Here's our view, after years of developing products and running plants: the tools built for human engineers work even better for bots.
That sounds backwards. Let's walk through it.
Why people found the tools hard
An FMEA is a big table. Every part, every way it could fail, three scores per row. Done properly for a small product, it might be a hundred rows.
People are good at the first ten rows. By row sixty they are tired, copying scores from the row above, and thinking about lunch. A lot of FMEAs end up as paperwork: filled in so someone can say it was done.
Why bots find them easy
A bot doesn't get bored at row sixty. It writes row sixty with the same care as row one. It doesn't mind a checklist with forty items. It can redo the whole table after a design change in a few minutes.
So the cost of using the tool properly drops close to zero.
The tool gives the bot structure
This is the bigger point. Left alone, a bot does what a smart, eager new employee does. It answers the question in front of it. It chases interesting detail. It can spend a long reply on something that doesn't matter.
Engineers have the same problem. They get bogged down in detail. The tools were built to fix that:
- An FMEA makes you list every failure before you fix any of them.
- A Pareto chart makes you rank, so you work the big items first.
- 5 Whys makes you keep going past the first easy answer.
- A control plan makes you say exactly what gets checked, and what happens when it fails.
Each tool is a shape for thinking. Give a bot the shape and it fills it in, step by step, instead of wandering.
It fits how computer intelligence works
A bot works well with clear steps, a defined output and a stopping rule. Most engineering tools are exactly that: a form, a sequence and a rule for when you're done. The method was there all along. Nothing here is new.
Where the idea stops working
The tools don't supply knowledge. A bot can fill a risk table with failures that sound right and don't belong to your product. People did that too, with pasted boilerplate. A bot can do it faster.
So the tool needs inputs from reality: your drawings, your test results, your real complaints. And a person still reviews the result and decides. The bot does the groundwork. People check it and decide.
Also, bots are similar to people here, not identical. A tired engineer skips rows. A bot doesn't skip rows, but it can make one up. Different failure, same need for review. See Don't make it up.
Try it
Next time you ask a bot to "look for problems," ask instead: "Fill in an FMEA table for this, one row per way it could fail, scored for severity, occurrence and detection." Compare the two answers.
Back to Engineering tools for bots.