The risk spectrum

Think of AI use cases on a spectrum from low-risk to high-risk:

LOW RISK                              HIGH RISK
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|                                    |
Drafting emails                    Medical advice
Brainstorming ideas                Legal interpretations
Explaining concepts                Financial decisions
Code scaffolding                   Production code review
Creative writing                   Safety-critical systems
Formatting data                    Academic citations

Low-risk use cases (verify optional)

These tasks have minimal consequences if the AI is wrong:

High-risk use cases (always verify)

These tasks have serious consequences if the AI is wrong:

Where AI excels

Understanding AI strengths helps you use it effectively in the right contexts.

Pattern recognition

AI is excellent at identifying patterns in large amounts of data:

• Finding similarities across codebases
• Identifying trends in datasets
• Recognizing common bug patterns
• Spotting inconsistencies in documentation

"Here are 50 error messages from our logs. 
Can you group them by similarity and identify common causes?"

Generating starting points

AI excels at creating first drafts and starting points:

• Initial code structure for a new feature
• First draft of documentation
• Outline for a presentation
• Template for a form or survey

Think of AI as a junior collaborator who can get you 70% of the way there, not a senior expert who delivers perfect final products.

Language and communication

AI is strong at language tasks:

• Rewriting text for different audiences
• Translating between languages
• Improving clarity and flow
• Generating variations of phrasing
• Summarizing long documents

Explaining known concepts

When explaining established concepts (not cutting-edge research), AI does well:

• How React components work
• Basic SQLWhat is sql?A language for querying and managing data in relational databases, letting you insert, read, update, and delete rows across tables. Ask AI for more queries
• Common design patterns
• Standard algorithms

The key is that these are well-documented topics with lots of training data.

Where AI struggles

Knowing AI weaknesses helps you avoid relying on it in the wrong situations.

Recent information

AI doesn't know anything that happened after its training cutoff:

• New software versions and APIs
• Current events and news
• New libraries or frameworks
• Recent security vulnerabilities
• Market conditions and prices

Nuanced reasoning

AI struggles with:

• Multi-step logical reasoning with branching paths
• Counterfactual thinking ("what if" scenarios)
• Understanding context and subtext
• Ethical dilemmas with no clear answer
• Complex trade-off analysis

Edge cases and corner cases

AI tends to give "typical" answers and misses edge cases:

• What happens with empty inputs?
• How does it handle extremely large values?
• What about special characters or unicode?
• Null or undefined handling?

// AI might write:
function divide(a, b) {
  return a / b;
}

// But miss edge cases:
divide(10, 0);        // Returns Infinity
divide("10", "5");    // Returns 2 (coercion)
divide(null, 5);      // Returns 0

Context outside the training data

AI can't know:

• Your company's specific coding standards
• Internal project context
• Your user's specific needs
• Industry-specific regulations
• Your team's conventions and preferences

The verification mindset

Professional AI users develop a default stance of healthy skepticism. This isn't paranoia, it's a workflow habit.

Assume it's wrong until proven otherwise

For high-risk tasks, start with the assumption that the AI output contains errors. Your job is to find them.

The three-question test

Before acting on AI output, ask:

1. "What would happen if this is wrong?"

• If consequences are serious → verify thoroughly
• If consequences are minor → quick check or skip

1. "Do I have expertise to evaluate this?"

• If yes → use your judgment
• If no → consult someone who does

1. "Can I verify this quickly?"

• If yes → do it now
• If no → flag for later verification

Build verification into your workflow

Don't treat verification as an afterthought. Build it into your process:

For code:

1. Get AI suggestion
2. Read and understand every line
3. Test in isolated environment
4. Check against documentation
5. Run your test suite
6. Code review with human

For research:

1. Get AI summary
2. Identify key claims
3. Find original sources
4. Verify citations exist and say what AI claims
5. Cross-reference with other sources
6. Note any contradictions

Building a personal verification toolkit

Develop your own set of trusted sources and verification methods:

Trusted documentation

For coding tasks, know where to check:

• Official language documentation (MDN, Python docs, etc.)
• Framework documentation (React, Vue, Angular)
• Package READMEs and docs
• Type definitions

Fact-checking resources

For general information:

• Wikipedia (with citation following)
• Google Scholar for academic claims
• Official government sources for statistics
• Reputable news organizations

Domain experts

Build a mental list of who to ask:

• Senior developers on your team
• Subject matter experts
• Professional communities (Stack Overflow, Reddit)
• Consultants for specialized topics

Red flags that require immediate verification

Some signals mean you should stop and verify before proceeding:

The "human in the loop" principle

For high-stakes decisions, AI should assist humans, not replace them:

• AI generates options → Human chooses
• AI drafts content → Human edits and approves
• AI flags issues → Human investigates
• AI suggests approaches → Human decides strategy

This isn't about distrusting AI, it's about recognizing that human judgment is essential for context, ethics, and consequences.

The goal isn't to avoid using AI, it's to use it wisely. Trust AI for what it does well. Verify what matters. And never let convenience override critical thinking.