What AI Can and Cannot Do
Reframing AI: The Super Intern Analogy
Think of AI as a super intern โ extraordinarily well-read, incredibly fast, endlessly patient, and never offended by "dumb" questions. This intern has absorbed an enormous amount of text across almost every domain imaginable.
But this intern has real limitations:
- Needs clear direction. "Help me with that thing" gets blank stares. "Draft an apology email to a client for a delayed shipment โ sincere but professional, under 200 words" gets a solid draft in 30 seconds.
- Has a knowledge cutoff. Training data ends at a specific date. Recent events may be unknown or inaccurate.
- Sometimes confidently wrong. When uncertain, AI can "hallucinate" โ generating plausible-sounding but incorrect information rather than admitting ignorance.
- Missing your specific context. It doesn't know your company's internal policies, your client relationships, or your boss's communication preferences โ you have to tell it.
Smart but needs guidance, knowledgeable but not omniscient, fast but requiring verification.
Why Many People Feel AI "Doesn't Work"
Three common failure patterns:
- Expectations too high. They expect AI to complete tasks autonomously from vague instructions. It doesn't work that way.
- Wrong tasks. They try to use AI for things it's bad at โ predicting stock prices, sourcing today's news, making emotional decisions.
- Too shallow. They ask once, see mediocre output, and quit. AI works best as a dialogue โ prompt, feedback, follow-up, iteration.
6 Types of Tasks AI Excels At
Text Production & Rewriting โ AI's clearest strength. Drafting articles, rewriting copy, adjusting tone, expanding sections, cutting for brevity โ anything involving getting words on the page.
Organizing Structured Information โ Taking scattered information and turning it into clear structure. Meeting notes โ action items, user feedback โ issue matrix, competitor data โ comparison table.
Batch Processing Repetitive Tasks โ 50 customer inquiries to respond to? 100 product descriptions to write? AI handles template-based repetitive tasks at scale, consistently.
Brainstorming & Ideation โ Stuck and can't think of ideas? AI is a tireless brainstorming partner. Give it a topic and get 20 directions fast โ pick the promising ones and go deeper.
Knowledge Lookup & Learning โ Understanding an unfamiliar field quickly. AI explains concepts, maps out knowledge structures, and recommends learning paths โ far more efficient than stitching together search results.
Tone & Format Adaptation โ Same content, different audiences. Formal report โ casual script. Technical doc โ user-friendly guide. Chinese brief โ culturally adapted English email.
3 Critical Weaknesses
Weakness 1: Real-Time Information
AI training data has a cutoff date. Events from recent months may be unknown or inaccurate โ and AI will often state outdated information with the same confident tone as facts it does know.
Fix: For time-sensitive queries, use AI with web browsing enabled, or gather the current information yourself and feed it to AI for analysis. Never trust AI for breaking news, recent product releases, or current pricing without verification.
Weakness 2: Precise Numerical Calculation
Large language models process numbers through language prediction, not arithmetic. Multi-step calculations have a surprisingly high error rate โ and errors look just as confident as correct answers.
Fix: Use AI to design formulas and calculation logic, but run the actual numbers through Excel, Python, or a calculator. Always independently verify any number that will appear in a financial document, proposal, or contract.
Weakness 3: Deep Emotional or Contextual Judgment
AI has no emotions, no real-world experience, and no genuine understanding of your specific situation. Decisions involving complex human relationships, organizational politics, or deeply personal values require your judgment โ not a generalized AI recommendation.
Fix: Use AI to structure your thinking โ pros/cons lists, scenario analysis โ but make the decision yourself. For anything involving your specific relationships, company culture, or personal values, real human judgment is irreplaceable.
Setting the Right Expectations
AI is a co-pilot, not autopilot. A co-pilot helps with navigation, alerts you to conditions, and can take some load off โ but you're still driving, still deciding where to go, still responsible if something goes wrong.
| AI Does (Co-pilot) | You Do (Driver) |
|---|---|
| Drafts the initial version | Judges if the direction is right, revises content |
| Organizes and summarizes information | Decides what matters, makes the call |
| Generates multiple options | Chooses what fits your actual context |
| Processes text per your instructions | Reviews output for accuracy and appropriateness |
| Speeds up execution | Defines the goal and quality standard |
One more truth worth saying directly: AI amplifies your existing ability, it doesn't create ability from nothing. An experienced strategist using AI gets insightful strategy. Someone who knows nothing about the domain gets polished-sounding nonsense. Your expertise is still the core asset โ AI helps you express and execute it faster.
3 Prompt Case Studies
Case 1: Writing an Annual Work Summary
HR requires an 800โ1,000 word year-end summary from every employee.
Wrong: "Write me an annual work summary." โ AI knows nothing about your work, your achievements, or your company. Result: generic filler text with zero relevance to your actual year.
Right: Provide specific accomplishments with real data, state the format and tone requirements, and give AI the raw material it needs to organize and express โ which is exactly what it's good at.
Case 2: Competitor Analysis
You need to analyze Competitor A's recent moves for next week's strategy meeting.
Wrong: "Analyze Competitor A's strategy over the past 3 months." โ AI's data is outdated. It may generate confident-sounding analysis that's actually fabricated or stale.
Right: Gather current information yourself from their website, industry coverage, and sales team feedback. Then ask AI to analyze strategic intent, find differentiation opportunities, and suggest 3 specific responses. Human + AI collaboration in its correct form.
Case 3: Responding to a Serious Customer Complaint
A key client sends an angry complaint email alleging product defects caused lost orders, demanding compensation.
Wrong: "A customer complained. Help me write a reply." โ AI doesn't know the client relationship, internal compensation policy, who's at fault, or your negotiating position.
Right: Provide the full context: client relationship (3 years, $120K annual spend), confirmed quality issue, compensation policy (up to $700), your position (acknowledge the issue, offer compensation without stating ceiling, protect the long-term relationship). AI can then draft what you actually mean to say, in professional language. Real situation handled in under 10 minutes.