Match AI to the right problem and the right user.
Start by figuring out whether AI is even a fit. Describe a user pain point below and get an honest verdict, mapped to the problem categories LLMs are actually good at — then jump into the Recommender, Dashboard, or Compare.
A simple test before you build
- Fuzzy input or fuzzy output? If yes, AI is probably a fit.
- Would a smart intern help here? If yes, an LLM probably can too.
- Is being approximately right valuable? If only exact answers count and there's no review step, prefer rules/solvers.
- Is the cost of being wrong recoverable? If not, design human-in-the-loop or skip AI.
Got a problem in mind? Find the AI opportunity.
Describe a user pain point or workflow. You'll get an honest take on whether AI fits, mapped to the categories below — plus cautions and next steps.
Summarization & Compression
Turn long, messy content into the gist a human actually needs.
User pain points
- Users drowning in long PDFs, emails, transcripts, or chat threads
- Knowledge workers re-reading the same docs to brief others
- Customer-support agents needing a 10-second case summary
Examples
When not to use AI
When users need exact source text, citations, or legal-grade fidelity without review.
Catalog use-cases
Extraction & Structuring
Pull structured data out of unstructured text, images, or audio.
User pain points
- Manual data entry from documents/screenshots
- Unstructured customer feedback that never reaches analytics
- Onboarding flows that ask users to retype info from a doc they already uploaded
Examples
When not to use AI
When the source format is fully predictable — a regex, parser, or OCR with rules will be cheaper and more reliable.
Catalog use-cases
Classification & Routing
Sort, label, prioritize, or route inputs without writing brittle rules.
User pain points
- Support inboxes that need a human to triage every ticket
- Content moderation backlog
- Sales lead scoring stuck on rigid heuristics
Examples
When not to use AI
Very high volume + tight latency: a small fine-tuned classifier or simple ML may be 100× cheaper.
Catalog use-cases
Semantic Search & Q&A over your data
Let users ask questions in natural language and get answers grounded in your content.
User pain points
- Help docs nobody can find the right page in
- Internal wikis that grow faster than anyone reads them
- Long product catalogs with poor keyword search
Examples
When not to use AI
When the corpus changes by the second, or when answers must be 100% verbatim from a source — pair with strict citation UX.
Catalog use-cases
Drafting & Content Generation
Get from blank page to first draft in seconds.
User pain points
- Blank-page paralysis on writing tasks
- Repetitive replies (returns, FAQs, intros)
- Personalization at scale (1:1 messages for thousands of users)
Examples
When not to use AI
When the output is published unedited under a brand or legal name without review.
Transformation & Translation
Convert content from one format, language, or style to another.
User pain points
- Localizing content into 20 languages
- Making technical writing readable for non-experts (or vice versa)
- Migrating code between frameworks/languages
Examples
When not to use AI
Legally binding translations, or anywhere a small human error could cause material harm.
Catalog use-cases
Reasoning & Decision Support
Walk through a multi-step problem and explain the answer.
User pain points
- Users forced to do calculations or comparisons across many docs
- Analysts needing an explained recommendation, not just a number
- Complex eligibility / pricing / config decisions
Examples
When not to use AI
When the domain has a deterministic solver (taxes, accounting, physics sim) — call the solver and let the LLM explain it.
Catalog use-cases
Agents & Tool Use
Let the model take actions across tools and APIs to complete a goal.
User pain points
- Multi-system workflows (search → schedule → notify → log)
- Power-user tasks that today require switching between 5 tools
- Internal ops scripts no one wants to maintain
Examples
When not to use AI
Irreversible high-stakes actions (large payments, prod deletions) without human approval gates.
Vision Understanding
Reason about images and screenshots in natural language.
User pain points
- Users needing to describe an issue but only able to share a screenshot
- Visual product search ('find me ones like this')
- Form/document understanding without templates
Examples
When not to use AI
Safety-critical perception (autonomous driving, medical imaging diagnosis) — use specialized models.
Catalog use-cases
Voice & Conversational Interfaces
Natural conversation as a primary interface.
User pain points
- Drivers, clinicians, field workers who can't use a keyboard
- Phone-based customer support
- Accessibility for low-vision or low-literacy users
Examples
When not to use AI
Noisy environments without good ASR, or when transcripts must be legally verbatim.
Personalization & Recommendation Explanations
Tailor content per user — and explain why.
User pain points
- Generic onboarding that ignores user role/goals
- Recommendation engines users don't trust because they're a black box
- Email campaigns that feel like spam
Examples
When not to use AI
When personalization could leak sensitive inferences (health, religion, sexuality) the user didn't consent to share.
Catalog use-cases
Pattern Recognition & Anomaly Spotting
Notice the weird thing in a pile of data.
User pain points
- Teams reading 1,000 reviews to find recurring complaints
- Ops staring at logs trying to spot drift
- PMs synthesizing user interviews
Examples
When not to use AI
Numerical anomaly detection at scale — pure ML/statistics is more accurate; use the LLM to explain findings to humans.
Catalog use-cases
Tutoring, Coaching & Explanation
Meet the user at their level and explain things until they get it.
User pain points
- One-size-fits-all training content
- Users abandoning products because they don't understand a concept
- Lack of patient, on-demand expertise
Examples
When not to use AI
Regulated advice (medical/legal/financial) without clear disclaimers and licensed-human escalation.
Creative Ideation & Variation
Generate options when the answer is 'show me 20'.
User pain points
- Creative blocks
- A/B testing that needs more variants than humans can write
- Solo founders without a creative team
Examples
When not to use AI
When originality and authorship matter for trademark or art purposes — outputs are statistical remixes.
Catalog use-cases
Code Understanding & Generation
From boilerplate to whole features, with the human in the loop.
User pain points
- Boilerplate fatigue
- Onboarding to unfamiliar codebases
- Maintenance work nobody wants to do
Examples
When not to use AI
Critical infrastructure or security-sensitive code without rigorous human review and tests.
Simulation & Role-play
Play a persona to help humans practice or explore.
User pain points
- Sales/support reps with no realistic practice environment
- Job seekers without mock interviewers
- Therapists/coaches needing role-play scenarios
Examples
When not to use AI
Therapy or crisis support replacement — always route to qualified humans for real distress.