# Practice Is the Point

I've spent 30 years playing guitar and 20 building software. I only want AI touching one of them.

By Jon C. Phillips · October 11, 2026 · https://joncphillips.com/practice-is-the-point/

Every musician in my feed hates AI right now, and they hate it with a unanimity musicians have never achieved about anything else. I agree with them.

I also ship AI-written code every single day and sleep like a baby. This article is me working out why both of those things are true at the same time.

## The Unanimous Feed

**Musicians can't agree on anything.**

They'll argue for days about tube amps versus modelers, or whether a $200 cable changes your tone at all. I've watched grown adults go to war over pick thickness.

Then AI showed up, and suddenly the entire feed speaks with one voice, from bedroom players all the way up to YouTube educators with millions of subscribers.

Rick Beato has gone at it [more than once](https://www.youtube.com/watch?v=7Hph0VSdh9w), [at length](https://www.youtube.com/watch?v=YNnvWLQlQM0), and the comment sections under those videos read like a union meeting.

I follow hundreds of musicians and I have yet to see one defend AI.

**And honestly, the industry keeps handing them reasons.**

There's a whole genre of accounts now that take a real guitarist's video, run it through a model, and repost a synthetic version of the performance. Someone's actual hands, with decades of practice in them, regenerated and re-uploaded for engagement. It’s sucking sad.

The comments fill with praise for playing that never happened, while the player who sweated for the original gets nothing, not even the view.

That's stealing, and every guitarist who sees one of those videos feels it right in the spine.

Then there's [Suno](https://en.wikipedia.org/wiki/Suno_(platform)), the text-to-song generator that major labels keep suing and [2 million people are paying for](https://www.forbes.com/sites/conormurray/2026/02/26/ai-music-platform-suno-reaches-2-million-subscribers-as-industry-backlash-grows/).

Its CEO went on a podcast and explained the problem his product solves. Making music isn't enjoyable, in his words, because ["it takes a lot of time, it takes a lot of practice."](https://www.musicradar.com/music-industry/its-not-really-enjoyable-to-make-music-now-says-the-ceo-of-an-ai-music-making-platform-and-he-wants-to-help) The labels found that quote so useful they [put it in their lawsuit](https://www.musicbusinessworldwide.com/universal-and-sony-sue-suno-for-a-second-time-claiming-platforms-v6-models-are-the-fruit-of-the-same-poisoned-tree/).

He described the entire point of music and called it the obstacle.

So yeah, I get the anger. Thirty years of practice lives in my hands, and I don't want a model wearing it like a costume.

And to be fair, plenty of developers feel exactly the same way about their own craft. They spent decades learning to write good code, and now a model spits out a working version in seconds. I'll get to them, because that grief is real too.

## Meanwhile, at My Desk

Here's the uncomfortable part. While my feed burns, I spend my days directing AI through codebases, and I feel nothing but gratitude.

And I'm no outlier, I'm roughly the industry average now.

| Who | How much code AI writes now |
| --- | --- |
| Google | [75% of new code](https://www.secondtalent.com/resources/how-much-software-written-by-ai/), AI-generated and engineer-approved |
| Working developers | [47% by agents alone](https://fosspost.org/ai-coding-statistics/), with 27% still written by hand |
| Meta | [About half of all development](https://tech.yahoo.com/articles/microsoft-says-30-code-now-114621008.html) within the year, by their own target |
| Anthropic | [70 to 90% of company code](https://www.hostinger.com/blog/vibe-coding-statistics/) |

Now think about where the musicians post their complaints. Instagram is Meta, and YouTube is Google.

The video essay about AI destroying music gets uploaded through transcoding pipelines and ranked by recommendation systems, inside an app where most of the new code came out of a model.

**The protest travels on the thing being protested, and nobody notices.**

And rather than hypocrisy, what that reveals is exactly where the line sits.

## Nobody Listens to Code

Nobody has ever opened an app to admire the controller logic. No one hums a database migration. In twenty years of shipping software, not one user has ever asked me how the code felt to write.

Code has never been consumed as itself. People experience what code does without ever meeting the code.

It's the wall between you and the thing you came for, and the only quality that matters to the person on the other side is whether the wall holds.

The obvious objection is that some people do read code for pleasure. There's a small priesthood that studies elegant source the way I study live bootlegs, and I've spent time in it.

But connoisseurs don't change what a thing is for. A few thousand engineers admiring a codebase makes code the product about as much as admiring a bridge's welds makes the welds the destination.

The other eight billion people drive across.

**A song is the opposite. The song is the destination.**

When you listen to a record, the human choices are the product, like the bend that lands a hair flat because flat hurt better, or the drummer rushing into the chorus because the chorus deserved it. Strip the person out and all you've automated is the removal of the point.

Which is why musicians posting on AI-built platforms aren't contradicting themselves. The code in those apps was always invisible to them, the way it was invisible to everyone, including the people who wrote it.

Caring only about what software does, while defending your own work for how it's made, is the same line drawn twice.

> The outrage maps exactly onto where the meaning lives.

