Parallelization
Do independent work at once
Some tasks contain parts that do not depend on each other. Running them one after another only adds waiting time.
Parallelization sends several model calls at the same time and combines their outputs in code. It has two common forms.
1. Sectioning: split the work
Each call handles a different part of the task. For example, one call checks a pull request for security issues, another for performance, and another for style.
Look for unsafe input handling and leaked secrets.
Runs at the same timeLook for slow queries and repeated work.
Runs at the same timeCheck naming, structure, and tests.
Runs at the same time2. Voting: ask more than once
Several calls answer the same question, and you compare the results. Agreement increases confidence. Disagreement is a signal to check more carefully or ask a person.
3. Combine in code
Decide how results are merged before you start: join them, take the majority, or keep the strictest verdict. Also decide what happens if one call fails or times out.
When is it useful?
When parts of the task are truly independent, or when one answer is not reliable enough on its own. It costs more calls, so use it where speed or confidence matters.