New here? Start with Which tool, when: it matches each job to the right tool, then sends you to the page that explains it.
Managing bots is a lot like managing people. Bots can reason and make judgment calls in ways similar to what people can do. Their intelligence isn't the same as ours, but it's similar enough that the systems organizations built to manage people carry over well.
In our view, computer intelligence is rediscovering something simple: time-tested tools work with whatever intelligence you pair them with. As builders set up teams of bots, they keep landing on principles, playbooks, skills and records. The best-known home for those tools is the ISO quality management system, so we borrow its structure. We recommend the tools, not ISO as such. This series takes one tool at a time and shows what it means for bots.
This is the umbrella hub. For the shop-floor methods, like FMEA, Pareto, control plans and machine safety, see Engineering tools for bots.
Want the big picture first? Read Managing bots is like managing people.
The pyramid
| ISO quality management system | Bot system |
|---|---|
| Level 1: Quality manual and quality policy | Principles |
| Level 2: Procedures | Playbooks |
| Level 3: Work instructions | Skills |
| Level 4: Records (retained documented information) | Saved files and logs |
A note on ISO
This series just says ISO when it names the source. One idea in ISO's principles is a good bridge to bots: recognize the limits and possible bias of data, analytics and automated tools.
The loop: Plan, Do, Check, Act
ISO is organized around Plan-Do-Check-Act. For bots, the loop is: write the skill, run it, test it and review the records, then change the default.
Plan
- The seven principles, for bots (Basics)
- Say what it isn't (Basics)
- Who owns what (Basics)
- Map the process before you hire the bots (Going Further)
- Risk sets the rigor (Going Further)
- Change on purpose (Going Further)
Do
- Write it down or it didn't happen (Basics)
- The current version at the point of use (Basics)
- Don't make it up (Basics)
- Nothing ships without a yes (Basics)
- Work instructions are skills (Going Further)
- Competence: right model, right tools (Going Further)
- Knowledge that survives the night (Going Further)
- Suppliers: connectors, models and other people's bots (Advanced)
Check
- Your tester bot is your internal auditor (Going Further)
- Management review for bots (Going Further)
- Measure what matters: KPIs and evals (Advanced)
Act
- Fix the cause, not the answer (Basics)
The whole system
- Your bots need a management system (Advanced)
Where bots differ
Not every tool carries over cleanly. We say so on the pages where it matters:
- Calibration. A gauge is stable. A model's output varies. Evals work more like sampling inspection. See Measure what matters.
- Customer focus. ISO's customer is usually outside. A bot's "customer" is usually its owner. See The seven principles, for bots.
- Suppliers. You can't audit a model provider's plant. You can choose, test and fence. See Suppliers.
- Engagement. Motivation doesn't carry over. Competence and clear roles do.
Engineering tools
The management system asks you to address risk and fix causes. It doesn't say which methods to use. The usual ones, FMEA, Pareto, control plans, 5 Whys, Lean and machine safety, are collected in Engineering tools for bots.
Want the background first?
If terms like model, context window or compaction are new, the foundation pages help: What is a model?, How bot memory works and Hallucinations.