6 ways music people should be using AI right now.

I am writing this because the AI conversation in music is still stuck on whether to use it.

That debate is over. The people who are going to win the next decade are already past it.

The interesting question is how. Because the gap between someone who opens ChatGPT and types "write me a bio" and someone running agents that scrape, sort, draft, and ship every night is enormous. And it’s not a tech gap. It’s a habit gap.

Six tips. Tagged for who they’re for. Use what fits. Skip what doesn’t.

1. Match the model to the task.

Who it’s for: everyone.

There is no "best" AI. There are different models for different jobs.

Opus 4.7 is the heavy lift. Use it when the cost of being wrong is high. Building your website. Reading a publishing contract. Mapping a six-month rollout. The kind of thing where you want the model to think hard before it answers.

Sonnet is the workhorse. Faster, cheaper, still very smart. Bouncing lyric ideas. Writing the first draft of a pitch email. Iterating on a deck. Most of your day-to-day belongs here.

Haiku is the cheap, fast, repeated stuff. Tagging things. Quick lookups. Sorting a list.

If you are an artist, the move is Opus when you build the site or rewrite the bio that has to land, Sonnet when you are kicking around hooks. If you are a manager or label, Opus when you read the deal, Sonnet when you draft the email.

Most people pick one model and use it for everything. That is like using a sledgehammer to hang a picture.

Pick the model that fits the cost of being wrong.

2. Stop prompting. Make it interview you.

Who it’s for: beginners moving to intermediate.

People ask me all the time where to start with AI. This is where I started.

The reason most prompts feel flat is that you are doing all the work. You are trying to describe the answer you want before you have figured out what the answer actually is.

Flip it. Make the model interview you.

The prompt:

"I want [goal or outcome]. Interview me thoroughly to extract the ideas and intent. Ultrathink. Plan mode on."

Then sit there and answer the questions. It will ask things you would not have thought to tell it. By the time the interview is done, the brief writes itself.

For an artist, this works on your bio, your rollout, the concept for your next EP. For a manager or label exec, it works on the deal memo, the artist strategy doc, the deck for the meeting.

You stop prompting. You start being interviewed.

Time and curiosity. That is the whole game.

The brief you do not write is the one that loses.

3. Thinking partner. Not oracle.

Who it’s for: everyone.

Most people ask AI for the answer.

Operators ask it to pressure-test theirs.

That is the entire difference between someone who gets generic output and someone who gets a real edge.

You already have a take. You already know what you want to do. The leverage is in handing the take over and saying "poke holes in this. What am I missing? What would the smartest person in the room push back on?"

For an artist, that sounds like: "Here is my plan to roll out this single. What is weak about it? What would a label A&R who has done this 100 times say is wrong?"

For a manager, label, or business: "Here is my read on this artist. Argue against it. What is the bear case?"

The artists building real careers right now are using AI like a co-writer who never sleeps. The operators running real shops are using it like a junior who reads everything and is not afraid to disagree.

The answer is not the product. The argument is the product.

4. Build your own context.

Who it’s for: intermediate users.

The model is only as smart as what you give it.

Generic prompt, generic output. Every time.

The artists and operators getting the most out of AI right now have private context the model can pull from. They feed it the things only they have.

If you are an artist: your streaming numbers month over month. The captions that actually moved your fans. Your last 10 setlists. The DM threads with your fans about what they want to hear. Your old bios. Your mission statement, if you have one.

If you are in the business: roster data, pitch history, signing memos, P&Ls, win-loss reasons on the last 20 deals you tracked, your scouting notes from the past year.

You feed it your work and you ask it to find patterns. Where am I winning. Where am I leaking. What is the through-line in the songs that hit. What is the through-line in the artists I passed on who blew up six months later.

This is the work that turns AI from a chatbot into a brain you have trained.

The prompt is just the door. The context is the room.

5. Automate the boring stuff first.

Who it’s for: intermediate to expert.

Before you build an agent, find the work you already hate.

Make a list. Five things you do every week that you wish you did not.

For an artist: caption drafts, DSP pitch text, replying to fan comments and DMs, show flyer copy, end-of-week recap of what is working.

For the business: weekly reports, scout briefs, royalty statement reads, deal summaries, the same five emails you write to five different people every Monday.

AI can knock 80% of that out today. You do not need agents. You do not need to code. You need a folder of saved prompts and 30 minutes to set them up.

Do this for a month. Two things happen. You get hours back. And you start to see exactly which of these jobs are worth automating for real.

That is when you are ready for the next step.

Automate the work you already hate. The rest will tell you what it wants to be.

6. Build agents that learn your work.

Who it’s for: experts.

An agent is not a chatbot. It is a process you have taught a model to run on its own. Every day. Every Friday. Every time something happens.

The architecture is three pieces.

A soul. Who the agent is. Its voice, its posture, what it cares about, what it will not do. This is the file the model reads first, every time it wakes up.

A set of skills. The specific jobs it knows how to do, with the rules for doing them right. Onboarding an artist. Pulling stream counts. Drafting a newsletter. Each skill is its own file the model loads when the job calls for it.

The scripts. The .mjs files that go grab the data, post the thing, send the email. The hands.

A note on platforms. We run our agents through OpenClaw and Hermes, both built to run 24/7 on dedicated hardware with cron jobs, device control, and full data pipelines. That is the heavy version. If you are not ready to run infrastructure, Claude Cowork is the on-ramp. It runs on your desktop and handles tasks while your laptop is open. Different jobs, different tools. Pick what fits where you are.

At BTD, we run a roster.

A scout that watches the TikTok For You feed every day and flags artists worth a closer look. A data agent that scrapes streaming and social numbers nightly across the full watchlist. A research agent that turns a name into a brief in 10 minutes. A newsletter agent that drafts every Friday at 9 AM. Crons fire them. Logs catch them. We review the output and ship.

For an artist, this looks different but works the same. A release-day agent that handles the DSP pitch, the posts, the email to the list, the thank-you replies. A fan-engagement agent that surfaces the comments worth responding to. A weekly review agent that tells you what worked and what did not.

None of them replace the people. They run the floor under the people.

Agents do not replace operators. They give operators a floor.

The pattern.

The people getting real leverage from AI right now are not the loudest in the discourse. They are quiet. They are shipping.

They started by picking the right model for the task. They moved to letting the model interview them instead of prompting it. They learned to use it as a thinking partner, not an oracle. They fed it their context. They automated the work they hated. And the ones who went furthest built agents that run their floor.

The artists and operators who are going to win the next decade are not arguing about AI on Twitter.

They are six tips deep, and they are not slowing down.