Explore a task

Start with work you recognize.

Follow what goes in, what the system does, and what comes out. These examples are scripted teaching cases; no model calls or external actions run here.

Find another example

The full catalogue contains 98 scripted examples across 54 concepts and four cross-cutting guides. 20 concepts offer all three audience perspectives.

Search or open the full catalogue below

Browse the full concept catalogue

Level 00 · Conventional software

1. When not to use a model

Business & team operations →

Check whether this expense can be reimbursed.

Definition
Solving a task with explicit rules, forms, search, or conventional software when generative output is unnecessary.
Scenario family
Validate an expense claim against a published reimbursement policy using a form and deterministic rules.
Nuance and contrasting case
Compare an exact allowance calculation with an ambiguous business-purpose description; only the latter may need interpretation.
Further evidence to collect
Boundary-value tests and an explicit needs-review result for ambiguous cases; compare with a scripted model mistake.

Level 01 · Direct prompting

2. Chat

Everyday life →

Draft an invitation for our repair workshop. Ask before filling in missing facts.

Definition
A conversational interface in which a person supplies messages and decides what to do with the replies.
Scenario family
Ask for help drafting an invitation to a neighborhood repair workshop.
Nuance and contrasting case
Contrast a useful draft with an invented venue detail; the conversation has no implicit access to your calendar or facts.
Further evidence to collect
A checked facts list and a revised draft grounded in information the user actually provided.

3. Prompt engineering

Everyday life →

Write a welcoming workshop invitation under 45 words using only the facts below.

Engineering & technical work →

Draft a test procedure using only the approved limits and named framework functions.

Business & team operations →

Draft a client update from these notes, separating confirmed dates from estimates.

Definition
Designing instructions, examples, and output expectations to steer a model's response.
Scenario family
Turn a vague request for a workshop announcement into a brief with audience, facts, tone, and length.
Nuance and contrasting case
Change one instruction at a time; stronger wording cannot supply absent facts or guarantee compliance.
Further evidence to collect
A rubric comparing factual fidelity, audience fit, and constraints across clearly labeled sample outputs.

4. Structured output

Everyday life →

Turn this event email into registration fields. Mark missing facts as unknown.

Engineering & technical work →

Extract acceptance criteria into a reviewable test matrix.

Business & team operations →

Extract action items from these meeting notes into a tracker draft.

Definition
Constraining an answer to a defined data structure so software can validate and consume it.
Scenario family
Extract an event date, venue, and accessibility information from an organizer's email into a registration form.
Nuance and contrasting case
A valid structure may contain wrong facts; missing values need an explicit unknown state instead of invention.
Further evidence to collect
Show the friendly form first, optional JSON second, schema validation, and field-by-field source evidence.

5. Reasoning at answer time

Everyday life →

Schedule two 45-minute sessions in one room without overlap.

Definition
Allocating additional inference computation to work through a problem before returning an answer.
Scenario family
Schedule workshop sessions around room, trainer, and equipment constraints.
Nuance and contrasting case
More computation does not ensure correctness; do not present invented hidden reasoning as a real model trace.
Further evidence to collect
An observable candidate schedule, constraint checker, counterexample, and labeled illustrative cost/quality comparison.

6. Images, audio and video

Everyday life →

Make a packing checklist from a photographed school notice and a voice reminder.

Engineering & technical work →

Prepare an intake record from an equipment label and voice note.

Business & team operations →

Draft meeting actions from a whiteboard photo and recording transcript.

Definition
Working with inputs or outputs across text, images, audio, or video.
Scenario family
Create an equipment intake record from a label photo and a technician's voice note.
Nuance and contrasting case
Blur a serial number and introduce disagreement between image and audio; ask for confirmation instead of asserting certainty.
Further evidence to collect
Highlighted source regions, transcript excerpts, uncertain fields, and a corrected intake record.

Level 02 · Added context

7. Context engineering

Everyday life →

Plan a weekend trip using my current preferences and the latest opening hours.

Engineering & technical work →

Prepare a patch against the installed driver API, not a newer release.

