In the last week of June, two seemingly unrelated things happened at once.
Tencent QQ Mail quietly launched Agently Mail — and it is not an AI assistant that manages your inbox. It is the reverse: AI received a mailbox of its own, with the @agent.qq.com suffix, completely isolated from personal email. The same month, Google’s A2A protocol passed 22,000 stars on GitHub with more than 150 organizations running it in production. The AP2 (Agent Payments Protocol), released in September 2025, added over sixty partners including Mastercard, PayPal, and Coinbase. Its companion x402 extension lets agents settle directly in USDC, with on-chain transaction costs under one cent.
The mailbox: acquired. The wallet: acquired. Put the two together and June 2026 should not be remembered as “Tencent shipped a feature” or “Google advanced a protocol.” It should be remembered as the month AI agent communication infrastructure was formally completed — and the month the human’s status as the default participant in digital communication began to wobble.
Not Agents Replacing Humans — Infrastructure Reassigning Who Counts as a Communicant
The popular AI-anxiety narrative says agents will take human jobs. That narrative misses the actual problem. Job displacement is gradual — it needs time, policy, and cost-structure shifts. But a change in the underlying logic of communication infrastructure happens overnight. Look closely at the two events above and you’ll see they operate on different layers: Tencent’s Agently Mail solves the identity layer — an agent needs an independent, recognizable digital identity to participate in correspondence, and the @agent.qq.com suffix grants it a legitimate marker distinct from humans. Google’s A2A + AP2 + x402 solve the payment layer — an agent needs independent payment capability to operate inside an economy, with triple-signature authorization keeping autonomous trading within granted bounds.
Each layer solves a concrete problem. But each is also doing the same other thing: quietly expanding participation in communication and commerce from “humans only” to “AI too.” This is not replacement. It is infrastructure redefining who counts as a legitimate participant.
A Paradox: Efficiency Tools Are Manufacturing More Overload
Tencent listed four showcase scenarios: automated invoice collection for expense reports, aggregated news briefings, automatic resume delivery, e-commerce reconciliation. Every one sounds efficient; every one sounds reasonable. Then the research data lands, and in this particular month it is glaring. As of April 2025, 51 percent of spam was AI-generated — surpassing human-written spam for the first time, on a climb that started the month ChatGPT launched in November 2022. By late 2025, Hoxhunt measured a 14-fold surge in AI-generated phishing. SentinelOne’s report put the year-over-year increase in AI-driven phishing at 1,265 percent.
The subtler shift is on the “legitimate use” side. One Reddit user subscribed Claude to thirty newsletters and has AI read them all. Someone built a fully automated news-briefing agent with Claude Code. Someone else wired Claude over MCP to auto-reply to Instagram comments and send DMs — AI conducting complete conversations on a user’s behalf. Technically, these stacks are nearly indistinguishable from a spammer’s: an LLM generates text, automation sends it, no human reviews word by word. The only difference is the intent of the operator.
Here is the absurd reality of the 2026 inbox: AI reads what AI writes, AI replies to what AI sends, and the human has moved from participant to bystander. The loop always lacked one thing — AI had no identity of its own, so it borrowed human addresses and operated in a gray zone. Agently Mail changed the premise: with @agent.qq.com, the AI is no longer “impersonating a human sender.” It participates in correspondence under its own name.
The Identity Infrastructure Trap: A Three-Layer Framework
Once agents hold independent identity, the spectrum of communication participants shifts visibly: human-to-human (the classic email era), then human-to-AI (AI-assisted drafting), then AI-to-human (AI auto-replying to people), and now AI-to-AI — agents corresponding autonomously while humans press the confirm button. Every step was sold as efficiency, and every step pushed the human one notch further from the loop. Dissect the agent infrastructure and it resolves into exactly three layers. The identity layer: QQ Mail supplied the @agent.qq.com suffix; Google’s Agent Card supplies cryptographic signature verification; the Linux Foundation governs the A2A standard. Without identity, an agent cannot be recognized, trusted, or authorized online — so identity must come first. The communication layer: agents need a standard protocol to discover, coordinate, and delegate. A2A solved it, with 150 organizations in production including Microsoft, AWS, Salesforce, and SAP — no longer a niche experiment but a functioning industry standard. The payment layer: agents need independent economic capability. AP2’s triple-authorization credentials define what may be bought, what was selected, and what is finally paid — every step cryptographically verifiable — and x402 drives transaction cost below a cent via stablecoin.
Every builder stresses safety. Tencent built a two-stage confirmation — the agent produces an operation summary and executes only after human sign-off. AP2 has triple signatures. Agent Cards verify domains. But the safety design is itself the most honest signal: without those valves, the system genuinely would run on its own — and it would most likely run toward chaos.
The Framework Explains More Than Mail
Run other recent headlines through the three layers. Vaudit versus the AI bill audit: a startup founded by a former Oracle director combed $34 million of AI spending across sixty enterprises and found $1.7 million in overcharges — mostly Claude Code usage. In framework terms, the payment layer is fully built (API billing) while the identity layer is missing — enterprises still manage AI consumption with human-budget instruments. When agents carry independent wallets and billing, does that ease or explode? Ford rehiring 350 veteran engineers: AI could not replace decades of seasoned judgment, so Ford brought retirees back. That is the physical-world trust layer remaining human. The digital layers — mail, payments, information processing — are exactly where agents are clearing customs fast. The Copilot CLI versus Claude Code token war: Microsoft pulled internal Claude Code licenses over token consumption and Uber torched a year’s AI-coding budget in four months. This previews the payment layer’s deepest problem: when agents consume resources autonomously, the very definition of “budget” must be rewritten. Today’s per-seat licensing is human budget logic stapled onto AI.
So What Should You Do
The three-layer infrastructure is not coming. It is running. Concretely: If you build agent products — look past model capability. Agent Card identity, A2A integration, AP2 payment rails: infrastructure compatibility decides whether your agent can join the larger ecosystem, and that matters more than another 5 percent of reasoning accuracy. If you run enterprise security — watch the “fake reply” traffic curve. Attackers A/B-test phishing with AI while your employees auto-reply with AI; both trends land in the same inbox. Design the next security policy at the identity layer — distinguishing AI mail from human mail — rather than the content layer. If you are a founder — the agent wallet is the thinnest, bluest of the three layers. Humans have Stripe and Alipay. Agents have AP2 on paper: who ships the developer-friendly SDK? Who handles settlement and reconciliation? Who helps enterprises audit a Copilot CLI token bill? If you are an ordinary professional — no need to panic about replacement, but do make one cognitive switch: whoever is on the other end of your inbox may no longer be human. That is not an alarm; it is a fact. And understanding it will shape how you judge sources, manage trust, and decide which messages deserve a personally written reply more than any specific AI skill will.
Sources: 36Kr, “Agents just got their own mailbox and wallet, and humans’ inboxes are already overflowing”; GeekPark.
