AI
Getting started with AI without making it your whole personality.
People ask me where to start, so this is the page I send. The tools I'd start with, what each one is actually for, where to learn, and what I wouldn't spend money on yet.
You do not need all of this. Almost nobody does. Pick one or two things and actually use them, which is the part everybody skips.
Written for people in music and creative work, because that's who asks me. Almost none of it is specific to that.
If you only do three things.
- Make an account with ChatGPT.
- Make an account with Claude.
- Use one of them on real work every day for two weeks.
Pick ChatGPT if you want the broadest general-purpose one.
Pick Claude if you spend your day writing, reading, or working through long complicated material.
Either one on its own is enough for the first month. The third week is where people quit, and it is the week it starts being worth it.
The assistants.
This is the decision that matters. Everything further down this page is optional and this is not. All three do most of the same things; the differences below are the ones you would actually notice.
The general-purpose one. Questions, documents, spreadsheets, drafting, images, and thinking out loud.
Best for. Anything, roughly. It is the widest of the three and the one with the most things bolted onto it.
Watch for. It knows nothing about your business, your artist or last week unless you tell it. Most of the bad answers people show me are missing context, not broken software.
The one for long reading and writing. Documents, drafts, and complicated things across a lot of back and forth.
Best for. Writing that has to sound like a person wrote it, contracts and decks you need read properly, and any problem that takes twenty messages rather than two.
Watch for. It will still be wrong sometimes, and it will be wrong in complete sentences. Fluency is not accuracy.
Google's. Sits inside Gmail, Docs and Drive, and reaches Google's image and video tools.
Best for. People whose entire working life is already in Google Workspace, and anyone who wants the video tools further down this page.
Watch for. It follows precise formatting instructions less reliably than the other two. Fine for a draft, less fine for something with a fixed shape.
Everything else, by what you're trying to do.
Grouped by the job rather than the technology, because nobody wakes up wanting an image generator. They want a poster.
None of it is required reading. Skim to the row that matches a problem you actually have, and ignore the rest until you have one.
Research and reading
Two different jobs that get confused. One searches the web and shows its sources. The other only reads what you give it.
PerplexityStart here
Search that answers the question and cites where each part came from.
Best for. Getting up to speed on a market, a company or a person before a call, with links you can check.
You upload the sources and it answers only from those, with citations back to the page.
Best for. A stack of contracts, a year of trade press, or forty PDFs nobody has read. It will not wander off into the open web.
Images
The three worth knowing do different things. Look at what you actually need before picking.
Image generation aimed at look and mood.
Best for. Mood boards, treatment visuals, and getting a feeling across before a shoot exists.
Image generation that handles text inside the picture.
Best for. Anything with words in it. Posters, mockups, tour art comps, signage.
Adobe's, trained on licensed and public-domain material, and wired into the rest of Creative Cloud.
Best for. Commercial work where somebody is going to ask what the images were trained on. That question has an answer here.
Video
Moves faster than any other category on this page. Treat anything you read about it, including this, as roughly current.
Google's video tool, built around their Veo generation and shaped like a filmmaking app rather than a prompt box.
Best for. Shots with sound, and stringing generated clips into something with a beginning and an end.
Video generation plus the editing tools around it, with control over camera moves and consistency.
Best for. People who already edit and want handles rather than a slot machine.
Video generation with multi-shot and lip-sync work.
Best for. Getting a lot of usable options quickly when you are still deciding what the thing is.
Voice and music
The category where this business has the most at stake, which is exactly why it is worth understanding rather than avoiding.
Voice generation, dubbing into other languages, and transcription.
Best for. Scratch voiceover, translating a piece of content for another market, and turning interviews into text you can search.
Watch for. Cloning a real voice is a rights question before it is a technical one. Get that answered first.
Generates finished songs from a text description.
Best for. Understanding what your artists and writers are actually up against. Spend ten minutes on it and you will have a far better argument than somebody who has only read about it.
