The scale challenge of AI
Ask any of the current AI tools to design a process for your business and you’ll get something genuinely impressive back. A marketing workflow with fourteen stages. A customer onboarding sequence with every touchpoint mapped. A content operation that produces, reviews, approves and publishes across six channels on a fixed cadence. It will look thorough, considered, professional. It will also, in most businesses, be completely undeliverable, and the reason why is worth understanding before you build anything on top of it.
AI is trained to be complete. When you ask it to design a process, it gives you the process as it should exist in a world with unlimited hands to do the work. It doesn’t know that your marketing team is two people, one of whom is on leave half of December. It doesn’t know that the person who’d own stage seven already has a full-time job doing something else. It doesn’t know that the approval step it has sensibly inserted will sit in someone’s inbox for a week because that someone approves forty other things a week. The plan is optimised for thoroughness, and thoroughness on paper has almost nothing to do with what a real team of real people can actually carry.
Tricked in to thinking you have a plan
This is one of the quieter ways AI distracts a business rather than helping it. The plan looks like progress. It gets shared, it gets nodded at, it might even get adopted. And then it slowly fails, not dramatically, just gradually, as the gap between the fourteen-stage process and the two people meant to run it does what it was always going to do. The team falls behind, feels like they’re failing, and quietly reverts to whatever they were doing before, which at least fit in the hours available. The AI didn’t lie. It just answered a different question from the one that mattered. You asked what a good process looks like. The question you needed answered was what a good process looks like for the people and the hours you actually have.
The fix isn’t to use AI less for this kind of work. It’s genuinely good at mapping processes, spotting gaps, and thinking through edge cases, and you’d be giving up something valuable by not using it. The fix is to put the constraint in front of it rather than leaving it out. Most people prompt for the ideal and then try to trim it down afterwards, which rarely works because trimming a fourteen-stage process still leaves you reasoning from the wrong starting point. It’s far better to make human capacity a hard input from the beginning. Tell it you have two people and roughly twenty hours a week between them. Tell it the approval bottleneck exists and isn’t going away. Tell it that whatever it designs has to survive someone being away for a fortnight. Ask it to design the best process that fits inside those limits, not the best process in the abstract.
Resource vs Scale
What comes back when you do this is different in kind, not just in length. Instead of fourteen stages it might give you four, with a clear view of what you’re deliberately choosing not to do and why that’s acceptable. It will tell you which steps to automate precisely because the humans don’t have time for them, and which steps to keep human precisely because they’re the ones that matter. The constraint doesn’t make the thinking worse. It makes it real. A plan built around your actual capacity is a plan that gets run, and a plan that gets run beats a beautiful plan that gets abandoned every single time.
There’s a deeper point here about where human resource sits in any AI-shaped process, and it’s easy to miss in the rush to automate. The people aren’t a limitation on the process to be designed around or engineered out. They are the part of the process that carries judgement, picks up the things the system didn’t anticipate, and takes responsibility when something goes wrong. A process that treats its humans as the weak link to be minimised tends to produce work that is technically complete and actually hollow, the kind of output that ticks every box and helps no one. A process that treats its humans as the part that makes the whole thing trustworthy tends to produce work people can stand behind. The difference shows up not in the design document, where both look fine, but six months later in whether the thing is still running and whether anyone trusts what it produces.
The fix; highlight the challenge
So the useful discipline, whenever AI hands you a plan, is to ask a plain question of it before you adopt anything. Who, specifically, does each part of this, and do they have the hours. If the honest answer is that the plan assumes a team you don’t have, you haven’t been given a process. You’ve been given a picture of one, and the work of turning it into something your business can actually run still lies entirely ahead of you. That work, the translation from the ideal to the achievable, is where most of the value in process design actually lives. It’s also the part AI is least able to do for you, because it’s the part that depends on knowing your people, which is the one thing the model was never given.