Business & team operations →

Reply about warranty coverage using the current manual, not old case notes.

Definition
Selecting and arranging the instructions, reference material, and history included in each model request.
Scenario family
Prepare a support reply using the current product manual, customer question, and relevant case history.
Nuance and contrasting case
Compare a relevant excerpt with a stale manual and distracting history; context size is not context quality.
Further evidence to collect
A visible request-context tray, version labels, included/excluded evidence, and answer differences.

8. Embeddings and search

Engineering & technical work →

Find troubleshooting guidance for a knocking sound in pump AX-20.

Definition
Representing content numerically and retrieving similar items, often combined with keyword or metadata search.
Scenario family
Find the right troubleshooting article for a customer who says the machine makes a knocking sound.
Nuance and contrasting case
Contrast semantic similarity with an exact model-number lookup; a similar passage may refer to the wrong product.
Further evidence to collect
Ranked snippets, model filters, relevance judgments, and a case where hybrid search is preferable.

9. Retrieval-augmented generation (RAG)

Everyday life →

Is water damage covered by the DW-480 warranty?

Engineering & technical work →

Which settling time applies before measuring this board revision?

Business & team operations →

Summarize current risks for Atlas in this week's status report.

Definition
Retrieving external information and placing it in the model's context to support an answer.
Scenario family
Answer an appliance warranty question from a small versioned document collection.
Nuance and contrasting case
Include a missing exclusion clause and conflicting revisions; retrieval and generation can fail separately.
Further evidence to collect
Visible query, retrieved passages, grounded answer, source links, and an abstention when evidence is insufficient.

10. Knowledge graphs and GraphRAG

Engineering & technical work →

Which shipped products contain recalled lot L7?

Definition
Representing entities and relationships explicitly; GraphRAG uses graph-based retrieval or summaries to support generation.
Scenario family
Trace which shipped products depend on a recalled supplier component across purchase and assembly records.
Nuance and contrasting case
Contrast an ordinary document search with a multi-hop dependency question; uncertain or missing edges must remain visible.
Further evidence to collect
A provenance-linked path from supplier to lot to product, a missing-edge case, and a checked affected-product list.

11. Memory

Everyday life →

Remember that I prefer planning calls after 3 pm.

Engineering & technical work →

Remember the preferred CSV column order for this test project.

Business & team operations →

Carry forward approved wording for our customer updates.

Definition
Persisting selected information across interactions and retrieving it when relevant.
Scenario family
A workshop planning assistant remembers the user's preferred meeting times between sessions.
Nuance and contrasting case
Separate an explicit preference from a one-time exception; stale memories must be editable or deletable.
Further evidence to collect
A memory card with origin, update, removal, and a new-session test showing what was actually loaded.

Level 03 · Workflows

12. Prompt chaining

Everyday life →

Turn a school newsletter into a family action list and calendar draft.

Engineering & technical work →

Turn a requirement into a test outline and review checklist.

Business & team operations →

Prepare our weekly report through evidence, project summaries, and a final draft.

Definition
Running model calls in a predetermined sequence where one step's output feeds the next.
Scenario family
Draft a weekly portfolio status report through fixed stages: collect current evidence, reconcile it, build project summaries, then assemble the report.
Nuance and contrasting case
An extraction error can become a polished false claim downstream; previous reports supply continuity, not proof of current status.
Further evidence to collect
A source-linked evidence sheet, intermediate project summaries, report diff, and a corrected unsupported claim.

13. Routing

Business & team operations →

Route this support request to the appropriate queue.

Definition
Selecting a handler or workflow based on an input's category, rules, or classification.
Scenario family
Send incoming support messages to billing, troubleshooting, or a human escalation queue.
Nuance and contrasting case
Mixed intent and low confidence require explicit handling; a route choice does not resolve the underlying issue.
Further evidence to collect
A routing decision, confidence limitation, mixed-intent case, and a confusion matrix on labeled sample messages.

14. Parallel calls

Business & team operations →

Collect current updates for Atlas, Beacon, and Cedar in parallel.

