Explore practical ways to combine artificial intelligence (AI) with ordinary automation. Choose an audience and goal, then open an idea for its workflow, review points and a way to test whether it helps.
IT · Classify informationTicket triage
Problem
Inconsistent service-request descriptions slow routing.
Where AI could help
Suggest category and clarifying questions.
Where traditional automation helps
Validate fields and route approved categories.
Example architecture
Ticket → Minimise → Classify → Rule checks → Queue
Potential benefits
Faster dispatch.
Risks & considerations
Priority errors and credentials in tickets.
Human review
Review urgent and access-related cases.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use reviewed examples from every category, including ambiguous cases. Track incorrect categories and make an explicit “needs review” outcome available.
IT · Search knowledgeRunbook search
Problem
Incident procedures are hard to find quickly.
Where AI could help
Retrieve relevant passages and cite sources.
Where traditional automation helps
Index approved versions and enforce permissions.
Example architecture
Question → Authorised retrieval → Cited draft → Engineer
Potential benefits
Less searching during incidents.
Risks & considerations
Stale instructions and source prompt injection.
Human review
Engineer verifies source before running commands.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Test known questions, missing answers and documents the caller cannot access. Measure whether relevant authorised sources appear before assessing the generated summary.
IT · Summarise informationChange summary
Problem
Long changes hide impact and rollback prerequisites.
Where AI could help
Draft impact and rollback summaries.
Where traditional automation helps
Check approvals, windows and checklist completeness.
Example architecture
Change → Extract → Checklist → Owner
Potential benefits
Clearer handovers.
Risks & considerations
Critical dependencies may be omitted.
Human review
Change owner approves final text.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare summaries with their source. Check that decisions, exceptions, numbers and unresolved questions remain accurate; a shorter summary is not automatically a better one.
HR · Search knowledgePolicy assistant
Problem
Employees repeat policy questions.
Where AI could help
Draft cited answers from published policies.
Where traditional automation helps
Restrict sources and escalate exceptions.
Example architecture
Question → Policy retrieval → Draft → HR escalation
Potential benefits
Fewer repeated searches.
Risks & considerations
Privacy and outdated policy.
Human review
HR reviews sensitive or ambiguous cases.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Test known questions, missing answers and documents the caller cannot access. Measure whether relevant authorised sources appear before assessing the generated summary.
HR · Reduce repetitive workOnboarding checklist
Problem
Starters need repeated coordination.
Where AI could help
Draft role-specific guidance.
Where traditional automation helps
Generate tasks from confirmed role and dates.
Example architecture
Starter record → Rules → Tasks → Guidance
Potential benefits
Consistent preparation.
Risks & considerations
Generated guidance must not grant access.
Human review
Managers approve access assignments.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Time the same task with and without the prototype. Include correction and review time, and check that it does not introduce duplicate actions.
Finance · Extract informationInvoice extraction
Problem
Invoices need manual entry from varied layouts.
Where AI could help
Extract supplier, amount and line items.
Where traditional automation helps
Check arithmetic and purchase-order matches.
Example architecture
Invoice → Extraction → Arithmetic → Approval
Potential benefits
Less data entry.
Risks & considerations
Fraud, bank-detail changes and OCR errors.
Human review
Finance approves payment and independently verifies bank changes.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare every extracted field with a manually checked example set. Count missing fields and incorrect values separately, especially dates, units and identifiers.
Finance · Classify informationExpense categories
Problem
Descriptions do not map cleanly to ledger categories.
Where AI could help
Suggest categories from receipts.
Where traditional automation helps
Apply thresholds and approved vendor mappings.
Example architecture
Receipt → Suggest → Policy check → Reviewer
Potential benefits
Quicker review preparation.
Risks & considerations
Tax treatment and private data.
Human review
Accountant confirms consequential choices.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use reviewed examples from every category, including ambiguous cases. Track incorrect categories and make an explicit “needs review” outcome available.
Sales · Generate contentFollow-up draft
Problem
Meeting notes must become customer messages.
Where AI could help
Draft a message and action list.
Where traditional automation helps
Validate contacts and dates; require send approval.