Watch for. This is the one people in music have real and reasonable opinions about. I am putting it here so you know what it does, not to tell you what to think.
Building things
You do not need this to get value out of the page. It is here because a surprising number of non-engineers now build small internal tools, and it works.
Runs in a terminal and writes, reads and changes code across a whole project.
Best for. Building the small piece of software nobody was ever going to prioritize.
A code editor with the assistant built into it.
Best for. Anyone who wants to see the files they are changing rather than work in a terminal.
Suggestions inside the editor your engineers already use.
Best for. Teams already on GitHub, because it is the least disruptive way in.
Wiring things together
Last, deliberately. This is where people waste the most money soonest, by automating a process they have not written down yet.
Connects the tools you already pay for so something in one causes something in another.
Best for. A handoff you do by hand every week and could describe to a new hire in two sentences. If you cannot describe it, do not automate it.
Start on work you already understand.
This is the only piece of advice on this page I'd fight about, and it matters more than which of the tools above you pick.
Most people start by asking it something they know nothing about, which is exactly backwards, because they have no way of telling whether the answer is any good. Start somewhere you're already an expert. Your format, your market, your artist, your own last campaign. You'll know within a minute whether it understood the question, and you'll catch the moment it says something confident and wrong.
That moment is the whole education. Everything after it is practice.
Things to try this week.
You signed up. Now what. None of these need a second subscription.
Writing
- Get a rough draft out so you have something to argue with.
- Paste in a forty-message thread and ask what was actually decided.
- Ask it to make the case against what you just wrote.
Research
- Have it explain something you have been nodding along to for a year.
- Ask what the strongest argument on the other side is.
- Turn a vague worry into five questions you could go and answer.
Numbers
- Upload a spreadsheet and ask what changed.
- Ask what looks unusual, then ask why it might not be.
- Ask what it would need to know before any of that means anything.
Work you keep redoing
- Describe a task you do every week and ask where the time goes.
- Have it write the instructions you would hand a new hire.
- Turn something you rebuild every time into a template.
Creative
- Push one idea out into ten and throw away nine.
- Draft a treatment you can react to.
- Ask what is derivative about a concept before somebody else tells you.
How I'd spend the first month.
You don't need this. Some people want a shape to follow, so here is one.
Week 1
Use one of them every day, on real work. Ask follow-ups. Tell it when it is wrong.
Stop treating it like a search box.
Week 2
Give it your own material. A spreadsheet, a long thread, notes from a meeting, a document nobody has read properly.
See how much of the answer was really about context.
Week 3
Build one small thing. A set of instructions you paste in every time, a template, a one-page tool.
Get out of the chat window.
Week 4
Take the thing you do every week and make it shorter. Not the most impressive one. The most repeated one.
End up with something you would miss if you turned it off.
Where to learn.
Three, and all three are free training from the people who build the things. That is not a coincidence. The official material is better than most of what is sold on top of it, and it does not have anything to upsell you.
OpenAI's own training.
Short, practical courses about using ChatGPT for actual work. Start here if you picked ChatGPT.
Anthropic's own training.
Same idea for Claude, and it goes further for anyone who wants to build something later.
A university course, originally from Finland, taken by a couple of million people.
The one for understanding what this technology actually is, rather than which buttons to press. Worth it if you want to stop nodding along in meetings.
No YouTube list. There are people worth watching and I'm not going to hand you four names I don't follow myself.
If you work in music.
What I'd try first depending on the seat you're in. None of it is exotic. Most of it is the paperwork around the work, which is exactly where the hours go.
Management and artist teams
- Turn a scattered campaign into one document everybody can argue with.
- Background on a market, a promoter or a brand before the call.
- Draft the announcement six ways and pick one.
- Write down the process that currently lives in one person's head.
Labels
- Pull a readable debrief out of a campaign that lived in forty emails.
- Compare two things that were reported in different shapes.
- Summarize a week of trade coverage into what actually matters.
- Draft the brief so the meeting starts somewhere.
Radio
- Show prep on a morning where there was no time for it.
