This is a topic into which we can accumulate observations and best practices.
I’m getting mostly underwhelming and unpredictable results from using the Playground prompt environment to build effects.
I suspect others might also be having a similar - less than successful - experience.
So, let’s work together to improve and find more joy!
I have a LOT of experience working with ChatGPT, Claude, Gemini and CoPilot chat-propmt environments. I know how to create prompts that to minimize ambiguity and avoid hallucinatins.
But the Playground is a new beast.
What are you guys finding that “works”? That is, what helps bend results to success?
What are you guys finding that “fails”? That is, what causes results to fail?
One approach that seems to work well is using the generative suggestions to help shape an effect.
Start with a fairly open-ended prompt, such as: “Give me ideas for a reverb that does X and X.” The system will usually provide three different suggestions.
You don’t have to choose just one. Instead, take the best parts from each idea and combine them. For example, you could say, “Combine the sound character of suggestions 1 and 3 with the controls from suggestion 2.”
You can also experiment with chaining ideas together—for example, running suggestion 1 into suggestion 3 and adding a series to parallel blend knob to mix between them.
I’m not a programmer, but I’ve worked professionally in software marketing almost all my career, so I’ve applied a software process to my approach to building ENDLESS prompts.
1st Layer - Ideation.
Getting down the ideas, the focus. Playing around with what could be/should be, need not be. The point is to have fun exploring ideas, and discovering.
2nd Layer - Concept Consolidation.
Taking the ideas and creating an MRD - Marketing Requirements Doc. That is, selecting from the smorgasbord of ideas to select a coherent functional set. The MRD is formal description that is designed to be shared with an engineering team. “This is what we want to create. Here are the key capabilities, UI, use cases…”
3rd Layer - Engineering Outline
This is where I take the MRD and use it to scope out what will be needed from an Engineering POV. Signal chain workflow. Components / computations that will be needed.
4th Layer - Engineering Spec
This is where the engineering outline language gets turned into building blocks. The goal is to move from probabilistic language to deterministic mathematically precise tables, values, and formulas. I move to eliminate LLM interpretation by building out the maths needed for various signal treatments.
Each layer I’ll do with ChatGPT, or Claude or Gemini. Each layer will consult with the Endless SDK, to confirm functionality and capabilities.
Also, I take special time to search for and eliminate inconsistencies, ambiguities, and anyplace in my end prompt where Instruction A might run at cross purposes with Instruction B - as these cause LLMs problems and result in what we commonly call hallucinations.
So, what I provide to the Playground LLM has gone through some thought. And the results I’ve achieved are very hit or miss. Indeed, the exact same prompt issued on different days will yield different results. It’s quite interesting on one level. Frustrating on another.
I know the effects I’m describing at novel. That’s part of my process - exploring to see if something already exists. But, alas, David Pedal Empire is off to a bumpy start.
Anyway, after many hours and many iterations, I’ve really not successfully achieved any of the effects I was setting out to make. So, after initial irrational exuberance and enthusiasm, I’ve scaled back my energy and expecations.
I can share more if anyone is interested.
ENDLESS is a beta product. I am a beta product user. Or, as the common language in software development goes, I am an “design partner.” 
I guess in reality, this thread is kind of a beta feedback thread - if folks want to turn it into that.
OTHER…My wishlist:
Tried again tonight to make an effect that I would care for.
Made no headway at all. 
I started several new sessions and each time the Playground LLM indicated it was already working on a delay, yet what I was asking it to do was nothing to do with delays.
Here’s an exchange:
PLAYGROUD: Something went wrong.
The request couldn't be completed due to a system error.
Please try again. If the issue persists, it may require further investigation on our side.
DAVID: what happned?
PLAYGROUND: Nothing “broke”—you just switched topics. So far, this workspace was about a Delay effect, and your new spec is a separate multi‑filter movement effect (“Filter Freak”). To proceed cleanly, do you want to replace the current Delay in this workspace with Filter Freak anyway?
1) Start a new session for Filter Freak (recommended)
2) Replace the current effect here with Filter Freak
After three times starting new sessions and seeing the same error come back, and having spend 2 hours building a short clean prompt, I gave up.
So, I tried something compeletly different. And way simple: A fuzz.
Here is the exchange:
"
Untitled effect
ID: 25620_6a656f6398a8e
Make a fuzz
What kind of fuzz do you want?
1) Vintage wooly sustain — Warm, classic fuzz with lots of sustain.
2) Aggressive gated splat — Spitty, gated fuzz with tighter cleanup.
3) Octave-ish synth fuzz (more harmonic) — More harmonic fuzz with an upper-octave blend.
Reply with 1, 2, or 3.
DAVID: 1
PLAYGROUND: Something went wrong.
The request couldn't be completed due to a system error.
Please try again. If the issue persists, it may require further investigation on our side.
DAVID What happened?
PLAYGROUND: Something went wrong.
The request couldn't be completed due to a system error.
Please try again. If the issue persists, it may require further investigation on our side.
My Observations:
Seems the Playground is currently locked to a task?
Overall - I have not created a single interesting new effect via the Playground LLM. I have tried 8 different effects. Each of them with multiple prompts.
Not sure if the Playground LLM is tied to an old / version of ChatGPT or Claude or Gemini - and that might mean it’s capabilities to understand / execute are not evolving. Or perhaps for various reasons its capabilities have been reduced?
Or perhaps, the only real way to create an effect is via direct coding and the SDK - which is already a cool thing, for those who can do that. But that’s not me.
So, while the idea behind the endless is fun and exciting and appealing. The actual operationalization and what I’m getting from it is not worth the effort I’ve put into it. Which at this point is quite a bit.
That said, I think I’ve tried some cool ideas for what would be interesting pedals.
Perhaps someone else would like to have a go and try building them?
Hi David, I’ve been having exactly the same problem for a few days now; I get this error message even with the simplest effect prompts. I think Playground is broken. I’ve contacted support but haven’t heard back yet. Best regards.
Well, I tried again tonight. Didn’t create anything I would want on my pedalboard.
Gotta say, I’m not having any luck with creating anything that makes me feel the effort is worth the frustration.
Have spent easily more than 60 hours in various ways with concepts, definitions, prompts and updates/variations based on what I get back.
Based on the early plates created by the Polyend team, the DSP can do things. I’m lead to believe directly programming via the SDK is one thing, but creating via the Playground LLM is not very effective - at lest for me so far.
Anyone care to share a Playground Prompt they used to create an effect they liked and kept (and that isn’t simple dirt, delay, modulation or reverb - or combination of those)?
I haven’t given prompting a go yet. Maybe i should at some point…
If i do, i’ll make sure to share my experience 