$ today_in_10_seconds
AT&T now processes about 45 billion AI tokens daily, up from 27 billion in March. That's a 67% jump in eight months, and it comes from redesigning processes around agents, not adding chatbots. Meanwhile, outside the firewall, platforms are starting to slam doors on agents they didn't invite.
$ lead_story
AT&T Rebuilds Business Processes Around Agents
AT&T has moved from experimenting with generative AI to systematically identifying business processes that can be redesigned around agentic capabilities. The key word is redesigned: AT&T is rebuilding workflows from scratch around what agents can do, not asking where AI can shave off a step. Its internal drag-and-drop builder lets employees automate multi-step processes across customer operations, network engineering, and internal support, and every workflow enforces human approval points, data isolation, and retention policies. For founders selling into large companies, the buyer is now asking which process you replace end to end, not which feature you add.
// why_it_matters: Agents are no longer a line item in the AI budget. They are becoming the company's operating model.
$ also_worth_attention
United Airlines blocks outside AI agents
United said it will restrict unauthorized personal AI agents and assistants from its booking and reservation systems, citing data security and unauthorized account changes. This is the first big consumer brand to formally close the door. Expect agent access policies to become as common as robots.txt.
link →AWS ships an emulator written entirely by agents
AWS released a free local tool that lets developers exercise 200+ AWS APIs without an account, and says agents wrote it with no human-written or human-reviewed code. That's a bold signal about trust in agent-written software, and a real test of what maintained means when no human knows the codebase.
link →ChatGPT gives agents a wallet
ChatGPT introduced a Wallet feature that lets AI agents shop and complete purchases on behalf of users. Put this next to United's ban and you have the week's central tension: model providers are giving agents the ability to buy, while merchants are deciding whether to let them in.
link →
$ bigger_picture
The trend behind the headlines
Agent adoption is splitting in two. Inside the enterprise, where the company controls identity, permissions and audit logs, agents are scaling fast. AT&T's token curve is the proof. Across company boundaries, agents act for a user on someone else's systems, and nobody has agreed on how to verify who they are or who is liable when something goes wrong. United's ban and ChatGPT's wallet are two sides of that same unresolved question. The next big agent infrastructure opportunity is probably trust between agents and the platforms they touch, more than smarter models.
Kitesurf (Cloudflare)
Cloudflare open-sourced Kitesurf, its stateless browser engine for AI agents, under the Apache 2.0 license. If your agents need to browse and act on the web, this gives you a production-grade engine you can inspect and self-host. That matters more as sites get pickier about which agents they let in.
$ subscribe --free
Get the next edition in your inbox
// DAIly Agent covers agentic AI three times a week
// free, no spam, exit anytime