Day 3: What AI Is Good At (And Where It Fails)
The Concept
One of the most useful things you can do before relying on any tool is to understand where it performs well and where it does not. People who get the most from AI are not the ones who trust it for everything — they are the ones who have a clear, honest picture of when AI adds genuine value and when it is the wrong tool for the job.
Where AI genuinely excels
Language tasks are AI's strongest area. Writing, rewriting, summarising, explaining, translating, formatting, and drafting of any kind are all things AI does with impressive consistency. If a task fundamentally involves producing or transforming words, AI is almost always worth trying.
Brainstorming and ideation are another strong area. AI generates a large volume of ideas quickly, organises them on request, and helps you refine or extend them further. It is particularly good at helping you move past a blank page — giving you a starting point you can edit and build on is often faster than starting from scratch.
Explaining and simplifying are things AI does exceptionally well. If you need a complex topic broken down, jargon from a specialist letter decoded, or the same concept explained three different ways until one sticks, AI is genuinely useful here. You can specify your level of existing knowledge and ask it to adapt accordingly.
Organising and structuring information is another consistent strength — turning rough notes into a structured outline, converting a meeting summary into action points, or grouping a long list into meaningful categories. These tasks take time when done manually and are done well by AI.
Where AI struggles
Facts and figures are the most important limitation to understand. AI was trained on historical data with a knowledge cutoff — it does not know what happened last month, or last year if it happened after training ended. Beyond recency, even for events within its training period, it can produce inaccurate information with the same confidence it would use for something correct. This is the hallucination problem introduced yesterday, and it matters most when you are relying on specific factual claims.
Arithmetic requires caution. AI can do some maths, and some tools connect to calculators underneath that make them more reliable — but language models are not reliable calculators by default. Any answer involving numbers is worth checking independently.
Recent events are often simply absent or outdated. Most models were trained several months to two years before the date you are using them. People, organisations, laws, and events that changed after the cutoff may be described incorrectly or not at all.
Personal judgement cannot be outsourced to AI. It can help you think through a decision, surface options, and articulate trade-offs. What it cannot do is weigh those options against your specific circumstances, your values, or the relationships involved. It can illuminate a decision; it cannot make it for you.
The rule to carry forward
Before reaching for AI, ask: is this fundamentally a language or ideas task? If yes, use it. If it involves specific facts that matter, numbers that will be acted on, or personal circumstances only you understand — use AI as a starting point and verify what matters before acting. This single filter will save you from both the most common mistakes and the habit of under-using a tool that genuinely earns its place in your workflow.
Prompt of the day
Copy this into your AI tool and replace any bracketed placeholders.
Prompt
I want an honest map of what AI language models are reliably good at and where they frequently fail or need checking. Please give me: 1) Six to eight task types where AI is consistently useful, with a brief reason for each. 2) Five to six areas where AI makes frequent mistakes or should not be trusted without verification, with an explanation of why each is a weakness. Be honest about the limitations — do not oversell the capabilities.
Your 15-minute task
Run the prompt. Then look at one real task you are working on this week. Identify whether it falls into AI's strength zone or its weakness zone. If it is a strength, do it with AI today. If it involves a weakness area, note which part you would need to verify independently before acting on the output.
Expected win
A clear personal map of when AI is worth reaching for and when to be cautious — which stops you from both over-trusting and under-using it.
Power user tip
For any task that involves AI's weakness areas, ask: 'What part of your answer would be most important for me to verify independently?' This one follow-up question catches the most common AI errors before they become problems.