There's a lot of noise about what AI is supposed to do. This page is about what it does now: where it's reliably useful, where it takes more effort, and where I'd hold off. I use these tools every day in my own business, and I update this page as things change.
✓
What AI does well
This list is long on purpose. More of your week probably fits here than you'd guess.
Drafting emails, proposals, and first-pass copy. It gets you from a blank page to a working draft in minutes, and it's good at matching a format or example you show it. You'll still want to edit before anything goes out.
Building simple tools, dashboards, and websites, fast! Internal calculators, trackers, simple sites. Work that used to mean hiring a contractor can now take an afternoon. This site is an example.
Watching inboxes & dashboards and flagging what matters. Once it's set up, it can read everything that comes in and flag the few things that need your attention.
Turning messy spreadsheets into clean data. Inconsistent names, stray columns, five different date formats. This kind of cleanup is quick work for these tools.
Answering questions from your own documents. Point it at your files and ask questions in plain English. Useful for policies, past projects, and finding things no one remembers filing.
Reformatting, editing and other clerical work. Renaming, restructuring, converting between formats, proofreading against a style guide.
Bringing datasets and tools together to get things done faster. It can connect your calendar, email, spreadsheets, and other apps, and move information between them so you're not copy-pasting.
Doing the above work on a schedule, or in response to some event. Any of the above can run every Monday morning, or whenever an invoice lands, without anyone remembering to start it.
!
Trickier, but doable
Nothing here is off the table. It just takes more setup, oversight, or budget.
Anything that has to sound exactly like you. Getting your voice right takes good examples, a lot of iteration, and a human read before anything goes out.
Ideation or management. It can generate options and help organize the work, but judgment calls about people and direction should stay with a person.
Heavy video or data work ($$). Doable, but the compute costs add up quickly. Worth estimating before you commit.
Building data-heavy infrastructure. Possible, but it's real engineering work, and AI assistance doesn't change how carefully it needs to be done.
Anything without occasional human-performed sanity checks. Automated systems drift over time, quietly. A periodic human review keeps them reliable.
×
Don't bother (yet)
This list gets shorter every year, but for now I'd hold off.
Work where a wrong answer is a legal problem (or a client-facing one). If a mistake means a lawsuit or a lost client, a person should stay in charge. AI can help with the work, but it shouldn't make the call.
Most "fully autonomous" promises. The demos look great, and the day-to-day reality disappoints. The systems that work today are the supervised ones.
Consumer robots. Not yet. Ask me again in a few years.
Trying it yourself? A few tips.
You don't need me to get started. These are the things I tell everyone first.
Iteration is essential. Give detailed feedback, and tell it to simplify or expand in the specific ways you find valuable.
It can write its own instructions. This is a significant time-saver, but proofread them to be sure they emphasize what you want emphasized. One prompt worth trying: "I want to do X. Interview me to build a skill I can use to do this repeatedly."
Audit your routines periodically. Over weeks, automated routines can accumulate conflicting instructions, data, and memory, and performance degrades. Reviewing on a schedule helps.
Learning is still human work. AI makes data collection and building fast, but reviewing what you know and deciding what to do next is not its strength.
your move
Not sure where your task fits?
Send it to me. I'll give you an honest read and tell you what I'd try first.