Write the artist brief

This course follows one independent single from a forty-second voice memo to a reviewed release packet. The packet holds the brief, labeled sketches, a listening decision, a track sheet, release copy and a list of open questions.

In a session, the take sheet is the log of every take, and someone circles the good one. That circle is the craft. The course calls the habit call the take: a person makes the decision, and the tool doesn't.

AI can now draft and check almost everything around a song. It can't decide what the song should make someone feel, whose voice belongs on it, or when it's ready to leave your hands.

There are two easy ways to lose the record:

• Lay out. You sit this part out and let other people decide what the new tool becomes.
• Reach for the preset. You let the machine's first pass ship because it sounds finished.

Example: an AI mastering pass makes the demo sound glossy, and the team uploads it without checking it against what the artist asked for. Nobody called that take. The preset did.

Leaders in the business are already taking AI seriously. When you hear their views, also note what each of them gains from AI.

Jean-Michel Jarre was president of CISAC, the world confederation of songwriters' and composers' collecting societies, for seven years until 2020. At the World AI Cannes Festival he told artists "we have to wake up". He meant treating AI as a positive disruption while staying alert to intellectual-property issues.

Robert Kyncl, CEO of Warner Music Group, describes the current state as "the total wild, wild west": many models of uneven quality, and no settled rules. His tie: in November 2025 Warner settled with the AI music companies Suno and Udio and licensed them, so Warner profits from licensed AI music.

Timbaland co-founded the AI company Stage Zero and is reported to advise Suno. He describes how he works: he brings his own material, trades ideas with the tool in rounds like a ping-pong rally, and does the final assembly himself. He says the work stays his.

All three are worth hearing. None of them is a neutral observer.

Harvey Mason jr., CEO of the Recording Academy, gives creators four steps: "understand, adapt, advocate, and compete."

He tells a story about his father, a session drummer, who was shaken when a Linn LM-1 drum machine showed up at a session. By Mason's account, his father took one home and learned to program it better than anyone else, which extended his career instead of ending it.

The practice for you: before you trust AI with the real single, run an understand phase on a sample song. Give it notes for a song that doesn't matter, and watch what it adds, smooths over or gets wrong.

Two songwriters put AI in a longer line of tools. Poo Bear calls AI exciting and scary at once. He says to "get out in front of it" and use it to save time while you develop a sound of your own. Skylar Grey compares it to Auto-Tune and computers, which people also complained about, and says there will always be songs where you can tell "a person wrote it."

For your brief, that means deciding which parts of the song have to come from a person.

Two sets of figures show how the field has changed. Each one comes with a limit.

Deezer (July 2026): about 90,000 fully AI-generated tracks are delivered to the platform each day, which is more than half of its deliveries on peak days. Yet those tracks draw only 1–3% of streams. The limit: this is one platform, measuring with its own detection tool.

Water & Music survey for Moises (March 2026): the survey covered 1,525 musicians. Among the 1,021 who used AI, 71% used it for stem separation, meaning splitting a recording into parts such as vocals, drums and bass. Only 24% used it to generate whole songs, the least common use. The limit: Moises, an AI music-tools company, commissioned the survey, and most respondents were Moises users.

The fair reading is that a flood of generated tracks exists, but listeners mostly aren't streaming them, and many musicians use AI around the song instead of in place of it. That is how this course uses it too.

An artist brief is a short document that turns rough notes into direction everyone can work from. A working brief has four parts:

1. Creative direction in plain words. A song about leaving your hometown without regret, and missing it anyway.
2. The feeling it should leave. Warm and a little unresolved.
3. References described by what you hear in them, not only by name. Drums like Sunday-morning records: brushes, low in the mix.
4. The fixed element, the one thing the song can't lose.

In the course single, the fixed element is a first verse of only voice and guitar, recorded close enough that you hear the breath before the band comes in.

Name it explicitly. Anything that evens out a record can flatten that moment without anyone noticing: a compressor, a mastering preset, a helpful AI arrangement that adds a pad under verse one. If the brief doesn't say "voice and guitar only, breath audible," nobody has a written reason to object.

Here are the sample notes for this exercise:

1. Song's about leaving my hometown and not regretting it, but missing it anyway.
2. Verse one is just me and the guitar. I want you to hear me breathe before the band comes in.
3. Drums like Sunday-morning records, brushes, not loud.
4. Bridge feels flat. Want something that lifts.
5. Voice memo attached, 0:40.

A chat assistant tends to fill gaps with confident guesses. Your prompt should stop that:

Using ONLY the numbered notes below, draft an artist brief with:
creative direction, intended feeling, references (described by
sound), and the fixed element. After each line, cite the note
number it comes from. Do not invent tempo, key, dates or
references. List anything the notes don't cover under
"Gaps", and list any assumptions you made separately.

Notice that the notes give no tempo, no key and no release date. A good draft lists them as gaps. It doesn't make them up.

Give each line of the draft one label:

• From notes (note #): the notes support it directly.
• Assumption, [person] confirms: it's a reasonable guess, so name who must confirm it. The artist confirms intent, the producer confirms arrangement, and the manager confirms release facts.
• Not supported, remove: nothing in the notes backs it, and it isn't needed yet.

Then read the draft line by line with the notes beside it. Don't skim for tone. For each line, ask: which note says this? If the assistant wrote "bridge modulates up a key," the notes only say the bridge should lift. The direction "bridge should lift" is from note 4, and "modulates up a key" is an assumption the producer and artist confirm.

This check is yours to do. The assistant can draft the brief, but deciding what's true about the artist's intent is part of calling the take.

Once the brief checks out, write down how you made it, so you or someone else can repeat it on the next song. A method has four parts:

• Inputs: the artist's notes and the voice memo.
• Prompt: the exact text you used, saved word for word.
• Checks: every line cited or marked, the line-by-line read against the notes, and each assumption given an owner.
• Output: the brief saved as a document (for example Single_Brief_v1.docx), with the gaps and assumptions listed at the end.

A written method can be repeated the same way next time, it shows where a mistake came in, and it can be handed on. In the Intensive, a proactive agent (an AI that carries out recurring tasks for you) can follow this method. It can only do that if the method is on the page and not in your head.

Next lesson, you'll use this brief to write prompts for contrasting directions of the bridge and label each sketch.