Example architecture
Notes → Draft → Field checks → Salesperson
Potential benefits
Faster preparation.
Risks & considerations
Invented promises and confidential notes.
Human review
Approve every outbound message.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Review drafts against a short checklist for factual accuracy, required content and audience. Keep an approval step before publishing or sending.
Sales · Improve qualityOpportunity hygiene
Problem
Opportunity records are incomplete.
Where AI could help
Summarise evidence and unclear next steps.
Where traditional automation helps
Flag missing fields and overdue activities.
Example architecture
Snapshot → Rules → Summary → Owner
Potential benefits
Cleaner pipeline reviews.
Risks & considerations
Writing style is not evidence of deal probability.
Human review
Owners confirm changes.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use examples with known errors and examples that are already correct. Count both missed errors and false alarms, and record whether the suggested fix is useful.
Customer Service · Assist humansResponse drafting
Problem
Agents repeat searches and response drafting.
Where AI could help
Draft from approved policy and conversation context.
Where traditional automation helps
Validate policy IDs and escalation rules.
Example architecture
Message → Retrieval → Draft → Agent → Send
Potential benefits
Less drafting effort.
Risks & considerations
Wrong promises and source prompt injection.
Human review
Agent reviews before sending.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask the intended user to complete a realistic task with the prototype. Measure whether it helps them make a correct decision, including the time needed to verify it.
Customer Service · Understand documentsComplaint themes
Problem
Recurring problems are buried in feedback.
Where AI could help
Group themes with representative examples.
Where traditional automation helps
Remove identifiers and count confirmed categories.
Example architecture
Snapshot → Minimise → Theme suggestions → Analyst
Potential benefits
Earlier awareness of common issues.
Risks & considerations
Sampling bias and unsupported interpretations.
Human review
Analyst checks source samples.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask questions with answers you have already verified in the source. Check whether the explanation preserves qualifications and points to the right passage.
Operations · Automate decisionsDocument intake
Problem
Incoming documents require team routing.
Where AI could help
Suggest document types.
Where traditional automation helps
Gate approved types and detect duplicates.
Example architecture
Document → Classify → Rule gate → Queue
Potential benefits
Fewer dispatch steps.
Risks & considerations
Confidential records may be misrouted.
Human review
Review unknown and sensitive documents.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Run in observation mode first: record proposed decisions without acting. Compare them with approved outcomes and test failure paths before enabling any effects.
Operations · Summarise informationShift handover
Problem
Notes span many events and systems.
Where AI could help
Summarise exceptions and outstanding actions.
Where traditional automation helps
Sort timestamps and check critical incidents appear.
Example architecture
Events → Sort → Draft → Supervisor
Potential benefits
Better shift continuity.
Risks & considerations
Missing unresolved hazards.
Human review
Supervisor signs off.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare summaries with their source. Check that decisions, exceptions, numbers and unresolved questions remain accurate; a shorter summary is not automatically a better one.
Marketing · Generate contentContent repurposing
Problem
One message needs several formats.
Where AI could help
Draft variants from an approved source.
Where traditional automation helps
Enforce length, links and publication timing.
Example architecture
Source → Variants → Checks → Editor
Potential benefits
Less repetitive drafting.
Risks & considerations
False claims and copyright concerns.
Human review
Editor approves publication.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Review drafts against a short checklist for factual accuracy, required content and audience. Keep an approval step before publishing or sending.
Marketing · Improve qualityCampaign checks
Problem
Dates, links and offers can conflict.
Where AI could help
Highlight wording contradictions.
Where traditional automation helps
Validate URLs, dates and disclosures.
Example architecture
Assets → Rules → Language review → Owner
Potential benefits
Fewer avoidable errors.
Risks & considerations
Subtle legal implications can be missed.
Human review
Owner verifies offers and disclosures.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use examples with known errors and examples that are already correct. Count both missed errors and false alarms, and record whether the suggested fix is useful.
Engineering · Summarise informationPull-request summary
Problem
Reviewers need concise change context.
Where AI could help
Draft a summary from the diff.
Where traditional automation helps
Run tests and attach actual outcomes.