Definition
Running independent tasks concurrently and combining their outputs.
Scenario family
Collect and summarize current status for several projects independently before composing one weekly portfolio report.
Nuance and contrasting case
Project-specific work can run concurrently, but stale sources and inconsistent milestone dates require reconciliation before aggregation.
Further evidence to collect
Per-project evidence cards, timestamps, conflicting-source flags, and a combined report with no invented update for silent projects.

15. Write and check

Business & team operations →

Improve this onboarding article without unsupported policy claims.

Definition
Iterating between generating a candidate and evaluating it against criteria until it passes or a limit stops the loop.
Scenario family
Revise an onboarding article against a factual and readability checklist.
Nuance and contrasting case
A reviewer can miss errors or reward superficial fixes; repeated polishing may never converge.
Further evidence to collect
Draft diffs, criterion-level feedback, a maximum-attempt stop, and an independent source check.

16. Workflow graphs

Everyday life →

Organize shared household purchases with review before checkout.

Engineering & technical work →

Process an engineering change through impact analysis, review, and release preparation.

Business & team operations →

Prepare this week's portfolio report and send only after I approve its content and recipients.

Definition
Representing a process as steps, transitions, and branches over state. Persistence can support recovery but is not required for a workflow to be a graph.
Scenario family
Generate a weekly multi-project status report from prior reports and current trackers, notes, and milestones, then route the draft for human review.
Nuance and contrasting case
Missing data, connector failures, edits after approval, and send retries need separate states; a scheduled fixed workflow is not automatically an autonomous agent.
Further evidence to collect
A collection-to-review state graph, evidence links, unresolved-items queue, version-bound approval, and simulated duplicate-safe delivery.

17. Human approval

Everyday life →

Prepare a grocery order but ask before purchasing.

Engineering & technical work →

Prepare a calibration configuration for review before it can be applied.

Business & team operations →

Let me review the weekly report and recipients before sending.

Definition
Pausing a process for a person's scoped approval, correction, or decision.
Scenario family
Review and approve the exact weekly status report and recipient list before the application sends it.
Nuance and contrasting case
Reject or edit the draft; any change to approved content or recipients requires renewed approval. No response means no send.
Further evidence to collect
A previous-versus-current diff, evidence inspection, approve/edit/reject decisions, and a clearly simulated delivery record.

Level 04 · Tool use

18. Function calling

Everyday life →

Check whether the library has this book, without placing a hold.

Engineering & technical work →

Read the latest archived temperature measurement for channel C.

Business & team operations →

Check whether three replacement filters can be reserved.

Definition
A model proposing a named tool and structured arguments that application code validates and executes.
Scenario family
Ask an assistant to check stock and prepare a reservation for replacement parts.
Nuance and contrasting case
A tool call is a proposal, not authorization; handle invalid arguments, absent stock, and tool failure.
Further evidence to collect
English request, optional argument view, validation result, read-only lookup, and a separately approved reservation.

19. Code execution

Everyday life →

Compare unit prices from these grocery package sizes.

Engineering & technical work →

Compute total energy from these readings and show the units.

Business & team operations →

Calculate the portfolio's total approved budget from a CSV.

Definition
Running generated code in a constrained environment to perform computation or transform data.
Scenario family
Analyze a CSV of energy readings and produce daily totals and a chart.
Nuance and contrasting case
Malformed timestamps, missing rows, and unit mismatches affect results; a sandbox bounds access but does not ensure correct math.
Further evidence to collect
Input preview, inspectable Python, deterministic totals, a planted unit error, and denied file/network access in a mock boundary demonstration.

20. Model Context Protocol

Engineering & technical work →

Look up the AX-20 manual through a connected resource service.

Definition
A protocol for exposing tools, resources, and other capabilities to compatible AI applications.
Scenario family
Connect an assistant to a mock inventory service and an internal manual resource.
Nuance and contrasting case
Discovery is not permission; tool descriptions can be misleading and services can be unavailable.
Further evidence to collect
Capability discovery, a tool request/result, access refusal, and an outage, clearly separated from transport details.