- Get up to speed on a story before you have to talk about it.
- Turn a long piece of industry reading into something usable on air.
- Write the recurring paperwork once and stop rewriting it.
Creative
- Treatments and concepts you can react to instead of staring at nothing.
- Reference boards, shot lists, and a schedule that survives contact.
- Edit notes turned into something an editor can action.
Promotion and touring
- Route history and market notes pulled into one place before a routing call.
- The same announcement rewritten for fifteen markets without losing the point.
- Settlement and advance paperwork read properly instead of skimmed.
Anybody
- The email you have been avoiding for three days.
- The deck somebody needs in an hour.
Before you paste anything in
Unreleased music. Contracts. Anything under an NDA. Artist data you were given rather than own. Personal information about anybody. Numbers a partner handed you under terms. Treat all of it like you were emailing it outside the company, because in a real sense you are.
Business accounts on these tools carry different terms from personal ones, and the terms are the entire point. Check your own agreements before you decide what is fine, not after.
You can get most of the value without any of this. Take the names out. Change the numbers and keep the shape. Ask about a made-up version of the same situation. It works nearly as well and then you never have to think about it again.
Where it gets dangerous.
The easy part is getting an answer. The hard part is knowing whether the answer has earned the right to change a decision.
A plausible answer and a correct answer look identical on the screen. There is no tone of voice for uncertainty, nothing that reads as a guess. It is equally fluent when it knows and when it is filling in.
Uploading a spreadsheet does not make the story it tells you about your business true. It can see what moved. It cannot see that the number moved because a release slipped, a station changed hands, or somebody left. Two things going up together is not one causing the other, and it will happily write you a paragraph that assumes it is.
Missing context is the failure I see most. Not the tool being stupid. The tool being asked a question that was missing the one fact that changed the answer, and answering anyway.
So check what you would have checked if a person had told you. Ask what it would need to know to be sure. Ask it to make the opposite case. And notice that the people getting the most out of this are the ones who knew the most before they started, which is the opposite of how it gets sold.
What you already know is the thing that makes any of it work.
What I use.
Separate question from what is worth knowing about. Most of what is on this page I think you should know exists. This is the shorter list of what I have open.
Shorter than you were expecting, probably. I've tried most of the rest of this page and kept almost none of it, which I think is the normal outcome and nobody says so. A tool I open twice and forget is not part of a stack, it is a browser tab.
What I wouldn't pay for yet.
Not because any of it is a scam. Because it is the wrong order, and you can't tell whether you need it until you have used the basic thing for a month.
- A subscription to everything that launches. You will not use four of them.
- Prompt packs. The good version is twenty minutes of your own notes.
- An app that turns out to be a thin layer over an assistant you already pay for.
- Automation, before you can describe the workflow on paper.
- Anything sold as an employee. It is software, and somebody still has to check it.
- Anything promising a strategy without knowing anything about your business.
From asking questions to building things.
There is a ladder, and most of the noise you hear is people on the top rung selling to people who need the second one.
- Ask it questions.Where everybody starts. Fine to stay a while.
- Give it your documents and your data.The step that changes the answers most.
- Reuse the instructions that worked.Where it starts saving real time.
- Connect it to the tools you already use.Once the work is genuinely repetitive.
- Build small systems around work that repeats.This is a project. Treat it like one.
Plenty of people get everything they wanted at two. If somebody is selling you five before you have done two, that is a sales decision rather than a technical one.
Why I'm the one writing this.
I came at this from the other direction. Fourteen years in radio, then years of creative work around artists, and data and software after that. I got here because of specific things I watched not work, which is why this page is about decisions and workflows rather than which tool is winning this month.
What I care about is whether a thing changes the actual work. Most of it doesn't. Some of it does, and the difference is worth knowing.
What I build now is over here, and the twenty years before it are here.
What to do next.
Scroll back to the top and do the three things. Everything else on this page keeps.
If you're trying to work out where this actually fits into something you're working on, tell me about it.