People treat setting up a team of bots like a brand-new problem. It isn't.
Bots can reason a little and analyze a little. Their intelligence isn't the same as a person's, but it's similar enough that the management techniques carry over. Whenever you're directing workers who think for themselves, those techniques apply. And organizations worked out how to run those workers a long time ago.
The system every plant already uses
Walk into almost any factory certified to an ISO-style standard, like AS9100 in aerospace, and you'll find a management system. It's usually drawn as a pyramid of documents with three main tiers:
- Manual and policy. The top-level document. What we do, what we don't do, where we're going, and what we're trying to achieve.
- Procedures. How each process is run across the company: who owns it, what goes in, what comes out.
- Work instructions. Exactly how to do one job or task, step by step.
Under those sits a fourth layer, records: the evidence that the work was done the way the documents say. (ISO doesn't actually require this exact structure. It's simply the layout most organizations use. Advisera's guide has a good overview.)
The order matters. Hand a group of people work instructions that don't fit together and you get conflict. Two people follow their instructions perfectly and still work against each other. The fix is never better instructions. The fix is agreeing on the system first, then writing instructions that fit inside it.
Bots work in a similar way.
The same three tiers, for bots
| Tier | In a plant | For bots |
|---|---|---|
| 1 | Manual and policy | Principles |
| 2 | Procedures | Playbooks |
| 3 | Work instructions | Skills |
Software teams building bot frameworks keep landing on almost exactly this: principles at the top, playbooks in the middle, and skills for each specific job. They come from software, not a factory floor, and still arrive at the same structure.
Thin principles are fine
There's one difference worth noticing. An ISO top-level document can run for pages. Bot principles are often one-liners, like "subtract before you add" and "prove it works."
That isn't a gap. Most companies' top-level documents are nearly identical. They take a standard template, put their own name on it, and change a few things. Nobody needs to rewrite that for bots.
So keep your principles thin. Write down only what makes your system different from everyone else's, and let the rest go without saying.
Do you need all three?
Only if your bots work together.
- A bot that works alone only needs its own instructions: its one job, and what it does and doesn't do.
- Bots that coordinate need the shared top tiers first. That means agreed principles, plus playbooks covering what each one owns and what passes between them. Then each bot gets its skills.
The more bots that share work, the more the top of the system matters, and the more you have to figure out before you build anything.
Computer intelligence rediscovered the proven tools
Here's the funny part. The artificial intelligence world didn't start from management systems. It started from models, prompts and code. Then it ran into bots that drift, contradict each other and forget the rules. One fix after another, it worked its way back to the same answer plants reached decades ago: a manual of principles at the top, procedures in the middle, and work instructions for each job.
Nobody set out to rediscover ISO or any other set of management tools. They got there anyway, because that's what it takes to make thinking workers pull in the same direction.
What this means for you
If you've ever written a procedure, trained a new hire, or sat through an audit, you already know how to manage bots. The vocabulary is new, but the job isn't.