What the Best Business AI Users Are Doing Different

The host says Meta’s early position in the competition for personal AI agents helped its stock rise 27% in September, despite a 5% decline during the week after OpenAI announced a competitor to Muse. It was Meta’s best monthly performance since November 2022, when the company began its “year of efficiency” with layoffs, hiring freezes, and a reduced emphasis on the metaverse. Muse’s early success has added roughly half a trillion dollars in market capitalization and, more importantly, convinced investors that Meta has a viable AI strategy. Most Wall Street analysts now rate Meta a strong buy, but Needham analyst Laura Martin remains at hold. She credited Meta with clearly pivoting away from the metaverse toward personal-agent AI, with Muse at the center, but questioned the eventual payoff because Meta is “notoriously slow at monetizing new products.” Muse is free except for the biggest power users, and Meta says user data will remain separate from its advertising business. Morningstar also called Meta’s ability to run an enterprise business “unproven,” even after the company launched an enterprise platform. The host’s conclusion is that this is still a major improvement: six months earlier, investors questioned Meta’s basic AI competence; now they are debating whether Meta can monetize a successful product.

Anthropic is reportedly trying to market its IPO during the week of November 9 and begin trading around Thanksgiving, although the schedule remains subject to change. The company plans an October 14 investor day for institutional investors. Leaked portions of its S-1 showed an $8 billion operating loss on $4.6 billion of revenue in 2025. The host rejects the idea that those historical losses will necessarily determine the IPO’s success, arguing that Anthropic is expected to grow more than tenfold in 2026 and should not be evaluated like a pre-AI-era company. Potential investors reportedly believe Anthropic can reach a valuation between $1.8 trillion and $2 trillion, which would make it the largest IPO in history, exceeding the $75 billion raised by SpaceX. One joke suggested waiting for an Anthropic “Black Friday” discount after the inevitable post-IPO dip. The host then discusses President Trump’s positive comments after a two-hour dinner with Anthropic CEO Dario Amodei and his wife. Trump said Amodei was “very smart,” and that his views were different from how they are portrayed in the media. This may indicate that some White House-Anthropic hostility has come from staff rather than Trump himself. The episode also reports that Trump spent hours talking with Grok about his legacy and Venezuela, asking how Venezuelans might react if the United States captured Nicolás Maduro. Officials said Trump viewed Grok as “ingenious” after celebrations followed Maduro’s seizure, prompting jokes that an LLM could be influencing later foreign-policy decisions.

SpaceX launched a Falcon 9 carrying a satellite with four Google TPUs, beginning the Suncatcher project, a collaboration with Google and Planet Labs announced the previous November. The project is testing whether solar-powered data centers in space could eventually support scalable machine-learning infrastructure. Four TPUs are far too little compute for a practical data center, so the current mission is a stress test for the chips, satellite systems, and communications. Travis Beals, Suncatcher’s senior director, said the team had communicated with the satellite and that everything was operating as expected. Over the coming weeks, Google will observe how the TPUs handle launch stress, radiation, and extreme thermal conditions. The chips will run only in 15-minute bursts to avoid overloading the satellite’s power and thermal-management systems. If the test works, the next stage is a pair of larger satellites handling heavier workloads and testing laser-based networking. Google’s medium-term plan calls for an 80-TPU satellite network, while it is also discussing a satellite the length of a soccer field. That scale depends on SpaceX’s Starship rather than Falcon 9. Beals described the effort as a multi-year or multi-decade moonshot, saying that if everything works perfectly within five years, the team probably has not taken enough risk or learned enough. The long-term goal is for serving a Gemini query from space to become unremarkable.

OpenAI fired three employees on its safety team for violating company policies governing access to and handling of sensitive information. The company said an investigation found that they had mishandled sensitive information outside established procedures and broken the trust required for their work. Sources told The Information that the issue involved sharing confidential information with an outside organization that conducts AI evaluations. Because the details of the information and the outside organization are unknown, the episode presents competing interpretations. Some commentators wondered whether the employees were effectively being punished as whistleblowers. AI general counsel Nathan Calvin noted that the dismissed researchers were lead authors of academic work on chain-of-thought monitoring and said losing them as safety monitorability was deteriorating was terrible. Others argued that working on AI safety does not create a moral exemption from confidentiality rules and that firing them would be reasonable if the allegations were accurate. The host says the truth is difficult to assess from outside and that people’s reactions will likely track their prior views on AI safety. The broader lesson, which the host believes both sides can accept, is that formal channels for reporting concerns and enabling third-party evaluations need to be established quickly. The episode closes the headline section with reports that some Anthropic users are being routed to a model called Fable 5.5, apparently with a newer knowledge cutoff and stronger design, 3D, and spatial capabilities. The host says the release could be significant even if Fable 5.5’s impact feels less dramatic because the existing Opus 5.5 is already extremely capable.

