Chad@chadbailey59April 15, 2026
Prompt engineering is dead. Long live prompt engineering.
In a non-trivial, open-ended agent application, you will see three LLM behaviors beyond the agent just doing what you want.
1. Whoah. Amazing "figuring it out" lateral thinking. We have AGI.
2. Um, the model just did something that was way too smart. I didn't want that.
3. Wait, why can't the LLM do this thing that seems super-basic to me?
Chad's video below is a great example of (2). The ship name Chad wanted wasn't available. But the agent went ahead and purchased a ship using a place-holder name while asking Chad to provide an alternative name.
In this case, the LLM being too smart/too quick doesn't cause any problems. But sometimes it does. We have a bunch of pulling-our-hair-out examples of this. Like trading a really expensive ship for a starter ship when the user's instructions about upgrading are ambiguous. Come hang out in the gradient-bang channel in the @pipecat_ai Discord for examples!
It's impossible for a prompt to cover every corner case in an open-ended application. The Gradient Bang game is a great, low stakes, test case for experimenting with this prompting dynamic. We use a lot of few-shot examples, try to be precise in the tool descriptions, and built a progressive skills loading system so the LLM can pull in detailed "strategies" for specific situations.
The core voice agent prompt is about 7,000 tokens.
You can see all the prompts (the whole game is open source).
Chad@chadbailey59We built AGI in a MMORPG with @pipecat_ai , and I'm just over here using it to give spaceships stupid names
Gradient Bang prompts are here[1]
Play the game: https://t.co/YgTkz1UakA
Sub-agents in (latent) space!
We’ve been working on a side project.
As far as I know, this is the first massively multiplayer, completely LLM-driven game. Come play Gradient Bang with us. See if you can catch me on the leaderboard.
This whole thing started because I wanted to explore a bunch of things I’m currently obsessed with, in an application of non-trivial size, that felt both new and old at the same time.
So … a retro-style space trading game built entirely around interacting with and managing multiple LLMs. Factorio, but instead of clicking, you cajole your ship AI into tasking other AIs to do things for you.
Some of the things we’ve been thinking about as we hack on Gradient Bang:
- Sub-agent orchestration
- Partial context sharing between multiple LLM inference loops
- Managing very long contexts, and episodic memory across user sessions
- World events and large volumes of structured data input as part of human/agent conversations
- Dynamic user interfaces, driven/created on the fly by LLMs
- And, of course, voice as primary input
If you’ve been building coding harnesses, or writing Open Claw agents, or doing pretty much anything that pushes the boundaries of AI-native development these days, you’re probably thinking about these things too!
This is all built with @pipecat_ai, the back end is @supabase, the React front end is deployed to @vercel, and all the code is open source.
