Investigate a recurring problem with AI
Describe the observable problem, compare possible causes and choose a small test before committing to a change.
Investigate a recurring problem with AI when the team keeps proposing fixes without agreeing what is going wrong. The problem-solver skill helps an agent distinguish a verified observation, an explanation and a possible intervention. It turns an uncertain cause into a question that can be tested instead of treating a tidy explanation as proof.
Suppose fulfilment keeps running late. Adding another person assumes a capacity problem. A repeated approval wait suggests a different cause. Measure the waiting and working time on a few comparable cases before investing in either solution. The aim of the first session may simply be a test with a named owner and a review date.
What it does
- State a symptom in terms that another person could observe.
- Separate what the team knows from what it suspects.
- Develop more than one explanation for the same outcome.
- Choose evidence that could distinguish between the explanations.
- Check whether a proposed countermeasure changes the original symptom.
Install and use
- Use an agent with an authorised example and whatever process evidence is available. A local workspace is useful for saving the test plan and results.
- Ask it to examine a repeated delay and propose a small measurement before suggesting a permanent fix.
- Agree what result would support each explanation and when the team will review it.
FAQ
Source
Made by Hieu T. Luu. Original skill: problem-solver, published on GitHub under the MIT licence.
View on GitHub