The host argues that enterprise adoption matters not only to people working inside large companies but also to freelancers, service providers, and anyone concerned with how business demand will shape AI’s future. Ramp’s latest AI Index found that token spending fell 5.2% week over week while token volume increased. Companies were therefore using more AI while paying less for each unit of usage. Ramp economist Eric Harasean said the change was not primarily due to open-source models, which account for less than 5% of business spending; he attributed it almost entirely to competition between OpenAI and Anthropic. The host highlights OpenAI’s enterprise marketplace as another sign of this competitive shift. Companies can use their OpenAI spending commitments to purchase other models sold through the platform, including cheaper open-weight models. Enterprises get the multi-model flexibility they increasingly want, while OpenAI retains customer lock-in and keeps the spending inside its ecosystem. The host warns that if investors conclude that usage will continue rising while spending falls, they may reassess the economics of AI infrastructure and how much infrastructure expansion they are willing to fund. The episode then turns to KPMG’s Q3 AI Pulse Survey, based on more than 2,100 senior leaders in 20 countries. KPMG separates organizations still experimenting with AI from those that have established return on investment, and the more mature organizations consistently show stronger adoption, governance, security, and management practices.

Among organizations with established AI ROI, 48% report significant employee adoption of AI agents, compared with only 15% among experimenters. Mature organizations are also much further ahead on AI-assisted cyber defense: 58% have implemented it, versus 8% of experimenters. Eighty-six percent of ROI organizations have a formal AI “harness layer”—an interface through which employees interact with AI—compared with 31% of experimenters. Data sovereignty shows an even larger gap: 53% of mature organizations have an enterprise-wide sovereignty strategy, versus 8% of experimenters. KPMG’s broader conclusion is that enterprise AI is shifting from individual experimentation toward a management layer that makes AI operate within an organization’s rules and context. Fifty-three percent of organizations place accountability for AI-informed decisions at the C-suite level, and 23% have built some model-routing capability. Across experimentation, planning, scaling, adoption, and ROI measurement, mature organizations are more likely to use formal cross-functional or enterprise-wide approaches. The objectives are changing as well. Productivity remains important but fell from 42% of organizations in Q1 to 37% in Q3. Human-AI collaboration, responsible AI and governance, trust and security, and adaptability and resilience all gained importance. A quarter of organizations are developing multi-agent systems, and one-fifth are orchestrating multiple agents. Revenue-focused strategies are rising while purely efficiency-focused strategies are declining, with most organizations pursuing both rather than choosing between them.

KPMG describes a shift from basic cost control to “AI economic management.” Cost visibility is now widespread, but organizations are increasingly trying to connect spending to business outcomes. Among organizations with established ROI, 48% consistently assess value against cost, compared with no such expectation for the experimenter category because those companies have not yet established ROI. AI cost-monitoring dashboards appear in 77% of mature organizations and 43% of experimenters. Cost reviews at approval points are used by 73% of mature organizations versus 50% of experimenters, while usage or token budgets appear in 46% and 31%, respectively. The host cautions that overly restrictive token budgets during experimentation can limit learning, although the survey does not reveal what each company’s budget rules actually look like. The overall pattern is that mature companies are not merely buying model access; they are constructing systems for controlling, routing, monitoring, governing, and evaluating AI across the organization. Compared with KPMG’s Q3 2025 survey, when companies were still being asked whether they had tried an agent at all, these management questions have become mission-critical. The host views that maturation positively and rejects the assumption that efficiency efforts necessarily mean lower investment. Average planned AI investment for the next 12 months rose from $186 million in Q1 to $210 million in Q3. His conclusion is that companies are learning to spend more intelligently while continuing to expand their overall investment in machine intelligence.