October 7, 2026

Global AI Edition

For the last few years, we’ve been told that all of us needed to learn a succession of new AI skills.

First, we needed to know how to build a custom GPT. OpenAI has already marked those for extinction.

Then came prompt engineering. That era is ending just as quickly. The newest models increasingly figure out for themselves how to break a problem into steps, what information they need and where to find it.

You can watch the next transition happening right now, in both consumer and business AI. Meta’s Muse — a personal agent that can work across your email, calendar, and accounts — launched less than a month ago and is now the No. 1 iPhone app in America. OpenAI has Dots, with Google, Grok and others moving in the same direction. At work, systems like Codex and Claude Code are increasingly able to create and coordinate agents.

The common idea is simple: instead of telling AI how to do every step of a job, increasingly you tell it what you want done and let it figure out how to do it. And increasingly, the AI may spot the opportunity before you do and suggest the work itself.

Which brings us to Ethan Mollick, the Wharton professor behind the One Useful Thing newsletter. Mollick writes that he had assumed the next skill we would all have to learn was how to manage these agents: creating teams of them, deciding which agent should do what, dividing up the work and coordinating the whole operation.

He now thinks he was wrong.

Mollick points to OpenAI’s use of what is now being called a “swarm” — thousands of agents powered by an advanced model and set loose on difficult problems.

“Under my old model, think about what managing this kind of work would have required. Ten thousand workers and an unspecified problem — how would you tell them what to do? How would a human manager decide which of 2.7 million messages mattered? How would they coordinate with each other? The swarm figured it out. I don't have 10,000 agents, but I now regularly see OpenAI's Codex and Claude Code using agents as needed.”

For companies — and the systems they are building — that could be a much bigger change than it sounds. We have largely imagined AI at work as a better and better tool operated by an employee: find the right system, write the right prompt, provide the right context, point it toward the right internal information, ask follow-up questions, check the answer and turn it into something useful.

But that entire workflow may be transitional too.

Imagine instead telling an AI what you actually need accomplished. It determines what information it needs, searches everything it is authorized to see, decides how many agents the job requires, assigns the work among them, has them challenge and check one another, and comes back with the finished product.

That is Mollick’s larger point. The machines are getting better not only at doing the work, but at figuring out how the work should be organized.

And if that continues, the next AI skill to disappear may be managing the AI itself.

Tip of the Week

Morning Monitor: How to Turn Your ChatGPT Dot Into a Daily Project Briefing Assistant (The Rundown AI)
‌

OpenAI recently launched ChatGPT Dots, a new always-on AI agent designed to shift AI from something you consult into something you assign.

‌

For business users, that means a Dot can keep track of a project, check for changes, run background work, use connected software and return when a decision needs human input. OpenAI recommends starting with work you want your Dot to “keep track of,” rather than treating every interaction as a fresh request.

‌

A good first use case is a morning work brief: give your Dot a simple set of project notes containing things like key deadlines, task owners, upcoming meetings, open decisions and status updates, then see how accurately it turns that information into a useful daily summary before giving it broader access or responsibilities. (Note: Dots are currently rolling out to paid Pro and Business Premium users.)

Here are some step-by-step instructions:

1) Open ChatGPT’s desktop app or web version and follow the Dot setup (you can’t create a Dot on mobile). Leave local-computer access off for your first test.
‌

2) In the mobile app, open Customize → Custom rules and select ‘Ask’ before taking action. Keep sending, publishing, and purchases outside the task.
‌

3) Use + to attach project notes (samples in the full guide) and prompt: “Use only [FILE] to prepare a morning project brief. Cite the source for each fact. Reply here only.”
‌

4) Check facts: owners, dates, and anything it can't confirm. Once a one-off brief passes, ask for a weekday brief at 8:00 AM and manage it under ‘Scheduled'.
‌

Pro tip: Before scheduling, attach an update that changes a deadline and ask for a revised brief. A smooth first answer tells you little about how your Dot handles changes.
‌

Why it matters: Dots represent a shift from AI that responds when asked to AI that can stay on top of changing information and recurring work over time, opening the door to more proactive project management, routine briefing and day-to-day coordination.

What I'm Reading

Follow the Money: The Top 100 Gen AI Consumer Apps, and Where Users Are Spending (a16z)

Andreessen Horowitz (a16z) just released the 7th edition of its Top 100 Gen AI Consumer Apps, ranking the leading web products by monthly visits and mobile apps by monthly active users, while adding a new 50-app ranking based on US consumer spending.

The leaderboard is becoming more established, with just 11 products making their first appearance this time, while ChatGPT still has roughly three times as many UA paid subscribers as its major rivals.

Here’s a look at the current hierarchy among generative AI mobile apps:

‌

AI use is now widespread, but paying for it remains much less common: nearly half of Americans report using AI, yet just 4.5% had a paid subscription to ChatGPT, Gemini or Claude as of August. The people who do pay are highly engaged, with the top 10% of spenders accounting for roughly half of observed spending and favoring coding, productivity and creative tools.

‌

With personal agents now emerging as the next major battleground, a16z’s recurring ranking is a useful snapshot of how quickly the market keeps shifting and makes us wonder what game-changing advancement will come next to shake up the race all over again.

‌

Frontier Return: Google’s New Argon Model Takes Aim at OpenAI and Anthropic, Raising the Stakes on Guardrails (New York Times)

‌

Google released Gemini 4 Argon, its newest frontier AI model, as it looks to regain ground against OpenAI and Anthropic in the intensifying race for more capable models.