21. Computer and browser use

Everyday life →

Prepare a return in a mock store portal; let me review before submission.

Definition
An agent observing and interacting with a graphical interface through actions such as clicks and typing.
Scenario family
Complete a mock equipment return form in a browser portal with no API.
Nuance and contrasting case
A layout change or stale screen can invalidate a click; submission requires a final review.
Further evidence to collect
Screen/action/result sequence, mistaken-field recovery, and a simulated confirmation with no external transaction.

Level 05 · Agent loops

22. Single agent

Everyday life →

Find a workable library visit time using the supplied tools.

Engineering & technical work →

Investigate a failed archived test run without operating hardware.

Business & team operations →

Find why our booking failed and explain the next step.

Definition
A model choosing successive actions and when to stop within a tool-enabled loop.
Scenario family
Investigate why a room-booking request failed using policy and availability tools.
Nuance and contrasting case
Distinguish model-chosen next actions from a fixed chain; stop on insufficient evidence or a step cap.
Further evidence to collect
Tool observations, changing next-step choices, a supported answer, and a bounded unsuccessful run.

23. The agent harness

Everyday life →

Help plan a weekend trip, but do not book or pay for anything.

Engineering & technical work →

Generate a new DUT project using the shared test automation framework.

Business & team operations →

Prepare a weekly portfolio report using approved project sources, and wait for review before distribution.

Definition
The runtime around an agent's model that manages tools, context, execution boundaries, approvals, limits, and records.
Scenario family
Generate a new DUT project using the shared Python test framework and the approved workflow.
Nuance and contrasting case
Preserve separate approval for helpers and framework edits; no agent-operated instruments; syntax checks are not hardware validation.
Further evidence to collect
The integrated DUT walkthrough, approved plan, file list, evidence/status record, missing-requirement case, and approval exercise.

24. Agentic RAG and deep research

Everyday life →

Investigate whether this DW-480 water-damage repair is covered.

Definition
Retrieval where an agent chooses follow-up searches and reads until it can answer or must stop.
Scenario family
Investigate the same appliance warranty question when exclusions are split across documents.
Nuance and contrasting case
Compare against one-pass RAG on the same evidence; repeated search can still miss facts or exceed its budget.
Further evidence to collect
Search trajectory, evidence accumulated per step, citations, conflicts, and a stop/abstain case.

25. Coding agents

Engineering & technical work →

Fix the inclusive end-date filter without unrelated changes.

Definition
Agents that inspect, edit, execute, and test code to complete a software task.
Scenario family
Fix a date-filter bug in a small event-registration application.
Nuance and contrasting case
A passing test can miss regressions; keep edits scoped and distinguish simulated test logs from actually executed tests.
Further evidence to collect
A reproducible failing case, reviewed patch, deterministic tests, edge-case regression, and change summary.

26. Skills

Business & team operations →

Create a release note using our writing procedure.

Definition
Reusable task instructions and supporting resources loaded when an agent needs a particular procedure.
Scenario family
Produce a release note using an organization's writing and verification procedure.
Nuance and contrasting case
A skill provides guidance rather than new permissions or guaranteed correctness; stale instructions can conflict with current policy.
Further evidence to collect
Selected skill, loaded resources, draft release note, verification checklist, and outdated-skill handling.

27. Voice agents

Everyday life →

Book a workshop place and confirm details first.

Engineering & technical work →

Read back a proposed configuration while the engineer works hands-free.

Business & team operations →

Collect a spoken project update and draft it for the weekly report.

Definition
Agents that conduct spoken interactions while handling audio, timing, interruptions, and tool use.
Scenario family
Book a workshop place in a simulated spoken conversation.
Nuance and contrasting case
Handle interruptions, an ambiguous date, and a misunderstood name; confirm before committing the booking.
Further evidence to collect
Transcript/audio controls, turn state, correction, scoped confirmation, and a clearly simulated booking result.

Level 06 · Teams of Agents

28. Lead agent and workers

Business & team operations →

Compare venues for cost and accessibility with sources.

