TOP DISCUSSION

VC Funding and Games: An Uneasy Marriage

A builder currently in a16z's Speedrun program shared their experience — $1M checks, demo day exposure to 1,000+ investors in-person and 12K online, and follow-on raises enabled by the brand. Their cohort raised $2.5M total by partnering with other VCs alongside the program.

Veterans with Series A experience pushed back, calling VC funding fundamentally incompatible with game projects at scale. VCs need scalable, exitable businesses, and shipped games rarely fit that mold. The fundraising process itself can become an intelligence-extraction exercise — founders educate investors for free and get kept warm with no thesis behind a check. Multiple builders noted that a16z has shifted energy toward worldlabs and adjacent AI infrastructure, leaving fewer slots for pure game studios.

Pre-seed and accelerator-stage funding still works — small checks, fast iteration, demo day leverage. But Series A gaming rounds have dried up post-COVID, with AI infrastructure absorbing most available capital.

PROJECTS & PEOPLE

Shatterwake's narrative AI pipeline one-shotted a 16K-word quest graph

One team building an AI-native MMO detailed their backend system for procedural quest generation. The architecture supports four quest emergence paths: a narrative agent that watches dialogue and converts organic NPC requests into quests, companion-initiated missions, prop-based clue discovery, and path-based character encounters — all personalized per player and designed around hero's journey principles. The system supports eight quest types (kill, investigate, social, heist, escort, defense, craft, bring) with full mechanical tracking for completion, XP, and rewards. The team reported that Claude 4.6 one-shotted a 16K-word JSON import containing variables, prompt templates, SQL queries, and sub-flow instructions — including sensible node positioning in the visual graph.

Multi-model round-robin for game spec generation

A veteran builder deep in a shmup project described a workflow where three models argue with each other in a round-robin loop. Each model's output gets funneled to the other two for critique and refinement, building out a massive spec broken into micro-prompts per feature. The core playable loop was hand-built for arcade-precise feel, with everything beyond the frame-perfect mechanics delegated to this agentic spec-writing process.

Auto mesh rigging via vibe coding

One developer reported progress on a vibe-coded auto mesh rig script with a modular top/bottom mesh approach for customizable characters, based on a recent research paper. The project is at its furthest point yet, though the builder noted it may need next-gen coding models to finish cleanly.

TECH & TOOLS

Cursor + Codex 5.3 as the quiet default stack

When asked to explain unusually high shipping velocity across multiple simultaneous game projects, one prolific builder's answer was simple: Cursor with GPT-5.3 Codex for planning and Composer 1.5 for execution, plus rules files and ChatGPT for reasoning through problems. No kanban, no remote agents, no orchestration layer — just focused tool use and head scratching.

"Game as MCP" as an emerging architecture pattern

A discussion on quest system design surfaced two contrasting approaches: heavily guardrailed predetermined flows (tracking quest state, inventory checks, NPC pathing per type) versus treating the entire game as an MCP server where the LLM has broader agency over game mechanics. The guardrailed approach prioritizes UX resembling traditional MMOs; the MCP approach trades predictability for open-ended emergence.

QUICK TAKES

• Prompt-to-game skepticism: Multiple builders pushed back on the idea that players will prompt their own games — the analogy to vibe coding apps doesn't hold when the product is entertainment, not utility.
• Play.fun economics: Average token fee earnings sit around $200 per game, with viral hits reaching $2K and only extreme outliers breaking $50K. The model mirrors pump.fun, with all the crypto friction that entails.
• Two-person studios: A veteran argued that a studio of two AI-native devs will outperform 25 traditional devs with AI — reduced ego friction and elimination of misaligned economic incentives.
• YC and gaming: Signals that YC may be opening a dedicated gaming track. Out of ~70 companies in the current Speedrun cohort, about 3 are game studios, including one traditional studio with no AI twist.
• Visual novels and dress-up games: Called out as overlooked genres ripe for AI-native approaches.
• GDC week: The Jabali AI Game Jam runs March 7-8 with $1K in prizes. Unity is reportedly preparing an AI beta that generates casual games from text prompts.

EDITORIAL

The funding conversation surfaced a real structural tension: accelerator-stage capital works for AI game studios, but Series A mechanics don't map well to a medium where every hit is its own snowflake. The VC playbook assumes scalable, exitable businesses — games are project-shaped, not platform-shaped, unless you're building tooling.

The more interesting signal sat underneath: the assumption that two-person AI-native teams can now compete with 25-person traditional studios. If that's even directionally true, it changes what funding is for. You don't need $5M to staff up — you need $1M to buy time and compute.