Example architecture
Diff → Tests → Draft → Author
Potential benefits
Faster reviewer orientation.
Risks & considerations
Explanations can misunderstand code.
Human review
Author checks against the change.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare summaries with their source. Check that decisions, exceptions, numbers and unresolved questions remain accurate; a shorter summary is not automatically a better one.
Engineering · Improve qualityTest suggestions
Problem
Specifications can omit edge cases.
Where AI could help
Suggest cases and failure modes.
Where traditional automation helps
Create fixtures from approved independent expectations.
Example architecture
Specification → Cases → Engineer → Tests
Potential benefits
Broader failure discussion.
Risks & considerations
Tests may merely mirror implementation.
Human review
Engineer chooses expected outcomes.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use examples with known errors and examples that are already correct. Count both missed errors and false alarms, and record whether the suggested fix is useful.
Legal / Compliance · Understand documentsClause comparison
Problem
Proposed wording differs from approved clauses.
Where AI could help
Explain semantic differences with source quotes.
Where traditional automation helps
Perform exact diff and check clause identifiers.
Example architecture
Documents → Diff → Explanation → Counsel
Potential benefits
Less time locating changes.
Risks & considerations
Subtle legal meaning and confidentiality.
Human review
Qualified reviewer approves conclusions.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask questions with answers you have already verified in the source. Check whether the explanation preserves qualifications and points to the right passage.
Legal / Compliance · Extract informationEvidence checklist
Problem
Audit evidence spans files and exports.
Where AI could help
Suggest evidence-to-control mappings.
Where traditional automation helps
Check dates, owners and integrity.
Example architecture
Evidence → Extract → Map → Auditor
Potential benefits
Better audit preparation.
Risks & considerations
Plausible evidence may not prove a control.
Human review
Auditor validates each claimed control.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare every extracted field with a manually checked example set. Count missing fields and incorrect values separately, especially dates, units and identifiers.
Management · Summarise informationDecision brief
Problem
Options and evidence need a compact review.
Where AI could help
Draft options, trade-offs and open questions.
Where traditional automation helps
Validate numeric totals and source references.
Example architecture
Notes → Draft → Numeric checks → Sponsor
Potential benefits
Focused preparation.
Risks & considerations
Selective evidence and false certainty.
Human review
Sponsor verifies and decides.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare summaries with their source. Check that decisions, exceptions, numbers and unresolved questions remain accurate; a shorter summary is not automatically a better one.
Management · Reduce repetitive workAction tracking
Problem
Actions are lost in meeting notes.
Where AI could help
Extract proposed owners and deadlines.
Where traditional automation helps
Create reminders only for confirmed tasks.
Example architecture
Notes → Candidates → Confirm → Tasks
Potential benefits
Fewer forgotten actions.
Risks & considerations
Inferred obligations may misrepresent agreement.
Human review
Attendees confirm assignments.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Time the same task with and without the prototype. Include correction and review time, and check that it does not introduce duplicate actions.
Home · Reduce repetitive workMaintenance planner
Problem
Household maintenance is scattered.
Where AI could help
Extract suggested tasks from manuals.
Where traditional automation helps
Schedule confirmed intervals.
Example architecture
Manual → Tasks → Confirm → Reminders
Potential benefits
Fewer missed routine tasks.
Risks & considerations
Safety instructions are model-specific.
Human review
Check the actual manual.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Time the same task with and without the prototype. Include correction and review time, and check that it does not introduce duplicate actions.
Home · Extract informationRecipe organiser
Problem
Recipes use inconsistent structures.
Where AI could help
Extract ingredients and preparation steps.
Where traditional automation helps
Scale verified quantities and group lists.
Example architecture
Recipe → Extract → Verify → Scale → List
Potential benefits
Less list preparation.
Risks & considerations
Allergens, safety and units may be misread.
Human review
Review ingredients and instructions.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare every extracted field with a manually checked example set. Count missing fields and incorrect values separately, especially dates, units and identifiers.
Money · Classify informationSpending categories
Problem
Transaction descriptions are vague.
Where AI could help
Suggest spending categories.
Where traditional automation helps
Sum confirmed categories against user budgets.