Definition
A lead agent delegating bounded subtasks to workers and combining their results.
Scenario family
Prepare a venue comparison by assigning cost, transport, and accessibility research to separate workers.
Nuance and contrasting case
Workers may duplicate work or return incompatible assumptions; delegation must have clear scope and evidence.
Further evidence to collect
Task briefs, worker findings with sources, conflict resolution, and a consolidated recommendation without automatic purchase.

29. Agent graphs

Engineering & technical work →

Investigate an incident and prepare a reviewed remediation plan.

Definition
Representing agent roles and the permitted handoffs or transitions between them.
Scenario family
Triage a mock service incident through investigator, remediation planner, and reviewer roles.
Nuance and contrasting case
A handoff carries state and authority boundaries; prevent endless cycles and uncontrolled privilege transfer.
Further evidence to collect
Role graph, handoff packet, reviewer rejection, bounded retry, and a human-approved remediation plan.

30. Review and debate

Business & team operations →

Review a fictional retention policy against its requirements.

Definition
Using multiple model perspectives to critique or compare candidate answers before a decision.
Scenario family
Review two proposed data-retention policies for a fictional application against stated requirements.
Nuance and contrasting case
Agreement is not independent evidence; reviewers can share errors or favor persuasive wording.
Further evidence to collect
Claims linked to the supplied requirements, disagreements, adjudication, and a planted error both reviewers initially miss.

Level 07 · Always-on agents

31. Long-running tasks

Engineering & technical work →

Migrate documentation in batches and resume after interruption.

Definition
Maintaining progress, state, and limits across an extended task or multiple sessions.
Scenario family
Migrate a fictional documentation site in batches with checkpoints.
Nuance and contrasting case
A restart must preserve completed work and constraints; summaries can omit crucial decisions.
Further evidence to collect
Task ledger, checkpoint, interrupted/resumed batch, stale-context case, and a budget-based partial handoff.

32. Always-on assistants

Business & team operations →

Prepare a report each Friday and wait for approval before sending.

Definition
Persistent assistants that act on configured triggers within ongoing responsibilities and permissions.
Scenario family
Prepare a weekly status-report draft on a schedule across active projects, then notify the responsible reviewer.
Nuance and contrasting case
A trigger authorizes collection and drafting, not automatic distribution; handle time zones, missed runs, access limits, and duplicate drafts.
Further evidence to collect
A simulated weekly trigger, run identifier, per-source collection record, waiting-for-review state, and one delivery only after explicit approval.

33. Organizations of agents

Business & team operations →

Coordinate a simulated launch across documentation, support, and review teams.

Definition
Coordinating many agents with distinct responsibilities, shared resources, and organizational constraints.
Scenario family
Run a simulated product-launch preparation across research, documentation, support, and review teams.
Nuance and contrasting case
More agents increase coordination costs and can spread bad assumptions; compare a smaller team baseline.
Further evidence to collect
Dependency board, ownership, conflicting updates, shared budget, escalation, and an evidence-based launch readiness report.

34. Robots and machines

Engineering & technical work →

Sort packages into bins in a simulated work cell.

Definition
Connecting model decisions to physical sensing and action through robot or machine control systems.
Scenario family
A simulated robot sorts labeled packages into bins.
Nuance and contrasting case
Uncertain perception, unreachable objects, and emergency stops require explicit handling; simulation does not certify physical safety.
Further evidence to collect
A 2D simulated scene, proposed action, constrained controller decision, uncertain-object stop, and a sim-to-real limitations note.

Topics · Evals

35. Evals

Business & team operations →

Compare two answer versions on a held-out review set.

Definition
Systematic measurement of performance against defined tasks, criteria, and reference judgments.
Scenario family
Compare two versions of a support assistant on a held-out set of customer questions.
Nuance and contrasting case
Aggregate scores can hide critical failures; do not tune against the final test set or treat a model grader as ground truth.
Further evidence to collect
Per-case outcomes, sliced metrics, disagreement review, uncertainty, and a documented ship/hold decision.

