Meta Muse Puts Distribution Ahead of Model Power, AI Shrinks Onchain Game Building From Months to Hours

Rare Friends’ public hackathon puts distribution ahead of model power, inviting participants to “bring an idea” while AI handles the code and the project supplies distribution. It is built around an open-source GitHub project and SDK, offers about $40,000 in prizes across separate project lanes, and does not require a Genesis asset. The live toolkit includes prompts and tools for building, exploring, and sharing creations, with proposed support for token economies, asset deployment, randomness, project distribution, and fees when assets trade. Developers can build experiences around Rare Friends NFTs without changing their locked metadata or images, though a game inspired by the project’s visuals need not be built directly on an NFT.

AI-assisted “vibe coding” is lowering the time and resources needed to make on-chain games: the host said an individual could build Minesweeper in hours rather than the months similar work might have taken in 2021–22. A live Minesweeper game associated with Rare Friends illustrates the opportunity for simple, nostalgia-driven experiences, though the host said Minesweeper itself is unlikely to transform the industry or onboard the next wave. Easier creation could expand the pool of builders and encourage collaboration across projects, but the same lower barriers during the 2021–22 mania might also have produced more low-effort scams.

The speakers described Meta Muse as a strong consumer product despite not using the most powerful model among the agents discussed. They praised its intuitive interface, said it is free and requires no extra hardware, and contrasted its easy setup with OpenClaw’s greater potential power and customizability. Their central argument was that Meta’s reach and familiar Facebook-style interface may matter more to mainstream users than model capability: Facebook has 3 billion monthly users, and Muse could introduce many of them to agent chat. One speaker said Muse had reached number one on the app store, while another found its conversational agent and personalized feed intuitive after fewer than 10 conversations.

That distribution could also help Meta monetize personal assistants, although forecasts remain speculative. One speaker argued that agents could make it easier for users to act on recommendations and spend through Meta, potentially multiplying its advertising business tenfold; another cited projections that 80% of advertisers plan to buy AI ad formats in the next 12 months and that the channel could attract an 82% figure, 20 percentage points above any other purchase channel. Speakers suggested some consumer searches and advertising budgets could shift from Google toward AI interfaces, but Gary argued that conventional SEO remains important because AI systems crawl indexed websites and content. The discussion also highlighted a trade-off: personalization may require sharing more personal information with Meta, while local models were described as the strongest privacy option. Participants differed sharply on how much they trust AI platforms.

The episode framed AI competition as a set of user and task-specific markets rather than a guaranteed winner-take-all contest. Muse may serve a broad consumer audience, while Claude and ChatGPT may appeal more to technical or enterprise users; coding and infrastructure tools may form separate segments. Ben Thompson called Muse a bearish signal for frontier labs because a capable, approachable personal agent could make users stickier than a chatbot, even though Meta’s Muse Spark is not state of the art. Speakers said that accumulated personal information and daily integrations could make agents harder to switch away from, while Meta’s existing distribution, product work, infrastructure, and possible future hardware such as its glasses could be advantages. The eventual form factor and competitive outcome remain uncertain.

The Other Side’s developer community is taking shape around ODK and Unreal, with Tropical Virtual offering four introductory classes. Hypha Labs’ Star Force Assault, an ODK 3D starfighter game with one-on-one and team combat and destroyable ships, had been tested for about a week and a half, with new dynamics added the prior week. The host also speculated that Kodas could gain a central role in Otherdeed’s resource game through bonuses or land protection, but acknowledged not knowing the game’s details. There are 10,000 Kodas in total, including 8,318 decoupled from Otherdeeds and 1,650 still coupled; if the game succeeds and gives Kodas active utility, the host predicted they could outperform Mutants, partly because their supply is about half as large. The host guessed the resource-economy beta might last less than six months, while stressing the timeline was uncertain.

Markets and tokenized-asset infrastructure supplied a broader backdrop. Spot Bitcoin ETFs took in $999 million, their largest single-day inflow in 11 months and third consecutive day of inflows, while ETH ETFs added $270 million, their largest day of 2026. Bitcoin was up 44% on the quarter, and CryptoQuant’s Julio Moreno called its move above the 365-day moving average final bull-market confirmation. Ondo said its in-kind conversion between tokenized stocks and ETFs and underlying shares is live on Ethereum and BNB Chain through Alpaca’s Instant Tokenization Network, removing the cash-first requirement. Ondo expects market makers to replenish inventory from shareholdings, potentially tightening spreads and deepening liquidity; the possibility of tighter spreads while traditional markets are closed was presented as uncertain.