AI at Work
How to Get Ahead at Work With an AI Decision Log
Use a weekly AI-assisted decision log to expose assumptions, compare options, learn from outcomes, and turn recurring choices into better checklists.
Many people use AI at work to write faster. A decision log uses it for a more valuable job: making assumptions visible, comparing options consistently, and learning from what happens after a choice is made.
The log does not let AI make the decision. It creates a record that you can review. This is especially useful for recurring, reversible choices such as prioritizing work, choosing an experiment, planning a handoff, or selecting among several reasonable approaches.
Do not put confidential company, customer, employee, financial, security, or legal information into an external AI tool unless your organization has approved that use.
Use five fields
For every decision, record:
- Problem
- Options
- Evidence
- Risk
- Outcome
Add the date, owner, and review date when helpful. Keep each entry short enough that you will continue using it.
Decision date:
Problem:
Options considered:
Evidence available:
Assumptions:
Risks:
Decision and owner:
What would change the decision:
Review date:
Outcome:
Step 1: state the problem precisely
Avoid questions such as “What should we do?” Describe the decision, deadline, constraints, and success measure.
Rewrite this decision problem so it names:
- The choice that must be made
- The deadline
- The people affected
- The constraints
- The success measure
Use only my notes. List missing information separately. Do not recommend an option yet.
A clear problem prevents the model from solving an imagined version of the situation.
Step 2: compare options using the same criteria
Choose two or three realistic options. Ask AI to evaluate all of them against the same criteria.
Using only the evidence below, compare options A, B, and C.
For each option, list:
- Expected benefit
- Cost or effort
- Main risk
- Assumptions
- Reversibility
- Evidence that would change the recommendation
Mark missing evidence clearly. Do not make the final decision.
Check whether the comparison favors the option described in the most detail. If so, add equivalent information for the other options and run the comparison again.
Step 3: challenge assumptions
Every recommendation rests on assumptions. Ask AI to separate facts from estimates and opinions.
Audit this comparison.
Create three sections:
1. Confirmed evidence
2. Assumptions or estimates
3. Missing information
Then identify the two assumptions most likely to change the decision and suggest the cheapest safe way to test each one.
For high-impact decisions involving customers, money, access, employment, policy, security, or legal obligations, involve the appropriate human reviewer. AI output is input to judgment, not approval.
Step 4: record the decision before the outcome
Write what you chose and why before you know the result. Include the evidence that would cause you to reverse the decision. This prevents hindsight from rewriting your reasoning.
Turn my notes into a concise decision record with:
- Decision
- Why this option was chosen
- Evidence used
- Risks accepted
- Trigger for reconsideration
- Owner and review date
Preserve uncertainty. Do not add facts or make the choice for me.
Review the record yourself before saving it.
Step 5: add the outcome
One week later, or at the appropriate review date, record what happened. Compare the observed outcome with the expected one.
Compare the original decision record with the observed outcome.
Identify:
- What went as expected
- Which assumption was wrong or incomplete
- Any risk that appeared
- Evidence we should collect next time
- One change to the process
Do not judge the decision only by the outcome. Consider whether the reasoning was sound with the information available at the time.
A good decision can have a bad outcome, and a weak decision can get lucky. Review both the process and the result.
Run a monthly pattern review
At the end of the month, remove sensitive details and review the log for repeated patterns.
Analyze these redacted decision-log entries.
Find recurring patterns in:
- Assumptions that fail
- Information that arrives too late
- Approval or handoff delays
- Risks we repeatedly underestimate
- Decisions that could become a checklist
For every pattern, cite the entries that support it. Separate strong patterns from weak signals.
Convert a repeated, low-risk decision into a checklist only after a human reviews it. Keep novel or high-impact decisions out of automation.
Weekly review checklist
- Did we state the actual decision?
- Did we compare realistic options with the same criteria?
- Did we separate evidence from assumptions?
- Did we name a risk and a reversal trigger?
- Did a person make and own the final choice?
- Did we record the outcome on the review date?
- Did we remove or protect confidential information?
Writing faster saves minutes. Improving a recurring decision can save weeks of rework. A small, consistent decision log gives AI the right role: challenging your reasoning and helping you learn, while responsibility stays with you.