Symphony Writings

We write what we find.

Letters from the founders on artificial intelligence, agent design, and the operations layer of music.

The Thesis

The bet, stated plainly.

Symphony

2026

AI is rebuilding every industry. In music, nearly all of that energy points at the product: generating the song, the voice, the master. However you feel about that, one thing is true: releasing music has never been easier, and cutting through has never been harder. The craft is still the craft. A great song still takes everything you have. But the song alone was never the whole business. What makes music a living is everything around it: getting it heard, getting it sold, getting it paid. The operations and the promotion. Almost nobody is pointing the AI there.

So that is where we point the AI.

We’ve run marketing and release strategy for music artists since 2016. Songs on the Billboard charts and Grammy stages. That decade is the training data. Most AI in music is trained on audio. Our agents are trained on what moves the business.

A record label, written down as software. Each agent does one job a label used to staff. Maestro conducts the team. Maude keeps the business accountable. Simmy runs the media. Cue keeps every release on track. Hire one, or hire the whole roster.

Personal computing started in a locked room run by specialists, then it landed on every desk and in every pocket. A record label is that locked room now: the staff, the budgets, the relationships, the hundred small operational moves that decide whether a release lands. Symphony puts that room in your hands and shows you the next move. Make the song in Suno or anywhere you like. The business around it was always the part that was missing. Running a music career should feel less like paperwork you keep losing to and more like a game you can win.

Anyone with a song now has the label.
Anyone running a label now has the staff.

The letters

01

Letter 01 · Memory

Why every agent has a memory file

Every artist project starts at level one. No save state, no memory card. We decided that was the fundamental problem.

Chuka Chase

May 2026

Read the letter →

02

Letter 02 · Architecture

Why we gave each agent one job and refused to add more

The first version tried to do everything and it was a disaster. The fix was scope: one agent, one job, genuinely expert.

Megh Vakharia

May 2026

Read the letter →

03

Letter 03 · Self-improvement

The moat that builds itself

Everyone has the same model access. The question is what your system learns between sessions, and who that learning belongs to.

Chuka Chase

May 2026

Read the letter →

04

Letter 04 · Catalog valuation

What is a catalog actually worth?

Most artists have never seen a real number on what they've built. The data to compute one is sitting in their inbox.

Megh Vakharia

July 2026

Read the letter →

05

Letter 05 · Design

Why we named every agent

Maestro. Maude. Simmy. Cue. Echo. The names aren't branding. They're accountability structures.

Chuka Chase

July 2026

Read the letter →

06

Letter 06 · Interface

The whole label fits in a text message

Everyone building AI for music is building a dashboard. Artists don't live in dashboards. They live in their texts.

Chuka Chase

July 2026

Read the letter →

07

Letter 07 · Economics

Credits, not seats: why you pay for work, not access

Seat pricing charges you for permission to do the work yourself. Agents do the work. The pricing has to say so.

Megh Vakharia

July 2026

Read the letter →

08

Letter 08 · Proactivity

Software waits. A staff shows up.

Every tool an artist owns is parked until it gets opened. We built the opposite: agents with a pulse.

Megh Vakharia

July 2026

Read the letter →

Want early access to new research?

We publish when we have something worth saying. No cadence. Just findings.

Sign up free →