I Built an LA-2A Plugin with AI (Twice)

My Smooth Dood plugin isn't available yet as it's not signed with Apple so it can't be installed via Logic plugin manager yet. Plus I want to tweak it.

Over a couple of days, with Claude doing the coding, I built a working LA-2A-style compressor plugin for Logic. Then I built a second one from the original schematics. On my first listen, I had a hard time telling mine apart from the Softube and UAD versions.

I'm not a plugin developer. I studied electronics at DeVry about 30 years ago and remember almost none of it. That was the point: I wanted to find out if a regular musician with a little curiosity and an AI could get close to a "must-have" plugin.

My angle going in: forget null tests. You'll never match a piece of hardware exactly, and you don't need to. If you understand what the box does, you can get close enough that nobody listening can tell.

The video shows the whole thing. This post is the short version of the prompts I used and what Claude did with them.

Which prompts mattered?

I didn't write anything fancy. These are the actual prompts, typos and all.

1. "we talked about making plugins. Let's make an LA2A." Claude broke the LA-2A into the parts that give it its sound: the light-based opto cell, the slow two-stage release, the tube warmth and the transformers. It wrote the plugin, tested it with test tones until the timing and ratios looked right, and put the project on my Mac with a one-line build script. I pasted one command into Terminal and it showed up in Logic.
2. "the manual: [Softube OPTO link]" I wanted to know what Softube's Opto was based on. Their manual says an early-1960s T4 opto tube compressor, so it was fair game for a three-way comparison with the free UAD LA-2A.
3. "keep an ongoing doc that describes what we're doing as I will be doing a video and showing what prompts were used." This is the best prompt I wrote. Claude kept a running build log the whole time, and it's basically the script for the video.
4. "I want this plug to have two sections: top is faithfully identical to the la/2a. bottom: the stuff you added... should have an on/off switch" Claude pushed back (nicely). It's a behaviour model, not a circuit copy, so we called the top section "Classic mode" instead of "identical." The bottom section, "My Mods," adds a Colour knob and HF Emphasis you can switch off.
5. Naming: "puns, plus usefulness." The plugin became Smooth Dood (model CD-2A), with a faceplate Claude Design mocked up. I now have names ready for a future channel strip: Rude Dood, Crude Dood, Subdued Dood, Stewed Dood. I'm 53, so yes, this is as uncool as it sounds.
6. [attached the LA-2A user's guide PDF] "build a new one, don't update the existing so I can test them." Claude read the schematics and built a second plugin, Smooth Dood SE, from the actual circuit.

What's the difference between the two versions?

So now I have two homemade versions, built two different ways.

Smooth Dood

• Built from: how an LA-2A behaves (timing, ratios, tone)
• Compression curve: drawn to match the known behaviour
• Tubes: a simple "warmth" curve
• Extras: My Mods: Colour, HF Emphasis
• Computer power: light

Smooth Dood SE

• Built from: the schematics in the manual
• Compression curve: falls out of the real resistor-and-photocell circuit
• Tubes: standard tube equations, in the real feedback loop
• Extras: rear panel R37 control, like the real unit
• Computer power: noticeably heavier

The surprise: even with the schematic, the heart of the LA-2A can't be read off a drawing. The T4 opto cell is a sealed light-and-sensor unit, and its famous slow release comes from the materials inside it. So both versions model that part by behaviour, using the timing numbers in the manual.

The schematic version did find something the first one missed. The circuit has a filter that makes the compressor react less to deep bass. You'd never guess that from listening, but it's right there in the drawing.

When we checked SE against the manual's spec sheet, it matched the gain, frequency response, distortion and release time. It fell short on maximum compression, which is about 26 dB versus the spec's 40.

What did I take away?

• You don't need the expensive version to get the sound. Understanding what a plugin does got me most of the way there.
• Ask the AI to keep notes. The running log made the video easy and keeps everything honest.
• Push back when it overpromises, and let it push back on you. "Classic mode, not identical" is a more truthful label, and better marketing too.
• Plain prompts work. Every prompt above is something you could type.

Watch the video to hear the blind A/B and see if you can pick out the homemade ones. The full build log, with every prompt word for word, is linked below.

FAQ

Can you build an LA-2A plugin with AI?

Yes. With Claude writing the code, I had a working LA-2A-style compressor running in Logic in a couple of days, and a second version built from the schematics shortly after. I'm not a developer.

Does the homemade plugin sound like the Softube or UAD LA-2A?

On my first listen I had a hard time telling them apart. The video has a blind A/B so you can try for yourself.

Why can't you build an exact LA-2A from the schematic?

The T4 opto cell is a sealed light-and-sensor unit, and its slow release comes from the materials inside it. Both of my versions model that part by behaviour, using the timing numbers in the manual.

How close is Smooth Dood SE to the real spec sheet?

It matched the gain, frequency response, distortion and release time. Maximum compression came up short at about 26 dB against the spec's 40.

Bottom line

Understanding what the box does gets you most of the way to the sound, and plain prompts are enough to get an AI to build it. If you want to hear the result, the video has the blind A/B. The compressors I normally reach for are on the gear page.