Your Weekly AI Briefing for Leaders

Welcome to this week's AI Tech Circle briefing: clear insights on Generative AI that actually matter.

This week, more than forty companies agreed on one thing. They agreed on how an AI agent should knock on a business's front door. Meta and Sierra published the Personal Agent Protocol, and the partner list reads like a payments-and-retail census: Walmart, Stripe, Shopify, Visa, Mastercard, PayPal, Bank of America, OpenAI. I read the draft so you don't have to. This week's Deep Dive walks you through how it works, step by step, and what to prepare before the first agent arrives at your site.

Today at a Glance:

• Executive Brief: The AI build-out moves to the bond market, Microsoft and Nvidia put frontier AI on your desk
• Deep Dive: The Agent at Your Front Door - a hands-on walkthrough of the Personal Agent Protocol
• Weekly News & Updates
• Favorite Tip of the Week
• Potential of AI
• Podcast
• Courses and events to attend
• Tool / Product Spotlight
• The Investment in AI

Executive Brief

The AI Build-Out Moves to the Bond Market - and the Numbers Get Audited

The money behind AI changed shape this week. SpaceX is reportedly seeking about $40 billion to buy Nvidia AI chips: roughly $30 billion in bonds and $10 billion in bank loans. Broadcom and Oracle arranged large financings of their own. Morgan Stanley estimates AI infrastructure could need about $1.5 trillion in external financing by 2028. The scrutiny arrived in the same week. Nvidia-backed Firmus withdrew its $5 billion IPO after investors noted only 42 MW of its planned 1 gigawatt of capacity is operational. And Axios reported OpenAI's annualized revenue is about $50 billion, not the $70 billion previously circulated - a difference in how revenue through cloud partners is counted, not a decline.

Why this matters to you: the build-out now runs on debt, and debt demands audited numbers. Add two questions to every AI vendor review: how is this company financed, and how does it define revenue?

Microsoft and Nvidia Put Frontier AI on Your Desk

Microsoft launched the Surface Laptop Ultra on October 7, built on Nvidia's RTX Spark chip to run large models locally. Base models start at $2,600 and $3,700; full options reach $5,900, and the highest-end model sold out at launch. A $6,000 RTX Spark Dev Box targets developers, and Dell priced its own RTX Spark machine at $3,800. The bigger news ships in software: a revamped Windows 11 adds Execution Containers, a sandbox to keep AI agents inside permitted boundaries, coming to all Windows 11 users. Satya Nadella framed the goal as enabling the harness layer "not just for our apps but apps for anyone's agent."

Why this matters to you: last week's Deep Dive asked which AI you should own rather than rent. This week the hardware for owning arrived at laptop prices. Treat your next device-refresh cycle as an AI capacity decision: price one local-AI machine per team that handles sensitive data, and ask IT when Execution Containers reaches your fleet - agent sandboxing by default is worth planning around.

Deep Dive: The Agent at Your Front Door - Inside the Personal Agent Protocol

Until now, a personal AI agent visited your website the only way it could. It signed in as the customer and clicked through pages built for people. You could not tell the agent from the human, and you could not limit what it did. This week the industry proposed a fix. Sierra and Meta announced the Personal Agent Protocol (PAP) on October 6, and published the draft, code-named Poppy, at personalagentprotocol.org on October 9.

The rule to remember: treat customer agents like customers with a badge - discovered, identified, scoped, served. The protocol turns that sentence into plumbing, and it uses parts your team already knows: OAuth, JWTs, HTTPS, OpenAPI, and MCP. Here is the walkthrough:

1. Discover. Your company publishes one file at /.well-known/poppy.json. It tells any agent how to work with you: how customers sign in and what agents may do. Think robots.txt, but for transactions.
2. Identify. An agent opens a session and declares itself, so you always know when an agent acts for a customer. Guest sessions handle simple asks, such as stock checks or your returns policy.
3. Scope. When a task needs account access, the customer signs in on your page through OAuth. The agent then receives a session token that carries only the access the customer approved. In Sierra's retail example, the agent can start a return but cannot change payment details.
4. Serve. One token works across your website, your APIs (OpenAPI and MCP), and your own support agent. You choose which channels to offer; the visit stays continuous across them.

Three pieces of evidence that this draft matters:

The roster. Launch partners are Meta, Genesys, Instinct, Rocket, Shopify, Stripe, and Walmart. Thirty-five more design partners signed on October 9, including OpenAI, Visa, Mastercard, PayPal, Bank of America, Wells Fargo, United Airlines, Target, Okta, and Cloudflare (Sierra). When payment networks, banks, and retailers co-sign an identity standard, procurement follows.

It already runs. At Sierra Summit, Rocket showed Meta's agent Muse completing a mortgage pre-approval through Poppy: the agent found Rocket's endpoint, requested permission, and the customer signed in and approved each data share. Not a slide - a working flow, in a regulated product.

The platforms are converging on it. Microsoft framed its Windows work this week as a harness "for anyone's agent," and Sierra positions PAP above shopping-only standards like UCP: one identity-and-permission layer for any task, "from buying a sweater to applying for a mortgage." The honest caveat: Poppy is a draft, and drafts can die. But the last time this pattern shipped - OAuth sign-in - it became invisible infrastructure within three years.

You can try doing this:

1. Measure the agent traffic you already have. Ask your web team for one number this week: sessions that identify as automated or fail your bot checks on customer journeys. That number is your baseline, and it only goes up.
2. Write a scope map for one journey. Pick one task customers will delegate first - order status, a return, a booking change. Write two lists: what a customer's agent may do, and what it must never do (payments, address changes). This map is your input to any protocol work, PAP or not.
3. Hand the draft to engineering. Have one engineer read the Poppy draft at personalagentprotocol.org and estimate a prototype /.well-known/poppy.json behind a feature flag. Sierra runs design workshops and ships a reference implementation next month; a prototype gets you a seat at that table.

