What can AI actually do right now? (and what it still can't)
Last checked:
Today’s free AI chat tools are good at drafting, rewriting, summarizing, planning, and explaining things, as long as you give them real details and check what comes back. They still make confident mistakes, don’t reliably know today’s facts or prices, slip on math, and can’t make judgment calls about your life. This page shows both sides, from published research and from our own testing.
Think of it as an extremely fast intern who has read a great deal, remembers some of it wrong, and never once says “I’m not sure” unless you ask it to.
For what an AI chat tool is, see Lesson 1: What an AI chat tool is (and isn’t) in our Start here lessons. This page is about what it does.
The quick version: good at, still bad at
We built this table from the guides on this site. We tested their prompts in a free AI chat tool in October 2026, and most of the “still bad at” column comes from what went wrong in those tests.
| Good at (with your details) | Still bad at (from our tests and guides) |
|---|---|
| Drafting and polishing emails: How to use AI to write a hard email | Sticking to only what you told it. In one test it added a cause (“for a Good Cause”) to a flyer that had none |
| Turning a messy to-do list into a plan: How to use AI to plan your day (and actually stick to it) | Math it does in its head. In one test it said 28 twenty-minute sessions came to eight hours and forty minutes (it’s nine hours and twenty) |
| Meal planning around a budget: How to use AI to plan a week of dinners on a budget | Prices. It doesn’t know what eggs cost at your store this week |
| Sorting and merging lists: How to use AI to make a grocery list (sorted by aisle) | Leaving blanks blank. In one test it invented quantities for items that had none |
| Explaining and quizzing: How to use AI to study for an exam | Being right about explanations. In our language test it “fixed” a correct Spanish word |
| Sorting your spending into categories: How to use AI to track your spending (without sharing account numbers) | Admitting it doesn’t know. In one test it guessed a category for a cryptic bank charge instead of marking it unsure |
| Ideas from a photo, if your tool takes one: How to use AI to design or decorate a room | Anything that needs your judgment: what to sign, what you owe, what’s safe for your health |
Notice the pattern. The left column is mostly about words and organization. The right column is mostly about facts, numbers, and knowing when to stop.
What the research says it can do
The Stanford HAI AI Index Report 2026, a yearly review of AI data from Stanford University, reports big gains. Its headline finding is that “AI capability is not plateauing. It is accelerating and reaching more people than ever.”
It also tracks AI “agents,” tools that try to complete tasks on a computer instead of only answering questions. On one widely used test of computer tasks, called OSWorld, the report says “accuracy rose from roughly 12% to 66.3%, within 6 percentage points of human performance.”
Workers are using AI chat tools for many of the same things our guides cover. In a Pew Research Center survey of U.S. workers published February 25, 2025, among those who had used AI chatbots for work, “the most common uses are research (57% have used them for this), editing written content (52%) and drafting written content (47%).”
Those workers were more impressed by the speed than the quality. Pew found: “Four-in-ten workers who have used AI chatbots for work say these tools have been extremely or very helpful in allowing them to do things more quickly. A smaller share (29%) say they have been equally helpful in improving the quality of their work.”
That matches our experience. The first draft shows up in seconds. Making it right still takes you.
What the research says it still can’t do
The same AI Index report is equally clear about the gaps.
- It’s uneven. “AI models can win a gold medal at the International Mathematical Olympiad but cannot reliably tell time,” the report says, describing “what researchers call the jagged frontier of AI.” Being brilliant at one thing tells you little about the next thing.
- Agents still miss a lot. Even after those gains, AI agents “still fail roughly one in three attempts on structured benchmarks.”
- The physical world is hard. “Robots succeed in only 12% of real household tasks.”
- The scores themselves are shaky. The report warns that benchmarks “face growing reliability and gaming concerns,” with “invalid question rates ranging from 2% on MMLU Math to 42% on GSM8K.” In plain terms, some of the tests used to grade AI have wrong questions in them.
So your laundry is still your problem. Sorry.
What regular people say
Many people haven’t tried it yet. In a separate Pew Research Center report published April 3, 2025:
- “One-third of U.S. adults say they have ever used an AI chatbot, compared with nearly all AI experts surveyed (98%).”
- Among people who had used one, experts found it more useful: “About six-in-ten AI experts who have used a chatbot (61%) say it was extremely or very helpful to them. Smaller shares of users in the general public (33%) say this.”
Our guess, and it’s only a guess, is that practice explains part of that gap. Experts know what to ask for and what to double-check. That part is learnable, and it’s what our guides are for.
How to get the “good at” column and avoid the other one
These habits come straight from our testing:
- Give it your real details. Names of dishes, the actual deadline, the email you’re replying to. Vague in, vague out. Our guide How to write a good prompt with five habits that work for any AI task shows how.
- Tell it what not to do. “Use only what I told you” and “Don’t add prices” cut down on made-up details.
- Check every number with a calculator, every fact with a real source, and every price at the store.
- Ask it to flag what it isn’t sure about. It won’t always do this on its own.
- Keep private things private. Don’t paste passwords, account numbers, or anything you wouldn’t want a stranger to read.
Try it yourself
The fastest way to see the good-at column in action is to try one guide. Most people start with dinner: How to use AI to plan a week of dinners on a budget. Or, if there’s an email you’ve been avoiding, try How to use AI to write a hard email.
Sources
- Stanford Institute for Human-Centered Artificial Intelligence (HAI), AI Index Report 2026, April 2026. https://hai.stanford.edu/ai-index/2026-ai-index-report (Technical Performance chapter: https://hai.stanford.edu/ai-index/2026-ai-index-report/technical-performance)
- Pew Research Center, “U.S. Workers Are More Worried Than Hopeful About Future AI Use in the Workplace,” February 25, 2025. https://www.pewresearch.org/social-trends/2025/02/25/u-s-workers-are-more-worried-than-hopeful-about-future-ai-use-in-the-workplace/
- Pew Research Center, “Artificial intelligence in daily life: Views and experiences” (chapter 1 of “How the U.S. Public and AI Experts View Artificial Intelligence”), April 3, 2025. https://www.pewresearch.org/internet/2025/04/03/artificial-intelligence-in-daily-life-views-and-experiences/
- PromptLantern testing notes, October 2026 (prompt tests for the guides linked above, run logged out in a free AI chat tool).
Filed under AI explained. Read more in our editorial policy and disclaimers.
Get one new AI how-to a week
Free, practical, and a little funny. One email a week, unsubscribe anytime.
We use Kit to send the newsletter, and Kit stores your email address. We only send the weekly how-to and never sell your address. After you sign up you'll land on Kit's page, then get an email to confirm. Privacy policy.