‌

Built for complex, long-running workflows, Argon is designed to reason through software engineering, financial and legal research and cybersecurity tasks, with a dramatically expanded 1-million-token output limit that allows it to work through much larger problems in a single run.

‌

Google says the model has already been used internally for tasks including large-scale code migrations, data-center memory optimization and specialized research, while outside testing cited by the New York Times found it outperforming rival models in coding, finance and other areas.

‌

The increased capabilities are arriving with additional safeguards. Google is initially limiting Argon to trusted testers and cyber defenders while it refines new guardrails designed to detect when the model goes beyond a user’s intended task and, if necessary, stop its activity, highlighting the broader industry debate over how quickly companies should deploy increasingly autonomous AI systems.

Equation Acceleration: OpenAI Releases Findings on 377 Math Problems, Further Roiling Field (New York Times)

‌

On Tuesday, OpenAI deluged mathematicians with hundreds of new findings that span a wide swath of topics including algebra, number theory, theoretical computer science, mathematical logic and topology.

‌

Levent Alpöge, a top mathematician at rival Anthropic called it “obviously the most significant moment in mathematical history.”

Market Launch: Moonshot Said to Eye Early 2027 IPO After Value Hits $50 Billion (Bloomberg)

Moonshot AI has closed the final round of private fundraising at a valuation of about $50 billion and is heading toward an initial public offering in Hong Kong in the first quarter of next year, according to people familiar with the matter.

Moonshot is one of the biggest stars of China’s AI scene, bursting onto the global stage in July with the debut of Kimi K3, an open model that goes nearly toe-to-toe with leading platforms from the likes of OpenAI and Anthropic on some metrics. Its planned IPO is one of the most hotly anticipated deals in Hong Kong, where fundraising driven by the industry is setting fresh records. 

Retail Intelligence: As A.I. Agents Begin Shopping, Brands Are Changing Their Sales Pitch (New York Times)

As AI agents increasingly shop on consumers’ behalf, marketers are beginning to rethink advertising for an audience driven less by emotion than by data, logic and product information. The stakes are rising quickly: AI-driven traffic to US retail sites jumped 393% in 2026, while Morgan Stanley estimates AI agents could influence as much as $385 billion in U.S. e-commerce spending by 2030.

Moving In: Apple to Launch Doorbell, Lock, Thermostat Developed With LG (Bloomberg)

‌

Apple is making a bigger play for the smart home, teaming up with LG on a new lineup that includes a doorbell, thermostat, smart lock and security cameras. The devices will plug into Apple’s new smart home hub, part of a broader Oct. 13 launch that also includes an upgraded HomePod mini and Apple TV box.

Inside Bloomberg’s ASKB: How the Terminal Is Rethinking Agentic AI (Hedge Fund Alpha)
‌

AI that draws conclusions from many documents at once has to rank news and company filings differently. Recency matters most for news, document type matters more for filings, and perspective matters across both. Bloomberg’s answer is ASKB, a conversational AI interface for the Terminal built on specialized agents rather than one massive model.

‌

In an interview with Hedge Fund Alpha, Wayne Barlow, Global Head of Terminal Products at Bloomberg, explained how that works: attribution down to the excerpt, guardrails against hallucinations, and where his team is deliberately holding agents back.

‌

If every Terminal user can synthesize the same data instantly, where does an investor’s edge come from? Wayne’s answer is ASKB Workflows, which are essentially pre-built agents or skills.

‌

“What that allows our customers to do is really customize the system through queries that set up workflows the way the individual user thinks about a problem,” he explained. “But the combination of allowing our users to bring in their documents and data allows them to construct their own ASKB Workflow templates that they can run. These are really complex. I think of them as complex prompts, like agents or skills that really are the client’s way of thinking.”

Read Wayne's full interview here.
‌

And in case you missed it: see ASKB in action and hear directly from Bloomberg teams about how the conversational AI interface is changing the way customers research, synthesize information and uncover insights across the Terminal.

‌

Watch the video below for a closer look at how ASKB is bringing Bloomberg’s approach to agentic AI to life.

‌

Best of Late Night

‌

On Trump’s announcement of a new “Super Intelligence Force,” a government initiative he says will bring together top talent and advanced AI capabilities to strengthen US technological leadership:

"All right, what the hell is that? It sounds like a badge you’d get in middle school for reading 10 library books.” 
-- Seth Meyers

“Everything he comes up with sounds like a game you played with your brother when you were 9 years old.”
-- Seth Meyers

“What happens if the President of China appoints a ‘Super-Duper Intelligence Force?’ How will we top that?”

-- Jimmy Kimmel

‌

And from Jimmy Kimmel Live: Jimmy Kimmel is back after a week of shows in Brooklyn, weighing in on Trump’s new Super Intelligence Force, Republicans’ midterm troubles and efforts to “buy” votes, before wrapping with a street-level edition of “What’s Your Problem?” featuring Brooklynites.

‌

Thank you for reading!

Please forward my email to anyone you think should read it. They can sign up here. If you no longer wish to receive the newsletter, you can unsubscribe here.

Follow me on X and LinkedIn for more updates. And get the latest updates from Mike on LinkedIn, X, Instagram and Threads.

Reply to this email with any feedback or tips, or send a note to dailyread@bloomberg.net.