Real Developers, Real Questions: What We Learned at the SPC Vibe Coding Forum

What happens when you code in front of people who actually know what they're looking at

Last Thursday, Kilo’s engineer Chris Hasson headed to South Park Commons (SPC) for their Vibe Coding Forum, led by John Bohannon. SPC isn't your typical tech meetup — it's a community of founders, highly-cited researchers, and first-time entrepreneurs all eager to build the next big thing.

Chris gave a quick recap of how Kilo Code came to be and jumped straight into coding. The group voted to recreate a classic game: Snake. It started with a voice prompt:

“We’re in a brand-new repo and want to build an interactive snake game in TypeScript as a single-page app as quickly as possible... where you can use the arrow keys to move your snake around. You can eat apples in a 2D grid, and the snake should grow as it eats apples. Standard Snake rules.”

He flipped into Orchestrator mode, letting Kilo Code delegate tasks while the community watched it scaffold the project, initialize the Git repo, set up the build system, and start coding the core game structure in real-time.

While Kilo Code was taking on the code, we dove into a deep discussion.

Specs First, Vibes Second

John opened with the reality check every developer faces: "Vibe coding something de novo from a single prompt only works for the simple stuff, and you have a bad experience as soon as you go into the deeper water. The irony here is that vibe coders are rediscovering software engineering best practices — you start with a spec.”

Chris emphasized the point: "I personally ask Kilo Code to write its own plan first."

Tirumari (Kilo's support and community manager) added: "One programmer created SpecGram, a repo focused on spec-driven development that takes advantage of how customizable Kilo Code is."

John summed it up: "It sounds like your philosophy is: Okay, we're open source, so let's just open source this thing."

It’s true; we believe the community of builders knows best. This is why we take time to review dozens of issues, discussion threads, and community PRs on a regular basis. Not to mention all the chatter on our Discord!

Cross-Model Critique Made Simple

The room had no shortage of ideas for our roadmap. There was abundant frustration with the standard cross-model workflows using the standard AI coding tools: "One of the things that excites me most about Kilo Code is that it looks like you're going to make it easier for me to use my own API keys to get different models to critique each other," said John. “It's one of the most valuable things, and right now it's a pain... If I'm coming from another app, I'm copy-pasting or pointing it at a file that might not have the whole context. It just sucks. So it sounds like you guys can make this a mode, right?"

“Yes!” confirmed Chris, after which he walked through the different modes and showed how you can install the modes marketplace, which is full of community contributions. "You can make as many custom modes as you want. They can all have their own system prompts and tool access."

As Chris showed, the implementation is straightforward: "Create a new custom mode and link it with whichever provider you want to do the review."

As John saw it, "That's most of the solution right there."

Context Management That Actually Works

Another challenge from the room was about context rot: "I am constantly babysitting my big seven-phase.md file as it goes through one case at a time, having to clear context manually. Can I accomplish programmatic context clearing?"

Chris jumped in, explaining the orchestrator approach: "You start in orchestrator mode and say, 'Delegate code tasks for each phase of this plan.' When it's done, it only returns the information the parent needs." That way, the parent branch never gets polluted with the growing context streams.”

During the demo, this delegation happened in real-time. "Once it delegates into a new task, you can scroll back in history to review each task," Chris explained.

Visualizing Context Usage

The demo also sparked discussion about context window visualization. One member requested something like SpaceMonger for hard drives: "I want to see what's in this context and think, 'Oh, we don't need this. We don't need that.'"

Chris showed how Kilo Code already flags context usage: "When you hit 50% or more, we start flagging it, because that's when hallucination increases."

Building With Developers

The SPC forum reinforced something we believe deeply: the best tools emerge from real developer feedback, not boardroom feature lists. This is also the power of open source; when developers aren’t locked into a proprietary toolkit, when they can bring their own API keys and experiment with 400+ AI models, when you’re a Discord message away from their feedback and feature requests, you get a vibrant community you actually want to ship for.

SpecGram exists because someone needed spec-driven development and had the tools to create it. That's exactly why we built Kilo Code the way we did — and why conversations like this matter more than any polished demo.

We're not trying to guess what developers need. We're building alongside them.

Play our vibe-coded snake game or explore the GitHub repo to see how it all came together.

Your Turn

Join our community on Discord and follow our Github repo. We’re nearing 10K stars! And psst, once we hit that milestone, we’re celebrating it by giving everyone $10 in Al model usage credits!

Ready to try it yourself? Get started at kilocode.ai and bring your team over!

Because the best AI coding tools aren't built in isolation — they're built in conversation with the developers who use them every day.