FOULKES / WORK & IDEAS
Better questions
make better builds.
A few principles I bring to AI and application projects: understand the source, define the decision, and make the result something a person can inspect and use.
Organize the evidence before asking AI to explain it.
Start with a representative sample of your information. Identify what each field means, who owns it, how often it changes, and which source wins when records disagree.
Preserve source references and uncertainty. A missing date or conflicting total should remain visible rather than being silently filled in.
Ask first: what would someone need to check before trusting this answer?
Choose the smallest useful decision.
“Add AI” is difficult to test. “Help a coordinator prepare a complete request for approval” gives us a user, a task, and an outcome to examine.
Agree what success looks like before choosing the technology. Compare the proposed workflow with the existing one using the same examples, including awkward cases.
Ask first: what can we prove with a focused pilot?
Plan for the day something goes wrong.
An integration needs to handle missing records, interrupted runs, and repeated requests. An assistant needs an honest response when it lacks evidence. Both need an owner who can understand what happened.
Decide which actions require approval, what is recorded, and how work resumes after a failure. Keep access limited to the task.
Ask first: can a person inspect, stop, and recover the process?
Keep exploring.
What’s the problem
you want to solve?
Tell me what you’re working around today. I’ll help you explore the options and be straightforward about the skills, information, and next steps involved.