SPEAKER_00 0:00

Welcome back to Sumo Growth AI Insights. I'm walking you through this week's AI news the way an operator actually uses it, not as a pile of headlines. Lead story. OpenAI releases 722 AI-generated math papers from an unreleased model. On Tuesday, OpenAI published 722 mathematical papers, organized into 372 groups of results on GitHub, all produced by an internal frontier model it has not released yet. The company said it fed the model about 4,000 problems. Most results came from a single prompt to a single agent, and the average result took roughly three hours of ChatGPT Pro thinking compute. The release spans number theory and algebraic geometry, includes lean formalizations of many proofs, and, according to the Wall Street Journal, touches three of the five remaining Millennium Prize problems, including the Riemann hypothesis. It lands a month after OpenAI's claimed Navy or Stokes proof, which used up to 10,000 agents over about 88 hours, and amid pushback from mathematicians over attribution and review. For agencies, the signal is simple. One agent with one prompt and a few hours of compute is now producing expert grade work, so the value you sell shifts toward verification, review, and accountability. Plus story, DeepSeek nears a $12 billion raise backed by Tencent and CATL. Chinese AI lab DeepSeek is close to securing at least 80 billion yuan, about $12 billion, in a new round, well past its original target of roughly 50 billion yuan, Bloomberg reported on October 6. Signed term sheets could push the total toward 100 billion yuan, about $15 billion, with Tencent and battery giant CATL among the largest investors. DeepSeek began this raise in July at a targeted valuation of about $74 billion, a month after taking roughly $7.4 billion in its first outside round, and it has hired CIDIC Securities to prepare a possible Shanghai Star Market IPO in early 2027. The money follows a $1 billion annualized revenue run rate in a new Huawei partnership on developer tools for Ascend chips. For agencies, low-cost open models are about to get a much bigger war chest, which keeps steady downward pressure on the model prices behind every stack you resell. Three more headlines. First, OpenAI is seeking $30 billion at a $1.4 trillion valuation. OpenAI is in talks with UAE funds, including Abu Dhabi's MGX plus BlackRock, to anchor around of at least $30 billion, Bloomberg reported. The UAE funds have discussed up to $10 billion together, and OpenAI has pushed IPO plans to at least next year after raising $122 billion at an $852 billion valuation in March. Second, Google is limiting free Gemini users to Flash Light starting October 9th. Free users lose access to Gemini Flash and Pro. Standard Flash will need the $4.99 a month Google AI Plus plan, which itself loses Pro, while Pro and DeepThink moved to the 1999 AI Pro and $99.99 Ultra Tiers. Third, Automation Anywhere is acquiring voice AI company Boost.ai. On October 7th, Automation Anywhere agreed to buy Boost.ai from Nordic Capital, adding Enterprise Conversational and Voice AI that carries customer requests through to back office execution. The deal is expected to close in the fourth quarter and follows its ISERA acquisition in late 2025. Tools of the day. Velocity at getvelosity.com is a shared workspace where your team and AI agents chat, meet, and run tasks with the same context on the AI subscriptions you already pay for. IrisGo at irisco.ai lets you show a task once with Watch and Learn and turn it into a reusable workflow, plus daily briefs and email triage for solopreneurs. Genpage at genpage.ai builds a personalized landing page for every ad, keyword, or target account, then lets autopilot keep testing it. Business AI tip of the day make every AI conversation end in a logged outcome. Automation anywhere is buying Boost.ai because a voice or chat agent that only talks is half a product. The value shows up when the conversation turns into finished work behind the scenes. Your clients want the same thing from every receptionist, chatbot, and follow-up agent you ship. For every sumo AI CRM, Go High Level, or client agent conversation flow this week, list the five most common intents like book, reschedule, quote, support, and cancel, and map each one to a single CRM action. Make the agent write the outcome to the contact record as a tag, opportunity stage, appointment, or task, never just a transcript. Add a human approval step for anything that sends money, signs, or promises a refund. The same rule velocity applies to agent output. Route every unknown intent to a named owner with a task due in two hours instead of a dead end. Track one metric per intent, conversations that ended in a completed action divided by total conversations, and review 10 transcripts every Friday, then turn the top failure into a new workflow branch. Clients do not renew for clever conversations. They renew for booked calls, closed tickets, and pipeline they can see in the CRM. Pro tip Sell a conversation to outcome audit as a paid add on. Pull last month's call and chat logs, show the client what percent ended in a logged CRM action, and quote the build that closes the gap. That is your sumo growth AI Insights briefing. Make every conversation end in an outcome this week, and I will catch you on the next issue.