lead comes in from Germany, route it to the DACH rep. You could hand the whole thing to a new hire on an index card. There's no judgment in it and no model to pay for - when this, do that.
AI handles the calls a rule can't
Read this email and tell me if the buyer is annoyed. Take these twelve messy job titles and work out which ones are decision-makers. You can't write that down as a rule, because the input changes every time, and sorting it out is the kind of judgment a model is built for.
A lot of what gets sold as "AI" is really just automation with a model's label stuck on it.
Why the difference costs you money
A rule runs for nothing and does the same thing every time. A model charges you on every call, because you're renting it by the token. Point one at a job the other should be doing and you've built something slower and more expensive, and often less reliable than the boring version you skipped.
And why it costs you trust
A rule is deterministic: feed it the same input and you get the same output, and you can prove that in an audit. A model is probabilistic, so it's right most of the time, which isn't the same as being right. For an email subject line, most of the time is fine. For deciding which leads get worked, or how a record gets updated, most of the time is how bad data gets in at scale.
So the useful question to ask about any given job is whether it needs judgment or just rules. Most of the time it's rules. And a lot of what gets sold as "AI transformation" is really the automation you should have built five years ago, finally getting built, with a language model bolted on so it sounds like the future.
We keep the two apart on purpose. Automations carry the rule-based work, and custom AI builds take on the judgment calls, but only where that judgment earns what it costs. Where we draw the line, and why, is its own manifesto. If you're not sure which half of your stack is which, the GTM diagnostic is a quick way to find out.
Use AI where it earns its place, and a rule for everything else. Your costs stay down, and the data landing in your CRM stays clean.