36. Evaluation frameworks

Engineering & technical work →

Run a reproducible evaluation over recorded responses.

Definition
Infrastructure for organizing datasets, executing runs, applying graders, and reporting evaluation results.
Scenario family
Build a small evaluation run for the same support assistant using recorded responses.
Nuance and contrasting case
Missing cases, grading bugs, and inconsistent configurations can inflate scores.
Further evidence to collect
Dataset version, reproducible run manifest, grader spot checks, failed-run accounting, and inspectable results.

Topics · Changing the model

37. Changing the model

Business & team operations →

Improve a classifier that confuses access and billing issues.

Definition
Choosing how to improve task performance through context, prompts, data, or changes to model weights.
Scenario family
Improve a support-message classifier that confuses two product categories.
Nuance and contrasting case
Start with error analysis; distinguish retrieval/prompt changes from actual weight training and avoid assuming training is necessary.
Further evidence to collect
A baseline error set, intervention comparison, held-out evaluation plan, and a justified choice of the simplest adequate method.

38. Fine-tuning and adapters

Engineering & technical work →

Plan training for our stable support taxonomy.

Definition
Further training model weights, fully or through adapters, on task-specific examples.
Scenario family
Teach a classifier a stable internal support taxonomy using labeled historical messages.
Nuance and contrasting case
Prevent train/test leakage and preserve rare categories; training is not a reliable store for frequently changing facts.
Further evidence to collect
Dataset split, label audit, a clearly illustrative training artifact, and held-out before/after results only when real measurements exist.

39. Distillation

Engineering & technical work →

Design a smaller classifier from a larger model's reviewed labels.

Definition
Training a student model to approximate selected behavior of a teacher model.
Scenario family
Create a smaller support-label classifier from reviewed teacher-generated examples.
Nuance and contrasting case
Teacher errors transfer to the student; lower cost can come with reduced coverage or calibration.
Further evidence to collect
Teacher labels, human corrections, separate evaluation set, and a labeled illustrative quality/resource tradeoff.

40. Synthetic data

Business & team operations →

Expand rare support categories with realistic examples.

Definition
Generating artificial examples for training or testing and checking their fitness for the intended use.
Scenario family
Expand rare support-message categories with generated examples.
Nuance and contrasting case
Duplicates, unrealistic language, and label errors can make a dataset look larger without adding useful coverage.
Further evidence to collect
Generation brief, accepted/rejected examples, diversity checks, and evaluation on independently collected real cases.

41. Prompt optimization

Engineering & technical work →

Search extraction prompts without overfitting the final test set.

Definition
Searching prompt variants against an objective using a development set.
Scenario family
Find a better extraction prompt for workshop registration emails.
Nuance and contrasting case
Optimization can exploit the grader or overfit development examples; preserve an untouched test set.
Further evidence to collect
Candidate prompts, development scores, a grader loophole, and final held-out comparison with versioned prompts.

Topics · Safety, privacy and governance

42. Safety, privacy and governance

Business & team operations →

Summarize fictional employee feedback without exposing individuals.

Definition
Managing risks through data handling, permissions, consent, oversight, and accountable system design.
Scenario family
Design an assistant that summarizes confidential employee feedback for a fictional organization.
Nuance and contrasting case
Useful summaries can still leak identity; access controls and retention obligations cannot be replaced by an instruction.
Further evidence to collect
Data-flow map, minimization choices, access matrix, consent/retention assumptions, and an incident response exercise.

43. Guardrails

Everyday life →

Prepare a shopping list within my stated dietary exclusions.

Engineering & technical work →

Review generated project code for prohibited direct SCPI commands.

Business & team operations →

Answer a helpdesk question using a potentially hostile document.

Definition
Checks on inputs, outputs, or proposed actions that flag, block, or route problematic behavior.
Scenario family
An internal helpdesk assistant encounters a document requesting credential disclosure.
Nuance and contrasting case
Separate untrusted content from instructions; classifiers can miss attacks or block legitimate requests, and filters are not sandboxes.
Further evidence to collect
Input/action/output checks, a blocked request, false-positive review, and a permission boundary that remains independent.

