Everyone on our team does the same task differently. Should our first AI project standardize it or automate it?
Standardize the minimum necessary workflow first, then automate its repeatable parts. Do not encode every personal variation into the first application, and do not assume AI can resolve disagreements about how the business should operate.
Have the people doing the work walk through actual examples together. Agree on the trigger, required inputs, expected output, and conditions that require an exception. You do not need an exhaustive procedure manual. You need enough agreement to recognize a correct result.
Separate the work into three groups:
- Consistent steps with clear rules.
- Interpretation tasks where AI could assist and a person can check the result.
- Exceptions that still require an accountable person's decision.
Build the first project around the consistent steps and a narrowly defined assistance task. For example, organizing information for review may be a better starting point than allowing AI to make the final decision. This is an illustrative scope choice, not a claim about a completed deployment.
Assign an internal champion to maintain the agreed workflow after initial coaching. Validate a small sample and measure whether manual effort or cycle time actually falls before adding more complex agents.
In an EOS-run company, unresolved process disagreements belong on the Issues List for resolution. A software build should not quietly settle them. The recommendation is a small, agreed operating routine supported by automation, not an elaborate application built around conflicting expectations.
Category: First AI Project