Vibe Code Brief, October 24: Epic MegaJam Makes AI Optional After Backlash

Today's brief runs four deep: Epic walked back mandatory AI in its own game jam within hours, a one-click hosting platform announced it is going away, a new local coding model arrived, and Stanford put vibecoding on the syllabus.

1. Epic will pay $65,000 to vibe code a game, and now you don't have to use AI

Image: Epic Games, via 2026 Epic AI MegaJam

What happened: On October 22 Epic announced the AI MegaJam: four days, November 12 to 15, build a playable game in Unreal Engine or UEFN, solo or in a team of up to five humans plus as many AI agents as you want. $65,000 in cash prizes, plus continuation grants up to $50,000 per team. The initial FAQ said AI use was mandatory. After pushback on X, Epic admitted the mistake the same day and rewrote the rule. The FAQ now reads, plainly: "Do I have to use AI? No."

Why it matters to designers and builders: A six-figure jam built around agents and game engines tells you where Epic thinks production is heading, and the walk-back is its own lesson: even the most AI-forward company on earth can overcorrect publicly when the feedback is loud enough. If you want a portfolio piece with a deadline, four days with agents is enough to finish something real, and the prize list literally rewards design craft: there is a Render Royalty category for best visuals and a Glass Box prize for the best dev log, which is basically a design process contest.

What to do: Join the jam on itch.io before November 12, pick Unreal Engine or UEFN (not both), set up your agents and MCP before kickoff, then scope tiny and polish. In-person building happens November 14 to 15 in San Francisco with teams from Anthropic, Google and OpenAI.

2. Zeabur is shutting down after two breaches. Rotate everything, then migrate

Image: Zeabur, via zeabur.com

What happened: The one-click deploy platform disclosed its second breach in two months: an attacker used internal credentials stored in plaintext, read 3,218,965 environment variable records tied to 321,810 accounts, and the company will wind down between December 1 and December 31.

Why it matters to designers and builders: Environment variables are where your app's secrets live: API keys, database URLs, webhook signatures. If you never look at that panel, the platform's ops discipline now decides your product's security. That is a design responsibility you outsourced without noticing.

What to do: Rotate every key that ever touched Zeabur, inventory where your secrets live on any host, and move your apps before December 1. I wrote the full playbook in today's deep dive on what vibe coders should learn from the shutdown.

3. Recursive's Flint: a 40 billion parameter coding model, mostly built by AI

Image: Recursive, via Introducing Flint

What happened: Richard Socher's startup Recursive launched Flint on October 23, an open-weight, 40 billion parameter mixture-of-experts model aimed at fast, local coding. Nearly every stage of the build was shaped by Recursive's automated research system, the Eureka Machine, and the company claims it is the best-performing MoE under 50B parameters on agentic coding benchmarks.

Why it matters to designers and builders: An open model that runs on a laptop means agents that can work on your machine, on client data, without shipping it anywhere. The pipeline story matters too: if AI systems can now iterate on training methods and architecture, the gap between frontier labs and everyone else is compressing.

What to do: Watch this one unless you specifically need a local coding model today, in which case the open weights make trying it low stakes.

4. Stanford is teaching "Effective Vibecoding" for course credit

Image: Frank Schulenburg via Wikimedia Commons, CC BY-SA 4.0

What happened: The Stanford Daily reports that Stanford's CS department rolled out CS193V, "Effective Vibecoding", which teaches students to get an LLM to build software without reading or writing the code, and CS193T, "Thinking With AI", a no-prerequisites course on agentic workflows that the vice provost personally recommended at convocation.

Why it matters to designers and builders: When a top CS department formalizes vibecoding, the skill being taught is not prompting, it is judgment. Keith Schwarz, who leads the course, says students still need to evaluate costs, maintainability and whether software actually works. Verifying output quality is the exact skill designers already train every day, which makes this the friendliest academic turn imaginable for our community.

What to do: Watch the syllabus when it surfaces. If Stanford says evaluation is the teachable core of AI-assisted building, expect hiring bars and client conversations to shift the same way.

Our take

The through-line this week: everyone is suddenly negotiating how much AI is mandatory. Epic tried to require it and backed down in hours, Stanford is teaching people how to use it responsibly for credit, and Zeabur proved that no amount of AI speed saves you when the boring ops layer fails. The professional skill is increasingly not "can you generate it", it is "can you evaluate it". That has been a designer's job description all along.

Building something from today's news? My number one pick for AI app builders is Base44. Want practice reps first? Try a Daily Vibe Code prompt.

Related on NewForm

• A Vibe-Coded Site Got Hacked 6 Days After Launch. It Was Not the AI's Fault
• United's CEO Vibe-Coded a Betting App Between Gym Sets. Here's What That Means for Designers
• OpenAI Just Put Up to $500 a Month of API Credits Inside Your ChatGPT Subscription