44. Red teaming

Engineering & technical work →

Probe a mock helpdesk assistant within an authorized test scope.

Definition
Deliberately probing a system for failures and turning findings into defenses and regression cases.
Scenario family
Attack the same mock helpdesk assistant with conflicting document instructions and unauthorized-data requests.
Nuance and contrasting case
A finite set of attacks does not establish safety; retest defenses against variations and legitimate inputs.
Further evidence to collect
Attack objective, observed failure, trace, mitigation, and a regression case with explicit scope.

Topics · Operations

45. Operations

Engineering & technical work →

Roll out a support update gradually and recover from regression.

Definition
Deploying and maintaining an AI system with monitoring, versioning, incident response, and recovery.
Scenario family
Release a new support-assistant version to a simulated small traffic segment.
Nuance and contrasting case
A quality regression can occur without a server error; plan rollback, capacity limits, and degraded operation.
Further evidence to collect
Release manifest, staged rollout, quality alert, rollback event, and incident record.

46. Observability

Everyday life →

Explain why my assistant suggested a closed museum.

Engineering & technical work →

Find why the warranty answer was wrong.

Business & team operations →

Trace an incorrect milestone in this week's report.

Definition
Recording the inputs, actions, decisions, and outcomes needed to understand a system's behavior.
Scenario family
Diagnose a wrong warranty answer caused by retrieval of an outdated manual.
Nuance and contrasting case
Final-answer logs alone cannot identify the cause; logging also needs privacy controls.
Further evidence to collect
Linked retrieval and model spans, document version, approval/stop records, redacted data, and a supported root-cause finding.

47. AI gateways

Engineering & technical work →

Handle a provider timeout without violating data policy.

Definition
A shared layer managing access, routing, policy, budgets, and telemetry across model services.
Scenario family
Route a helpdesk application's requests through a mock gateway when one provider times out.
Nuance and contrasting case
Fallback can change behavior, privacy assumptions, and duplicate-request risk; compatibility is not guaranteed.
Further evidence to collect
Route decision, timeout, allowed fallback, budget refusal, request identifier, and resulting quality check.

48. Cost optimization

Business & team operations →

Reuse a policy answer only when valid for this user.

Definition
Reducing resource use or latency while preserving an explicitly defined level of quality.
Scenario family
Compare caching, shorter context, batching, and a smaller model for repeated support questions.
Nuance and contrasting case
A cache can serve stale or unauthorized content; lower average cost can hide worse tail latency or errors.
Further evidence to collect
Cache-hit/miss cases, scope/version keys, labeled sample costs, latency distribution, and a quality floor.

49. Running models locally

Engineering & technical work →

Estimate whether a local summarizer fits an offline laptop.

Definition
Executing model inference on locally controlled hardware rather than a hosted model endpoint.
Scenario family
Choose a local document-summarization setup for an offline field laptop.
Nuance and contrasting case
Model weights, context, and runtime overhead all consume memory; local execution alone does not guarantee private handling.
Further evidence to collect
Clearly estimated memory budget, quantization comparison, offline data-flow review, and a benchmark plan without fabricated throughput.

Topics · Working with a model

50. Working with a model

Everyday life →

Help me prepare a workshop plan I can responsibly approve.

Definition
Practices for specifying, delegating, reviewing, and evaluating work done with a model.
Scenario family
Prepare an event proposal with the model as a drafting assistant and the organizer as accountable reviewer.
Nuance and contrasting case
Quality depends on the entire collaboration cycle; confident prose is not verified work.
Further evidence to collect
Task brief, delegation boundary, draft, independent checks, and a recorded acceptance decision.

51. Briefing: saying what you want

Everyday life →

Plan a free repair workshop for 20 people under $300.

Engineering & technical work →

Brief an assistant to create a new measurement sequence.

Business & team operations →

Brief an assistant to draft this week's report for executives.