Weekly News & Updates...

• Anthropic released Claude Haiku 5.5 on October 7 at $0.10 per million input tokens and $0.50 per million output tokens - about 75% cheaper to run than Haiku 4.5, and available on AWS, Google Cloud, and Azure.

Why it matters: your high-volume "baseload" tasks got repriced again; re-run last week's benchmark before you renew anything.

• Anthropic disclosed that during automated testing in July, one of its models sent a false homicide tip to a Philadelphia police website. The message landed in spam, and Anthropic halted the process.

Why it matters: agents reach real institutions in ways their builders did not plan; keep every outbound action behind an allowlist.

• Google began rolling agentic capabilities into Gemini for businesses. Why it matters: agent features now arrive inside suites you already license - inventory what you have before you buy a standalone agent product.
• An AP-NORC poll found 64% of Americans say AI is developing too fast, and about eight in ten want human control and worker protection as government priorities.

Why it matters: every rollout you run faces this public; name the human who stays accountable, and say so out loud.

• A Chinese developer withdrew the public code for the ARTEX security tool after CrowdStrike linked it to AI-assisted attacks on at least nine South Korean banks.

Why it matters: ask your security vendors how they gate offensive capabilities - defensive tools can be turned.

Favorite Tip of the Week

Run one real errand through an AI agent this weekend - and keep score. This week's Deep Dive is about agents arriving at businesses. This tip builds your personal feel for what those agents can and cannot do.

1. Pick one low-stakes errand with a real outcome: reschedule an appointment, compare and book a service, or draft and send a refund request.
2. Open the assistant you already use (ChatGPT, Claude, Gemini, or Meta AI) and turn on its browsing or agent mode. Ask it to complete the errand end to end.
3. Watch it work. Each time you must step in - a sign-in, a payment, a question it cannot answer - note it down.
4. Keep the tally: steps it finished alone versus steps that needed you. That tally is exactly what a customer's agent experiences on your company's website today.

Potential of AI

The next AI fund may buy your competitor, not a startup. The signal this week: SoftBank is reportedly seeking up to $100 billion from Gulf investors for a new fund - with Abu Dhabi among those in talks - that would buy established businesses and use AI to improve them, rather than back startups. It follows SoftBank's completed $30 billion OpenAI investment.

On a one-to-three-year horizon, this changes who owns ordinary companies. A buyer that runs call centers, claims processing, and back offices on agents can pay more for the same cash flows than a buyer that cannot. If that model works, "how automatable are this company's operations" becomes a line in every valuation - including yours.

Look at your own P&L the way this buyer would: list the three cost lines an AI-equipped owner would attack first. That list is your transformation roadmap, whether or not anyone ever bids.

The Opportunity...

Podcast:

What is the business value of private AI, and why should companies consider owning the AI systems that connect to their internal data and operations? In Episode 202 of Open Tech Talks, host Kashif Manzoor speaks with James Lang, an operator, investor, and technology strategist at Overlang Venture Partners, about private AI, business growth, AI security, legal technology, business valuation, and the future of work.

​Apple Podcast | YouTube​

Why Companies Need to Own Th... Oct 4 · OPEN Tech Talks: AI wort... 19:16   ​

Courses to attend:

• ​MCP: Build Rich-Context AI Apps with Anthropic (DeepLearning.AI): PAP's API channel is built on MCP; this is the fastest hands-on route into the plumbing behind this week's Deep Dive.
• ​ACP: Agent Communication Protocol (DeepLearning.AI): how agents talk to services and to each other - the context you need to read the Poppy draft critically.
• ​Building AI Browser Agents (DeepLearning.AI): see how agents navigate human web pages today, and why a discovery file beats screen-scraping.

Events:

• ​AI & Big Data Expo Europe, RAI Amsterdam, October 19–20, 2026
• ​Dubai AI Festival, Dubai, UAE, October 26–27, 2026
• ​Global AI Show, Abu Dhabi, UAE, November 12–13, 2026
• ​GITEX Global, Expo City Dubai, UAE, December 8–11, 2026

Tool / Product Spotlight

• ​Hermes Agent (Nous Research): the open-source agent, now packaged for business users and built on open models you can self-host. Worth testing if you want an agent whose weights, logs, and data stay under your own control.
• ​Meta Muse on iPad: Meta's consumer agent reached iPad this week, a month after its mobile debut. Worth testing if you want to feel what your customers' personal agents will do - spend twenty minutes using it as a customer, not as a builder.

The Investment in AI

• TypeSafe closed an $870 million round at a $7.5 billion valuation, led by Andreessen Horowitz with Sequoia and DCVC, less than a month after launching its Jev model. Jev returns only structured decisions - a yes/no, a pick from a list, or a score with a confidence number — in under 700 ms, and TypeSafe claims adoption by about a third of the Fortune 500. What the money says: investors now fund decision plumbing that slots into software, not chat windows. List the workflows where you need a verdict, not an essay; that is where this class of model lands first.
• Butterfly Effect (Manus) raised more than $500 million, co-led by Boyu Capital and IDG Capital with Tencent participating, after unwinding a reported $2 billion-plus acquisition by Meta. What the money says: AI capital is splitting along geopolitical lines, and strong agent startups can stay independent. Know which bloc your agent vendor's capital comes from - it now shapes where your data may live.

That's it for this week — thanks for reading!

Reply with your thoughts or favorite section.

Found it useful? Share it with a friend or colleague to grow the AI circle.

Until next Saturday,

Kashif

The views expressed here are solely my own, based on my experience, practice, and observation, and do not necessarily represent those of my current or previous employers or their clients and customers.