SKD Sounds

I make Audio plugins, music, mobile games and digital books (based on lessons learnt).

[SKD Sounds @skdsounds]

@Andrew_Blumson @threejs @WebGPU @ElevenLabs This is probably the coolest thing i've seen anyone build with 5.2. Hats off to you good sir!

[SKD Sounds @skdsounds]

@Berna7224 💯 this is the way

[SKD Sounds @skdsounds]

@parhelion318 Lovely!

[SKD Sounds @skdsounds]

@billyjhowell Definitely possible! I’m building some for myself, and want to make a few tracks using only the stuff I built

[SKD Sounds @skdsounds]

@codyavant Cool! I’m doing similar stuff!

[SKD Sounds @skdsounds]

Here is the full panel:

[SKD Sounds @skdsounds]

козерог-82 is a Vermona-inspired synth you can play in the browser. No install.

If 100 people want it as a VST, I'll build it. Feedback welcome.

https://t.co/0nVbXyStXw

[SKD Sounds @skdsounds]

@fadihares95 Agree! Starting is easy, but I wouldn’t even start unless i check a few things first:

[SKD Sounds @skdsounds]

@jamesckemp @threejs Very cool!

[SKD Sounds @skdsounds]

A vertical slice is the right first target, not the finish line. I shipped two mobile games as a non-programmer with AI, and I still write three docs before any code: GDD, vertical slice, and pipeline. The slice only becomes a game on a real device. Ideally also played by someone else than yourself.

[SKD Sounds @skdsounds]

@script_ing Anyone can do it these days, there are plenty of pitfalls along the way tho, and the hard part is judging whether its fun or not (especially with no audience). I wrote a post about it the other day: https://t.co/cZwMSX016t

[SKD Sounds @skdsounds]

If I started a new game tomorrow, I wouldn't open an engine. I'd open a chat.

From Part 7 of my book — what I'd do now, before the first line of code:

Step 1: Ask what's already out there
The first thing I do with any idea is explain it to a model, in plain words, and ask one question before anything else: are there games like this already? What do they do well, and what could be done better?

That answer decides whether I carry on. If there's already a proven game doing the same thing well, a solo developer is going to have a very hard time competing with it, however good the AI is. Most ideas stop here, and that's the point: stopping here costs an evening, not six months.

Step 2: Test whether anyone wants it
A model can tell you what exists. It can't tell you whether anyone wants yours, and several models agreeing with each other isn't demand either. This is the step I skipped on both shipped games. So before the design gets serious, run one cheap test with real people, for a week, with the deciding number written down before it starts.

Step 3: Argue the design through several models
An idea that survives the first question gets a proper design conversation, and not with one model. I take the same design to several, and ask each one to improve it and to find what's wrong with it. Then I take what I've learned back to the others. Each round, the design gets a bit sharper and a bit harder to knock down.

Step 4: Three documents before any code
Everything from those conversations goes into three documents, kept where every model can read them:

• GDD — what the game is, who it's for, the core loop, and what makes it different
• VerticalSlice — the smallest playable version that proves the game is fun, and the test that says it does
• Pipeline — the engine, the tools, the models and who does what, how a build gets onto the phone, and what "done" means

None of them is a page of notes from my phone. Each one has been through several rounds of different models trying to refute it. When the AI starts building, it's building from a design that has already survived an argument.

Then the setup before the AI builds anything:
• One main partner, one reviewer (different companies)
• One source of truth + a decision log
• A check that can fail loudly
• The real device from week one
• No art until it's fun

What this won't do: it won't make the game good. Nothing in this setup knows whether your game is fun, and neither does the AI.

What it will do is make the mistakes cheaper. You'll still make some of the twenty-seven. I'd be surprised if you didn't. But you'll find them in a week instead of a month, on the phone instead of in a store review, and before you've spent the money rather than after.

Full field report — 27 Mistakes I Made Building Mobile Games with AI ($15, PDF + EPUB):
https://t.co/sA5BpjAvVi

[SKD Sounds @skdsounds]

@ayyar I think AI is a great tool to help find "the fun" quicker. Judging whether or not its actually fun, is still up to us mortals imo.

[SKD Sounds @skdsounds]

@dev_bre Hell yea!

[SKD Sounds @skdsounds]

@BlendiByl Amazing work

[SKD Sounds @skdsounds]

@itsfazley @threejs @AnthropicAI @claudeai Insane

[SKD Sounds @skdsounds]

@chongdashu Inspiring stuff!

[SKD Sounds @skdsounds]

@SyntheticBeef Great idea!

[SKD Sounds @skdsounds]

@fizzychicken Not a fan of ”I one shotted this game” posts. Using ai to speed up the process is totally fine though imo. Most of the trouble for a solo dev is lack of audience / scope control / lack of marketing experience & budget

[SKD Sounds @skdsounds]

@Giulio6686 Core mechanic that’s original

[SKD Sounds @skdsounds]

@guilleflorvs Not if there isn't an audience for it. Making a community around the idea is the hard part.