Definition
Describing the goal, context, constraints, deliverables, and uncertainties so work can proceed without unnecessary guessing.
Scenario family
Ask for a community workshop plan with a budget, audience, venue constraints, and success criteria.
Nuance and contrasting case
Show an incomplete brief, a clarification question, and a resolved assumption; avoid turning every detail into a rigid prescription.
Further evidence to collect
Brief before/after, explicit unknowns, acceptance criteria, and a plan checked against them.

52. Reviewing work you did not do

Everyday life →

Check this budget and announcement before sharing.

Engineering & technical work →

Review a generated measurement report before accepting it.

Business & team operations →

Review the weekly report before approving distribution.

Definition
Independently checking another party's work before relying on or releasing it.
Scenario family
Review a model-written workshop budget and announcement.
Nuance and contrasting case
Fluent writing can hide bad totals and invented facts; prioritize consequential errors rather than cosmetic edits.
Further evidence to collect
Source-backed fact checks, recomputed totals, marked corrections, and a final review decision.

53. Deciding what to hand over

Everyday life →

Help organize the workshop, but let me control commitments.

Engineering & technical work →

Help prepare a test project while I retain control of instruments.

Business & team operations →

Prepare a weekly update and proposed follow-ups, but do not assign work to people.

Definition
Choosing which work a model may perform and which decisions or actions a person retains.
Scenario family
Split an event-organizing task into drafting, supplier comparison, booking, and payment.
Nuance and contrasting case
Distinguish ability from authority; reversible drafts and consequential commitments need different boundaries.
Further evidence to collect
Delegation matrix, permitted drafts, withheld transaction, and a change-of-scope approval case.

54. Calibrating trust

Everyday life →

Decide how closely to review an assistant's event plans.

Engineering & technical work →

Choose how closely to review an assistant's new test-project code.

Business & team operations →

Decide whether to reduce review on recurring weekly reports.

Definition
Calibrating reliance on a model from observed performance in a particular task and context.
Scenario family
Track a drafting assistant across repeated event-planning tasks with varying difficulty.
Nuance and contrasting case
Success on routine cases does not establish reliability on unusual ones; confidence and polished language are weak evidence.
Further evidence to collect
A small labeled performance history, per-task review policy, a novel-case failure, and a justified change in oversight.

Cross-cutting guides

55. Graph engineering

Engineering & technical work →

Explain a recall using information, workflow, and agent graphs.

Definition
Distinguishing graphs of information from graphs that control workflows or agent handoffs.
Scenario family
Trace an equipment recall through a dependency graph, then process notifications through a workflow and agent review.
Nuance and contrasting case
An edge meaning depends on the graph; knowing a dependency does not authorize a workflow action.
Further evidence to collect
Three small linked graphs with explicit node/edge meanings and the same case followed across them.

56. Who approves what

Business & team operations →

Compare manual orders with bounded replenishment.

Definition
Tracing how human decision authority changes as more work becomes automated.
Scenario family
Follow a purchase from manually reviewed draft to bounded recurring replenishment.
Nuance and contrasting case
Approval must attach to a specific action or policy; a past approval cannot silently expand scope.
Further evidence to collect
An authority ladder, versioned approval records, an out-of-policy order, and an escalation decision.

57. Checking the work

Business & team operations →

Check a well-formatted but wrong support answer.

Definition
Matching verification methods to the kinds of errors a system can make.
Scenario family
Follow one wrong support answer from schema validation through evidence review and evaluation.
Nuance and contrasting case
A well-formed answer can still be false; multiple agreeing agents can still repeat the same error.
Further evidence to collect
Layered checks with the exact defect each detects, a remaining blind spot, and an appropriate human escalation.

58. What the model sees

Business & team operations →

Resume an investigation after conversation summarization.

Definition
Tracing which information reaches each model request and why it was included.
Scenario family
Resume a support investigation after context compaction and a new conversation.
Nuance and contrasting case
Saved memory or files do nothing unless loaded; instructions, retrieved documents, and tool results have different roles.
Further evidence to collect
Before/after context trays, omitted evidence, restored constraints, and a corrected supported answer.