Example architecture
Local export → Minimise → Suggest → Totals
Potential benefits
Easier spending review.
Risks & considerations
Financial privacy and transfers misclassified as expenses.
Human review
Confirm categories; no automated financial decisions.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use reviewed examples from every category, including ambiguous cases. Track incorrect categories and make an explicit “needs review” outcome available.
Money · Reduce repetitive workBill reminders
Problem
Bills and renewals have varied dates.
Where AI could help
Extract dates and cancellation terms.
Where traditional automation helps
Remind on confirmed dates only.
Example architecture
Bill → Extract → Confirm → Reminder
Potential benefits
Fewer missed deadlines.
Risks & considerations
Dates and terms can be ambiguous.
Human review
Check the original bill.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Time the same task with and without the prototype. Include correction and review time, and check that it does not introduce duplicate actions.
Travel · Assist humansTrip checklist
Problem
Preparation differs by trip.
Where AI could help
Draft from a supplied itinerary and constraints.
Where traditional automation helps
Combine confirmed dates and task dependencies.
Example architecture
Itinerary → Draft → Traveller edits → Reminders
Potential benefits
Less repetitive planning.
Risks & considerations
Entry requirements and timetables change.
Human review
Verify current requirements independently.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask the intended user to complete a realistic task with the prototype. Measure whether it helps them make a correct decision, including the time needed to verify it.
Travel · Understand documentsItinerary notes
Problem
Bookings span emails and PDFs.
Where AI could help
Extract confirmation codes and dates.
Where traditional automation helps
Sort verified events and detect overlaps.
Example architecture
Bookings → Extract → Verify → Timeline
Potential benefits
Clearer personal itinerary.
Risks & considerations
Private data and timezone mistakes.
Human review
Confirm dates and timezones.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask questions with answers you have already verified in the source. Check whether the explanation preserves qualifications and points to the right passage.
Learning · Generate contentStudy questions
Problem
Passive reading gives little feedback.
Where AI could help
Draft questions from supplied material.
Where traditional automation helps
Shuffle approved questions and check answers.
Example architecture
Material → Questions → Learner → Source check
Potential benefits
Active recall practice.
Risks & considerations
Incorrect answer keys.
Human review
Check against authoritative sources.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Review drafts against a short checklist for factual accuracy, required content and audience. Keep an approval step before publishing or sending.
Learning · Understand documentsConcept comparison
Problem
Related concepts are easy to confuse.
Where AI could help
Draft contrasts from supplied references.
Where traditional automation helps
Present a fixed comparison checklist.
Example architecture
References → Comparison → Verify → Quiz
Potential benefits
Clearer distinctions.
Risks & considerations
Invented facts or omitted conditions.
Human review
Verify technical claims.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask questions with answers you have already verified in the source. Check whether the explanation preserves qualifications and points to the right passage.
Productivity · Extract informationEmail actions
Problem
Next steps are hidden in long messages.
Where AI could help
Suggest actions and deadlines.
Where traditional automation helps
Create reminders only after confirmation.
Example architecture
Selected text → Actions → Confirm → Tasks
Potential benefits
Less rereading.
Risks & considerations
Sensitive text and ambiguous obligations.
Human review
Approve each task.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Compare every extracted field with a manually checked example set. Count missing fields and incorrect values separately, especially dates, units and identifiers.
Productivity · Assist humansDaily planning
Problem
Long lists need a workable schedule.
Where AI could help
Draft groupings from supplied priorities.
Where traditional automation helps
Check time blocks and over-allocation.
Example architecture
Tasks → Priorities → Draft → Capacity check
Potential benefits
More realistic plans.
Risks & considerations
Effort estimates are uncertain.
Human review
Choose the final schedule.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask the intended user to complete a realistic task with the prototype. Measure whether it helps them make a correct decision, including the time needed to verify it.
Organisation · Reduce repetitive workFile naming
Problem
Downloads have inconsistent names.
Where AI could help
Suggest meaningful titles.
Where traditional automation helps
Apply a naming convention and flag collisions.
Example architecture
File → Title → Name checks → Preview
Potential benefits
Easier organisation.