Everyone already knows where the line is. They feel it instantly. They've never had to name it before, because until a few years ago nothing could fake the parts that matter.

## The Line Runs Through Music Too

Here's where I'll lose some of the feed, because the line doesn't sit neatly between code and music. It cuts straight through music itself.

Last year a "band" called [The Velvet Sundown](https://www.abc.net.au/triplej/news/velvet-sundown-liz-pelly-australia-ghost-artists-ai-music/105703542) hit roughly a million monthly Spotify listeners in weeks, with members who never existed and songs that reportedly came out of Suno.

And Deezer now says [more than half of its daily uploads](https://newsroom-deezer.com/2026/07/ai-music-exceeds-50-percent-daily-uploads-deezer/), around 90,000 tracks a day, are fully AI-generated.

But look at where all that synthetic music actually lands. It pools in background playlists and study beats, the mood wallpaper people put on without ever checking an artist name.

Fully AI tracks pull [1 to 3% of Deezer's streams](https://www.engadget.com/entertainment/music/deezer-says-ai-made-songs-make-up-44-percent-of-daily-uploads-163642921.html) despite being half the uploads. The flood is landing in exactly the corner of music that was already consumed as function.

I watched the same split happen in photography. Stock photography went first, because nobody ever bought a stock photo caring who shot it.

They needed a picture of a handshake for a slide deck :/

Wedding photographers are doing fine, because what they sell is witness, and no model can attend your wedding.

Music that works as furniture is falling first, while music that works as presence is holding. [The floor rises](https://joncphillips.com/the-slop-floor/) exactly as far as the function extends, and it stops where the human was the product.

## The Demo Problem

Now the confession, because this essay would be dishonest without it.

I write songs, and when I'm demoing one I need a lot more than guitar. I need drums, keys, maybe a string section, maybe a sax line in the bridge if the song is feeling dramatic.

I can book players for all of that, which means money, scheduling, a room, mics, and days of back and forth.

Or I can generate perfectly usable parts in an afternoon. I pick the AI parts every time, **for the demo and nowhere else**. I need to be clear about that.

The only people who'll ever hear that demo are the players I eventually hand it to. What a drummer wants from that demo is the idea, delivered fast, so he can go play it better than the machine will ever do, with the quirks and all.

The demo is just scaffolding, **it’s a sketch that exists so the real thing can get made**. Spending four musicians' time on scaffolding is a strange way to honour the craft.

But those generated parts are never going on the record. **If the song earns a proper recording, humans play it.**

A purist would tell me to book the players anyway, on principle. The purist isn't funding my demos.

Thirty years in, I've made peace with the idea that protecting the sacred part sometimes means being ruthlessly practical about everything around it.

The same rule runs through the rest of my house, room by room.

The featured image on this article came out of an image model, like every header on this site for the past year.

I also built a product that uses AI to generate Lightroom presets for photographers, and I don't lose a minute of sleep over that either. The preset was never the photograph. The header was never the article.

Lay it all out and the pattern gets almost boring.

- AI drums and strings on a demo for the band to learn from? **Fine.**
- AI drums and strings on the finished record? **No.**
- AI image as a blog header? **Fine.**
- AI image entered in a photo competition? **Hell no.**
- AI code in a checkout flow? **Nobody on earth ever came for the checkout flow.**

The question, in every single case, is whether anyone came for this part.

When the answer is no, hand it to the machine and spend your life on the parts where the answer is yes.

## What the Devs Lose

Back to the developers, because they're owed more than a footnote.

Some of them feel about their code the way I feel about my guitar. They've spent decades building taste and style, and they're watching it get commoditized in real time.

I've written before about what we [quietly give up](https://joncphillips.com/judgment-bottleneck/) when the machine does the typing, and I won't pretend that loss is imaginary.

But there's a difference, and it matters. The purpose of code survives its automation.

The app still ships and the user still gets what they came for, while the craft moves up into judgment and architecture.

The purpose of a song is the person inside it. Automate that and what comes out the other end is an empty jacket.

Programming has also spent seventy years automating itself on purpose. Compilers replaced assembly and frameworks replaced the boilerplate, with every generation of the craft existing to make the previous generation's typing unnecessary.

AI is a violent acceleration of a tradition we built ourselves. Music has no such tradition. Its labor was the art the whole time.

## Thirty Years

Suno's pitch, straight from its CEO, is that practice is the unpleasant part and the song is the deliverable.

**For code, that logic is airtight.**

The toil really did stand between people and the thing they came for, and I'll let the machine eat as much of it as it wants, starting again tomorrow morning.

**For music, the same logic gets everything backwards.**

I didn't spend thirty years practicing guitar as the entry fee for some future output.

The practicing is the relationship, plateaus and calluses included, right down to the slow afternoons spent chasing a phrase my hands couldn't reach yet.

**The thirty years were the point all along.**

So I'll keep living on both sides of this line without apology.

Tomorrow I'll direct an AI through a codebase all day, come home, pick up the guitar, and play something badly that no model will ever be asked to fix. The machines can have the walls.

The songs are ours.