AI

AI mastering is not good yet#

AI mastering has been stuck in not-very-good territory for years. Right now it is more likely to harm your audio than improve it. It reads a crest factor, a loudness value and a tonal weighting off well-produced reference tracks and stamps them onto your song. It knows nothing of the chain before it. Copying those statistics doesn't even work when the song is identical: match a cover version to the original's EQ, dynamics and loudness and you still don't get the original. A competent mastering engineer hears what can't be put on a chart, and can tell you that if you want the impact you've described, you'll have to sort it out in another part of the chain.

I have no objection to AI as a utility, and at some point it will probably beat humans at this. Maybe in another ten years. Likewise an AI inside my EQ just slows me down, and stem separation of old stereo recordings isn't solved: some tools are very good, most results sound bad.

AI beats a bad online course and loses to a good one#

I tested the free chatbots as a mixing coach. Some answers were right. Many were wild goose chases: a 30 Hz high-pass for a boomy kick, reverb for dry drums that needed the room mics turned up and crushed, the right answer buried in an overwhelming list. An amateur wouldn't know which bit to trust.

So AI beats a paid course from someone who doesn't know what he's on about, and there are plenty of those. It falls far short of a well-curated course from a working engineer with a coherent method. An ocean of unfiltered, raw knowledge is just as bad as no knowledge at all. See education.

AI changes the cost of code and nothing else#

For a long time LLMs were useless for serious code. Ask one to fix a bug and you'd come back to a nuked codebase and four wasted hours. The most recent models changed that. In my experience they now write code better and faster than I do, and I've been programming since the late '90s. Refusing them now is digging a tunnel with a shovel instead of a boring machine.

But code was never the hard part. Most of my earlier plugins I typed by hand, often twelve hours a day. What makes a plugin worth having is the idea, the design, the research, how it sounds and how it feels to use. That doesn't come out of an AI; AIs are profoundly unoriginal with ideas. I understand and direct the code; I just no longer obsess over implementation details, so a change to the sound or the interface takes fifteen minutes instead of a day.

AI is an amplifier#

AI amplifies whoever is using it. Put a bad DJ in charge of a club's tens of thousands of watts and the club empties. A good DJ uses the exact same rig to make it the best night in town.

Anyone can pay for the same models, just as anyone can shop at the chef's market. You still go to the restaurant for what the chef does with the ingredients. The code is free; you're paying for the ideas, the taste, the execution and the support.

Generic plugins were around long before AI#

A plugin made from one prompt and one brain cell is another compressor, another clipper, another saturator, with bog standard DSP. Textbook DSP sold at premium prices behind a flashy interface predates AI by many years. AI didn't invent slop; it made it cheaper to produce. See plugins.

So ask the usual questions. Is it new in sound or workflow, or could a stock plugin match it? Does the developer understand the code? Is there a real, accountable business behind it? Will it survive the next OS update? A plugin runs with your DAW's privileges, so know who made it before you install it.

AI as a tool is fine, AI as the end product is not#

Using an LLM to dig into genres I'd never heard of is fantastic. A Reaper script from an LLM that saves three hours of cutting audio is fine: it's still your art. Press a button and get a track out, and the AI is the end product.

Music's meaning is inseparable from its human context: the story, the struggle, the limitations, the live show. That's why "it's not about the tools, it's what comes out of the speakers" is wrong. AI music has none of that context; in my personal opinion it's amalgamated plagiarism. Outsourcing talent isn't new either: MIDI chord packs and ghostwriters got there first. There's no objective wrong in preferring it; our values just differ radically. That's part of why I don't think AI-generated music is going to take over the world overnight.