MCG Live · October 23, 2025 · 1 hr 07 min
Spencer from Orgo discusses the V2 launch of their computer-using agent platform, expected within weeks, featuring multi-agent collaboration, Windows support, team features, and GPU access. Most striking: Claude 3.5 Sonnet jumped from 43% to 62% on the OSWorld benchmark in just one month, outpacing expectations for AI capability improvement.
Orgo V1 launched on Believe as an MVP allowing users to spin up Linux computers and connect them to agents like GPT; V2 will be a major product iteration releasing by end of next week.
V2 features include team collaboration (multiple people working on one project), increased memory limits, GPU support for running graphics-intensive software like AutoCAD, and Windows support alongside Linux.
Early use cases include multi-agent pipelines (one agent researches online, another enters data into spreadsheet), and testing workflows with 5-10 personas of agents evaluating SDKs and websites.
Claude 3.5 Sonnet achieved 62% accuracy on the OSWorld benchmark for computer tasks (baseline human 72%), jumping from 43% a month ago; Gemini 2.5 also launched computer use capability recently.
October 8 SDK update reframes UI around projects containing multiple resources: multiple computers, multiple human collaborators, and upcoming file storage (Dropbox-like) for agents to submit completed work.
Orgo plans M2M (machine-to-machine) communication interface between computers, allowing agents to send and receive data without relying on screenshots, optional visual/headless operation modes.
Orgo developer base grew from 1,000 to 1,400 organically in one month; goal is 10,000 developers by year-end; still in product-focused phase before major marketing push in coming two months.
Orgo closed its first funding round; investor details and token distribution changes (50% locked supply changing hands) will be announced alongside Believe content release.
Token utility planned as 'Costco membership' for token holders offering wholesale compute discounts and valuable access on Orgo; paid monetization targeted for February-March after free developer phase.
Claude models built-in safety guardrails prevent bots from creating Gmail accounts or evading CAPTCHA, recognizing when to hand tasks to humans; agents demonstrate awareness of needing human intervention for restricted actions.
This summary was written from the recording's transcript by a language model. For the exact words, ask the MCG Live archive: every answer there is quoted word for word and linked to the second it was said. Market Bubble Search is independent and not affiliated with the show. Nothing here is advice.