Project managers call it the triple constraint, or the iron triangle. The classic version in PMI's literature is scope, time and cost. Quality is often added as a fourth, or swapped in for scope. The shorthand is "fast, cheap or good: pick two."
The idea: if one side changes, at least one other is likely to move. Shorten the schedule and you usually spend more, or cut scope or quality.
How it shows up with bots
Bots have their own triangle.
- Time. How fast you get an answer.
- Cost. Usage charges. Bigger models and longer prompts cost more.
- Quality. How carefully it reasons and how much it checks.
Some real levers:
- Model choice. On xAI's published pricing, grok-4.3 costs less per token than grok-4.7. The bigger model may reason better on hard work.
- Reasoning effort. grok-4.7 offers low, medium, high and extra-high reasoning settings. More reasoning takes longer and uses more tokens.
- Checks. Every verification step adds time and tokens.
- Context. Loading every file costs more than loading the right one.
See Cost and model choice for the figures.
Make the trade on purpose
The mistake is not choosing. A bot set to the biggest model and maximum checks on every small job wastes money. A bot set cheap and fast on a job where a wrong answer costs real money is a false saving.
Decide per job:
| Job | Hold fixed | Let move |
|---|---|---|
| Customer-facing figures | Quality | Time and cost |
| Morning news summary | Cost | Quality a little |
| Urgent draft for a meeting | Time | Cost |
Where bots differ
On a human project, adding people often slows things down at first. With bots, a faster setting really is faster. And a bot can redo the work cheaply, so "quality" partly becomes "how many times do we check."
The triangle still applies, though. Nothing makes all three free.
Back to Engineering tools for bots.