Risks & considerations
Private documents and accidental overwrites.
Human review
Preview every rename.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Time the same task with and without the prototype. Include correction and review time, and check that it does not introduce duplicate actions.
Organisation · Search knowledgeKnowledge index
Problem
Useful notes become hard to locate.
Where AI could help
Retrieve notes and draft cited summaries.
Where traditional automation helps
Use local indexes and explicit folder filters.
Example architecture
Notes → Local index → Retrieve → Summary
Potential benefits
Faster rediscovery.
Risks & considerations
Private notes and stale content.
Human review
Read sources before acting.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Test known questions, missing answers and documents the caller cannot access. Measure whether relevant authorised sources appear before assessing the generated summary.
Creativity · Generate contentWriting outline
Problem
An idea lacks structure.
Where AI could help
Suggest outlines and alternative structures.
Where traditional automation helps
Check word budgets and manage drafts you control.
Example architecture
Brief → Outlines → Writer → Draft
Potential benefits
Less blank-page friction.
Risks & considerations
Generic voice and invented facts.
Human review
Writer reviews the final piece.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Review drafts against a short checklist for factual accuracy, required content and audience. Keep an approval step before publishing or sending.
Creativity · Generate contentCreative exercises
Problem
Practice becomes repetitive.
Where AI could help
Suggest variants within chosen constraints.
Where traditional automation helps
Combine approved themes and time limits.
Example architecture
Constraints → Candidates → Select → Exercise
Potential benefits
More varied practice.
Risks & considerations
Factual or cultural claims may be wrong.
Human review
Select prompts that fit your intent.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Review drafts against a short checklist for factual accuracy, required content and audience. Keep an approval step before publishing or sending.
Shopping · Improve qualitySpecification comparison
Problem
Descriptions hide comparable details.
Where AI could help
Extract features from supplied product descriptions.
Where traditional automation helps
Normalise units against explicit requirements.
Example architecture
Specs → Extract → Unit checks → Compare
Potential benefits
Less manual table work.
Risks & considerations
Prices and availability change; claims may be inaccurate.
Human review
Verify current specifications and pricing yourself.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use examples with known errors and examples that are already correct. Count both missed errors and false alarms, and record whether the suggested fix is useful.
Shopping · Classify informationShopping list
Problem
Mixed lists make errands inefficient.
Where AI could help
Suggest categories for ambiguous items.
Where traditional automation helps
Group known items and deduplicate.
Example architecture
List → Categorise → Group → Checklist
Potential benefits
Fewer duplicate purchases.
Risks & considerations
No inferred medical or dietary suitability.
Human review
Confirm substitutions and quantities.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Use reviewed examples from every category, including ambiguous cases. Track incorrect categories and make an explicit “needs review” outcome available.
Personal projects · Assist humansProject breakdown
Problem
A hobby project feels too large.
Where AI could help
Draft milestones and dependencies.
Where traditional automation helps
Calculate budgets and track confirmed tasks.
Example architecture
Brief → Milestones → Checks → Owner plan
Potential benefits
A clearer first action.
Risks & considerations
Unrealistic estimates and missing constraints.
Human review
Review feasibility.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Ask the intended user to complete a realistic task with the prototype. Measure whether it helps them make a correct decision, including the time needed to verify it.
Personal projects · Automate decisionsSensor rules
Problem
Project events need repetitive responses.
Where AI could help
Explain unusual patterns from a supplied log.
Where traditional automation helps
Use fixed thresholds, debounce and fail-safe rules.
Example architecture
Sensor → Validate → Rules → Reversible action
Potential benefits
Less repetitive monitoring.
Risks & considerations
False triggers and physical safety risks.
Human review
Use independent deterministic safety controls.
Start with a small pilot
Choose a small set of representative examples you are allowed to use. Include an easy case, an ambiguous case and a missing or malformed input. Keep the first version as a preview so someone can inspect its output before any action.
How to judge success
Run in observation mode first: record proposed decisions without acting. Compare them with approved outcomes and test failure paths before enabling any effects.
A bundled knowledge base. Exploring ideas never calls a model. The architectures describe systems you might build separately.
