What AI is genuinely good at — and what it cannot do
The problem with how AI gets talked about
AI news tends toward two extremes. Either AI is about to make every human job obsolete, or it’s a glorified autocomplete that gets things wrong half the time.
Neither is true. The reality is more boring and more useful: AI is a tool with specific strengths and specific weaknesses. Knowing both makes you a significantly better user.
What AI is genuinely good at
First drafts of almost anything. Email replies, reports, social posts, summaries — AI produces a usable first draft faster than most people can open a blank document. You still edit. But starting from something is much easier than starting from nothing.
Explaining concepts at any level. Ask it to explain quantum computing to a 10-year-old, then ask for the graduate-level version. It adjusts. This is useful for learning, for preparing presentations, or for translating technical content into something your team can understand.
Thinking through problems. Describe a decision you’re facing. Ask for pros and cons. Ask it to argue the other side. AI won’t make the decision for you — but it will raise things you hadn’t considered.
Editing and improving writing. Paste something you’ve written. Ask it to make it clearer, shorter, more formal, or more friendly. It’s a tireless editor that never gets annoyed when you ask for the fifth revision.
Handling repetitive text tasks. Reformatting data, extracting information from documents, converting bullet points into paragraphs — anything involving text manipulation at scale.
Think of AI as a very capable intern who has read everything but experienced nothing. They're fast, eager, and full of knowledge. But they've never actually done the thing you're asking about — so always check their work before sending it to a client.
What AI genuinely cannot do
Remember your previous conversations. Each new chat starts completely fresh. AI has no memory of what you discussed yesterday. Some tools are adding memory features, but by default — it doesn’t know you.
Reliably know current facts. AI is trained on data with a cutoff date. It doesn’t know what happened last week, last month, or potentially last year. Some tools can search the web, but the base models cannot.
Always be right. This is the most important limitation. AI will give you a confident, well-formatted, completely wrong answer sometimes. It doesn’t know when it doesn’t know something. It fills gaps with plausible-sounding information it has invented.
Do things in the world. Base AI tools can only produce text. They cannot book flights, send emails, make calls, or take actions unless specifically built to do so with extra tools.
Understand context it hasn’t been given. AI only knows what you tell it in that conversation. It doesn’t know your company, your situation, or your history. The more context you give, the better the output.
The hallucination problem
“Hallucination” is the term for when AI invents facts that aren’t true. It’s not a bug that will be patched — it’s a fundamental property of how these models work. They predict plausible text, not factual text.
In practice this means: never trust AI for specific facts, numbers, dates, quotes, or citations without verifying them elsewhere. It’s excellent for generating ideas and drafts. It’s unreliable as a fact source.
Ask an AI tool: "What are the most cited academic papers on habit formation?" Then check whether those papers actually exist. You'll likely find that some are real, some are invented, and some are real papers by the wrong author. This is hallucination in action — and the best reason to always verify specific facts.