[sdiy] AI code generation
Benjamin Tremblay
btremblay at me.com
Fri Jul 24 16:42:52 CEST 2026
A recent project was reading from the data and control pins of a Boss DR-202 LCD port.
I had no idea what to do, so I was pretty much the assistant who allowed Claude to read the serial port on an mcu as we tried many pin combinations.
Some wires were wrong on the MCU side. It wasn’t until 2 weeks of testing and fortifying the bus with 74HC buffers that I, the frustrated human, decided to inspect the wiring and got it fixed in 5 minutes. We had overbuilt the circuits and wrote too much defensive code, because Claude doesn’t know what needs to asserted first and foremost, and I allowed my own common sense to be snookered.
Benjamin Tremblay
> On Jul 24, 2026, at 10:11 AM, Chris McDowell via Synth-diy <synth-diy at synth-diy.org> wrote:
>
> 1 has worked fairly well for me in some cases
>
> 2 and 3 have had pretty mixed success. I had perplexity make up citations and claude did everything it could to not read the sources lol. it can't hold a whole book of context at once so the move that -has- helped is get it to build a cheat sheet so it knows where to look in the reference material.
>
> 4 I am super interested in this one here. I have done a bit of analog visibility work on recent projects, using a digital scope and instrumenting more parts of the circuit that I would have in the past. this has always been on projects where I have an stm32 on board as my AI stm32 infrastructure is pretty developed, claude will build dashboards in the browser and stream data directly via swd for display and debugging. on projects with analog and MCUs playing together, it is awesome.
>
> I have noodled a bit on trying to standardize some kind of "big analog visibility" for future projects, but nothing has felt like a complete enough idea to put on a board yet. Through that exercise I've learned that my "clips and clamps" tooling is way worse than it needs to be, so maybe just way better probes and a salae would get me somewhere satisfying.
>
> The various claude models behave very differently. Fable is incredible, Opus 4.7 and 4.8 are insufferably neurotic and often ineffective, 4.6 is still a workhorse. ymmv of course. I don't use openai as the company gives me the willies.
>
>
> Just as an anecdote for folks who are curious but haven't tried this stuff yet: I had a recent project with a raspberry pi talking to my custom hardware over USB, with claude writing code for and real-time debugging both at the same time (ethernet to the pi, stlink to my hardware). it was a not particularly novel problem but one with a big list of requirements, and I did meetings and other projects while it just chugged away in the background and essentially completed the project in a few hours.
>
> Cheers,
> Chris McDowell
>
>> On Jul 24, 2026, at 6:33 AM, cheater cheater via Synth-diy <synth-diy at synth-diy.org> wrote:
>>
>> I see some bad experiences with AI and analog electronics. Have any of
>> you tried any of the following approaches?
>>
>> 1. Rather than use an image of a schematic diagram, provide it with a
>> plain-text netlist + BOM and have it parse that
>> 2. Use a "research" type AI, which you feed documents. Feed it books
>> and papers about the kinds of things you are doing, and then ask it
>> questions about your circuits
>> 3. Point the AI to trusted resources such as "better quality" forums
>> or even this mailing list's archives in order to get it to answer
>> questions
>> 4. People have mentioned giving the thing access to be able to measure
>> some signals when they were working on DSP code. When working on
>> analog hardware, has anyone tried to give it access to AD/DA and a
>> digital oscilloscope, and build your circuits on a board that has all
>> (or very many) nodes tied to an i/o multiplexer so it can measure
>> things? (I think azonenberg's open source oscilloscope software can be
>> helpful here as the oscilloscope interface)
>> 5. I wonder if some form of modular pick and place machine would be
>> helpful for building circuits. Put all the parts on DIPs, have sockets
>> that can open and close, have the thing connect things... basically a
>> big breadboard style thing. Leave it alone in a dark room for a month
>> and do other stuff in the meantime.
>>
>> Has anyone tried spending some time on focused learning on how to use
>> AI for analog circuits? Maybe other forums might have answers, like
>> idk, eev blog or diy audio or gearspace or kvr. It would be
>> interesting to see what others come up with, I think this sort of
>> knowledge is going to be shared across silos but sharing between silos
>> only happens when someone does that on purpose.
>>
>>>> On Fri, Jul 24, 2026 at 11:31 AM Mike Bryant <mbryant at futurehorizons.com> wrote:
>>>
>>> I didn't think it would need an AI to point out that most USB audio dongles are s**t 😄
>>>
>>> ________________________________
>>> From: Synth-diy <synth-diy-bounces at synth-diy.org> on behalf of rrsounds via Synth-diy <synth-diy at synth-diy.org>
>>> Sent: 24 July 2026 08:39
>>> To: Eric Frampton <eric at ericframpton.com>
>>> Cc: SDIY <synth-diy at synth-diy.org>
>>> Subject: Re: [sdiy] AI code generation
>>>
>>> Using Claude to design audio DSP on my Mac Mini, I ran into a situation where the audio I could hear was occasionally cutting out, with some pretty bad digit-itis squwawking. I wanted to blame the DSP, but Claude couldn’t find an error in the code that would likely result in such a series of glitches. Since AI is only as good as the questions you give it, I carefully described what I was hearing and it was able to discern by probing into OS system signals that the USB driver for my cheap AES-out dongle (or the dongle itself) was probably at fault. I attached my Yellotec PUC-2 in its place, and the problem disappeared entirely. And the audio sounded clearer, too, so I guess the dongle was f*cking with me.
>>>
>>> David Reaves
>>> David Reaves Audio.
>>>
>>> Sent from my iPad
>>>
>>>>> On Jul 24, 2026, at 6:08 AM, Eric Frampton <eric at ericframpton.com> wrote:
>>>>
>>>>
>>>> Also yeah, Claude’s rate for diagnosing hardware issues is pretty bad - maybe 10% of the time it gives me something useful, but when it is useful, it’s spot-on.
>>>
>>>
>>>> This is the Synth-diy mailing list
>>>> Submit email to: Synth-diy at synth-diy.org
>>>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>>>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>>>> Selling or trading? Use marketplace at synth-diy.org
>>>
>>>
>>> ________________________________________________________
>>> This is the Synth-diy mailing list
>>> Submit email to: Synth-diy at synth-diy.org
>>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>>> Selling or trading? Use marketplace at synth-diy.org
>>> ________________________________________________________
>>> This is the Synth-diy mailing list
>>> Submit email to: Synth-diy at synth-diy.org
>>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>>> Selling or trading? Use marketplace at synth-diy.org
>>
>> ________________________________________________________
>> This is the Synth-diy mailing list
>> Submit email to: Synth-diy at synth-diy.org
>> View archive at: https://synth-diy.org/pipermail/synth-diy/
>> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
>> Selling or trading? Use marketplace at synth-diy.org
>
> ________________________________________________________
> This is the Synth-diy mailing list
> Submit email to: Synth-diy at synth-diy.org
> View archive at: https://synth-diy.org/pipermail/synth-diy/
> Check your settings at: https://synth-diy.org/mailman/listinfo/synth-diy
> Selling or trading? Use marketplace at synth-diy.org
More information about the Synth-diy
mailing list