What the Best Business AI Users Are Doing Different

October 2, 2026 · Episode Links & Takeaways

Author

AI Daily Brief
October 02, 2026

HEADLINES

Muse Gives Meta Its Best Month Since 2022

An early win in the personal agent wars sent Meta stock up 27% in September, its best month since the November 2022 start of the "year of efficiency," even after a 5% slide this week on news of an OpenAI Muse competitor. Muse has added roughly half a trillion dollars in market cap and, more importantly, convinced the market Meta has a viable AI strategy. Not everyone is sold: Needham's Laura Martin held her hold rating, calling Meta "notoriously slow at monetizing new products," and Morningstar called its enterprise ambitions "unproven." Still, six months ago investors were questioning Meta's basic competence in AI. Questioning how it monetizes a successful bet is a huge shift.

Anthropic Targets a Pre-Thanksgiving IPO

Anthropic is pushing to start marketing its IPO the week of November 9, with an investor day for institutional buyers set for October 14 and trading targeted for early Thanksgiving week. The leaked S-1 showed an $8B operating loss on $4.6B of 2025 revenue, but treating 2025 numbers as decisive for a company that has more than 10x'd in 2026 misreads what Anthropic is. If the IPO underperforms, it won't be the 2025 operating loss that did it. Potential investors reportedly believe the $1.8–2T target valuation is achievable, which would make it the largest IPO in history, ahead of SpaceX's $75B raise.

Trump Warms to Dario, and Turns Out to Be a Grok Power User

In a wide-ranging TIME interview, President Trump said he "liked him and his wife a lot" after a two-hour dinner with Dario Amodei, adding that Dario's views are "much different" than portrayed in the media. That suggests at least some of the White House–Anthropic acrimony was staff-driven, recalling Wired's June report of staffers relieved that Tom Brown wasn't "being a weirdo like Dario." The bigger reveal: Trump reportedly spent hours with Grok, including asking how Venezuelans would react to Maduro's capture, and came away thinking the chatbot was ingenious. That led plenty of people to wonder what Grok might have said about Iran.

Google's First Space Data Center Prototype Reaches Orbit

For anyone who thought data centers in space were a marketing ploy, a Falcon 9 just put a Planet Labs satellite carrying four Google TPUs into orbit, the first launch under Project Suncatcher. Four TPUs is a stress test, not real compute: the chips will run in 15-minute bursts to gauge radiation and thermal tolerance, and early signs are that everything is operating as expected. The roadmap runs to an 80-satellite network and a soccer-field-length mega-satellite, which hinges on Starship. Suncatcher lead Travis Beals framed success as the point where nobody thinks twice about a Gemini query being served in space.

OpenAI Fires Three Safety Researchers Over Information Sharing

OpenAI parted ways with three safety team members for mishandling sensitive information, which sources say involved sharing it with an outside AI evaluation organization. Reactions split predictably along AI safety lines: some read it as retribution against whistleblowers, Encode AI's Nathan Calvin lamented losing chain-of-thought monitoring researchers, and others argued a safety label is no exemption from confidentiality rules. Without knowing what was shared, the justification is impossible to judge from the outside. What most can agree on is that formal channels for reporting and third-party evaluation need to be up and running as soon as possible.

First Sightings of Fable 5.5

Some Anthropic users report being routed to a model with an updated knowledge cutoff and noticeably sharper design taste, widely assumed to be Fable 5.5. Early testers are pointing to design, 3D work and spatial intelligence as standouts, with the caveat that Opus 5.5 is already strong enough to make the jump feel less dramatic than it is.

MAIN STORY

What the Best Business AI Users Are Doing Differently

KPMG's Q3 AI Pulse, a survey of 2,100+ senior leaders across 20 countries, now splits organizations still experimenting from those with established ROI, and the gap between them reads like a map of where business AI is heading over the next several months. The leaders are building model routers, sovereignty strategies and a far more robust management layer on top of AI. Enterprise adoption matters even outside the enterprise: it shapes how model providers compete and how much Wall Street is willing to fund the buildout.

KPMG AI Pulse

HOW AI IS WORKING AT WORK

Ramp AI Index
Token volume up, token spend down 5.2% week over week.
The decline isn't open source, which remains under 5% of business spend; it is driven almost entirely by OpenAI–Anthropic competition. If "volume up, spend down" becomes the accepted trend, expect implications for how much investors will backstop infrastructure.

OpenAI's Enterprise Model Marketplace
Multi-model flexibility for customers, lock-in for OpenAI.
An underappreciated Dev Day announcement: enterprises can now apply OpenAI spend commitments to other open models sold through OpenAI, keeping spend in its ecosystem even when it flows to other token merchants.

The Agent Gap
48% of ROI leaders see significant agent adoption; experimenters, 15%.
2026 was the year agents became real, and the leaders are pulling away. 58% of established-ROI organizations run AI-assisted cyber defense, versus 8% of experimenters.

The Management Layer
86% of ROI leaders have a formal AI harness layer.
53% have an enterprise-wide sovereignty strategy (vs. 8% of experimenters), 53% place accountability for AI-informed decisions in the C-suite, and 23% have built model routing. The more mature the organization, the more formal and enterprise-wide its approach on every dimension.

Beyond Productivity
Revenue-focused agent strategies rising, efficiency-focused strategies declining.
Productivity as a key goal fell from 42% in Q1 to 37% in Q3, while human-AI collaboration, responsible AI and governance, and adaptability and resilience all gained. Most organizations now pursue efficiency and revenue together rather than trading one off against the other.

AI Economic Management
Cost visibility is widespread; linking it to value is next.
77% of ROI leaders have cost monitoring dashboards (43% of experimenters), 73% run cost reviews at approval (50%), and 46% have usage or token budgets (31%). Overly restrictive token budgets during experimentation risk being limiting.

Spending Keeps Climbing
Average planned AI investment rose from $186M to $210M.
A year ago the Pulse was still asking whether organizations had tried an agent; now sovereignty, routing and AI economics are mission-critical decisions. Cost discipline isn't shrinking budgets, and spending on intelligence is barely scratching the surface.