AI Agent Digest: Week 24, 2026 - Apple Embraces Agents, Agentforce Hits $1B, and Agents Get a Wallet

Apple finally picked a side, Salesforce's agents started out-working its humans, and the payment networks quietly built the plumbing that lets agents spend money on their own. Week 24 was the week agents stopped being a feature and started being infrastructure. Eight stories, with our takes.

1. Apple embraces agents at WWDC 2026

At WWDC on June 8, Apple revealed its long-delayed Siri overhaul and, more interestingly, started shipping real agentic behavior: the Passwords app now uses Apple Intelligence and Safari to "agentically take action on your behalf" to fix insecure passwords. Claude also became a supported model option on iPhone.

Hot take: The Passwords demo matters more than Siri. Apple, the most caution-obsessed company in tech, just shipped an agent that takes unattended action on your accounts. When Apple decides agentic action is safe enough for a billion phones, the "are agents ready" debate is over. The question is now who you trust to act on your behalf, which is exactly the right question.

2. Salesforce's Agentforce hits $1B while its agents out-handle humans

Agentforce crossed $1 billion in revenue, with Salesforce reporting its agents now handle double the inquiry volume of its human support staff.

Hot take: A billion dollars is the real headline, but read the fine print on "double the inquiries". Inquiries handled is not problems solved, and volume is not satisfaction. The milestone is genuine and the deflection is real, but the company whose support brain now lives entirely inside Salesforce should ask what it costs to ever move it out. Impressive, and a lock-in lesson at the same time.

3. McKinsey now counts 20,000 agents as workforce

McKinsey revealed it operates with 40,000 humans and 20,000 agents, aiming for a 1:1 ratio before the end of 2026.

Hot take: "Virtual employee" just stopped being a metaphor. When the most buttoned-up consultancy on earth puts agents on its headcount, the term is a line on the org chart, not marketing. And notice the ratio: two humans per agent, not the other way around. The winning model is augmentation at scale, not replacement, at least for now.

4. Mastercard and MetaMask give agents a wallet

The payment rails arrived in force. Mastercard launched Agent Pay on June 10 with 30-plus partners, bringing structure and governance to machine payments, days after MetaMask shipped an Agent Wallet (June 8) that lets agents transact onchain under mandatory security checks. Visa and PayPal are wiring in too.

Hot take: This is the quiet milestone of the week. An agent that can read, write, and now pay is a different kind of thing entirely, and the networks know it: every one of these launches leads with governance and permissions, not speed. That is the correct instinct. An agent with a credit card is only as safe as the limits you put around it, and "what can it spend, on what, up to how much" is about to be the most important setting in your stack.

5. Microsoft Agent 365 goes GA, KPMG deploys it globally

Microsoft's Agent 365 reached general availability, and on June 9 KPMG announced a worldwide rollout to manage, monitor, and secure agents under its Trusted AI framework.

Hot take: "Managing your agents" is now a paid product category, the way device management became one a decade ago. That validates the whole space, and it also tells you what to value: identity, monitoring, and audit logs for every agent. If a Big Four firm is buying a control tower for its agents, the lesson for everyone else is that owning your agents' identity and logs is not optional.

6. Verizon's CEO says AI will take "a large percentage" of customer service

At Bloomberg Tech, Verizon's CEO said plainly that AI will handle a large share of the company's customer service.

Hot take: He is right, and the panic is the wrong response. The repetitive, high-volume, 3am questions should be automated; the angry, complex, high-stakes ones should not. The companies that win redeploy their best support people onto the hard conversations instead of cutting them. "AI takes a large percentage" and "humans matter more, not less" are the same sentence if you mean it.

7. Agent startup funding is hot, frothy, and consolidating

Average agentic AI rounds hit $155M, nearly double last year, European startups like Parloa and Lovable raised over $640M combined, and analysts are openly using the word "bubble" while smaller teams seek acquisition as a soft landing.

Hot take: The money is flowing to infrastructure (orchestration, security, interoperability) and away from thin wrappers, which is the market getting smarter, not dumber. But "bubble" is fair for the wrapper tier. If a startup's whole product is a prompt and a logo on top of someone else's model, the next funding round is an acquihire. Buyers should plan for the vendor churn that is coming.

8. Anthropic calls for a global AI pause

Anthropic publicly called for coordinated global governance on frontier AI, a notable move from a lab in the middle of shipping ever more capable agents.

Hot take: Easy to call this hypocritical; more useful to call it telling. The most safety-focused frontier lab looking at its own roadmap and asking for guardrails is a signal worth taking seriously, not a PR stunt. For businesses, the read is simple: regulation is coming, so build on a stack where you control the data, the permissions, and the audit trail now, rather than retrofitting compliance later.

What we're watching next week

• Gemini 3.5 Pro, still the big pending model drop after Google's "next month" promise.
• Mastercard Agent Pay's first partners going live, the real test of agent payments at scale.
• More agent-management products following Microsoft's Agent 365, as rivals answer.
• Whether the funding "bubble" talk turns into the first visible down rounds in the wrapper tier.

Bottom line

This was an infrastructure week. Agents got a wallet, a place on the org chart, a management product, and Apple's blessing, all at once. The capabilities are no longer the conversation. Everything that happened this week was about the boring, load-bearing stuff that decides whether agents are safe to run: identity, permissions, governance, and trust. That shift, from "what can agents do" to "how do we run them responsibly", is the whole game now, and it is exactly the ground a self-hosted, owned employee is built to win.

Want to test the most advanced AI employees? Try it here: https://Geta.Team