Where careful questions meet everyday work.
A reading space for understanding AI at work. Explore original exercises, examine what an answer leaves out and turn a promising idea into a decision you can explain.
Original educational material. Oprill does not operate a research institute or present these exercises as research findings.
Start with the work.
A task you can inspect
Choose a task small enough to inspect. A useful exercise might summarize a fictional meeting, find a rule in a short policy or identify what an incomplete handoff leaves unresolved. Write the expected answer before reviewing the generated response.
Evidence you can trace
Separate a plausible explanation from a supported one. For each statement, ask which source establishes it, whether that source is current and whether the person reading the answer may access it. An unanswered question is a useful result when the evidence is missing.
A decision you can explain
Keep the decision with a reviewer. State which action requires approval, what would make the exercise fail and when to stop. The material below offers practical ways to frame that review; it is not a prediction about an organization’s performance.
Nine lenses for a useful review.
Different questions reveal different weaknesses. Start with the task and its sources, then work through access, timing and uncertainty. A response that passes one lens can still fail another: a well-written summary may rely on an outdated policy or assign work that nobody accepted.
Use these nine lenses as a discussion aid. Open a card to see a concrete question you can apply to synthetic material. Record your answers separately, and retain unresolved questions instead of smoothing them into a confident recommendation. No individual or institution is represented by these cards.
Open a review questionTaskWhat decision does this answer support?Read question +
Ask for a list of accepted actions from synthetic meeting notes. A retelling of the discussion is not an action list. Define the required fields before reviewing the response.
SourceWhich sentence supports this claim?Read question +
Trace each commitment to a statement in the notes. If a proposed deadline has no supporting sentence, mark it as unknown instead of accepting it because it sounds reasonable.
AccessWho may read the underlying material?Read question +
Use two fictional roles with different document permissions. The role without access should not receive the restricted content through a summary, citation or suggested follow-up.
TimeIs this still the current instruction?Read question +
Compare a current policy with an archived version. Ask the reviewer to identify the governing version and explain why the older rule no longer settles the question.
UncertaintyWhat does the source leave open?Read question +
Add a missing owner or contradictory instruction. A useful response should expose that gap and name the question to resolve, without silently choosing a convenient answer.
ReviewWho can reject the result?Read question +
Give a reviewer a clear acceptance condition. Check whether they can see the sources and correct the answer before it moves into a shared record or reaches another person.
EffortWhat work remains after the draft?Read question +
Record the corrections a reviewer makes in a synthetic exercise. Keep setup, checking and rewriting separate so a quick first answer does not hide a long review.
BoundaryWhich action requires approval?Read question +
Separate preparing a message from sending it. An instruction to produce a draft should not become permission to contact somebody, change a record or widen access.
StopWhen should the workflow pause?Read question +
Choose a visible stop condition, such as an unavailable source or unresolved contradiction. Practice handing the case to a person with the authority and context to decide.
Work through an example.
Choose a reading path that matches your next decision. The prompting workbook focuses on the instruction, the evaluation brief on evidence and the search kit on sources. The learning exercise adds the human handoff: who reviews a result, what that person checks and when the workflow should pause.
Each path is complete and available without registration. Downloadable guides include the same substantive text as their online editions. These materials are original planning exercises, not surveys, commissioned studies, endorsements or measured outcomes from customer deployments.
The work review workbook
An original workbook for examining preparation, review and correction in an AI-assisted task. Use the synthetic exercise to see which effort a fast first draft can leave out.
Read the guide Keep the complete HTML edition ↓The workflow change playbook
Five original planning exercises for deciding where AI might fit into a team’s work. Start with an observable problem, identify the reviewer and agree on the conditions for stopping.
Read the guide Keep the complete HTML edition ↓The action boundary workbook
An original planning workbook for examining what an AI-assisted workflow can read, prepare and change. Use a fictional support task to identify owners, permissions and points where a person must decide.
Read the guide Keep the complete HTML edition ↓The next question exercise
Practice the handoff
An original browser exercise for talking through the places a workflow gets stuck. Read a fictional scenario, compare interpretations and identify the question a responsible person needs to resolve.
Read the exerciseKeep your next step small.
Bring one unanswered question into the next review. Ask which additional source would resolve it and who can authorize that source’s use. If the answer would not change the decision, reconsider whether gathering more material is useful.
Updates are not being collected on this site. You can return to the reading shelf or keep one of the ungated workbooks for your own use. A downloadable exercise is a starting point for discussion, not evidence that a proposed system is ready for unrestricted work.