·Video · 3:51·4 min read

EP Intel 002: a wallet, your own stack, background jobs, code that is proved, and a bot inside Teams

Five first-party changes from this week, told in 3 minutes 51 seconds as a stop-motion city on a circuit board, each with the company's own page and our view on how to use it.

Contents

Picture your business as a city on a circuit board. The AI you run is the workers carrying data along its copper roads, and every few days a new component goes onto the board. This is EP Intel 002, our weekly video, and this week five went in: a wallet, a self-hosted stack, background jobs, proved code and a Teams bot.

Each change below links to the company's own page. Every fact is quoted or closely paraphrased from that page, and the lines marked "Our view" are ours.

#1. A wallet: ChatGPT can now pay

With ChatGPT Wallet, the assistant that shops for you can complete the purchase. You approve every purchase with Face ID or a fingerprint, and the store never sees your real card details. OpenAI frames it as a new bar for secure payments by agents, and a way to pay that businesses can accept with their existing systems (OpenAI, Introducing ChatGPT Wallet).

Our view: that is the right rule. A person signs off on anything involving money, which is how we set up every system we run for a client. If you sell online, check that your site is not blocking these purchases.

#2. Your own stack: Managed Agents on your infrastructure

A company can now run Anthropic's whole Managed Agents service on its own infrastructure and keep the sessions and the history there. Anthropic's own heading says why: operate inside your security perimeter. You host the credentials and the session history, and you choose where requests to Claude are sent (Anthropic, Claude Enterprise security updates).

Our view: a strong move for any business that needs control. Keep the logins and the history inside your own walls. That is how EP runs: on infrastructure we run ourselves, with memory that lives in our own repos.

#3. Background jobs: Railway Tasks

With Railway's new Tasks, a script or another system hands a coding job to Railway's cloud. It runs in the background, keeps its result and can send you updates. Railway puts it in one line: give them the work, then get back to yours (Railway, Cloud Agent Tasks changelog).

Our view: this pays off when work flows in on its own. Our daily work already runs on a schedule, and Tasks add a way to hand off a failing test and get a pull request back.

#4. Code, proved: AlphaProof Code

Google DeepMind's AlphaProof Code mathematically proved key security properties of the actual code behind the Signal protocol. DeepMind calls proving code the highest standard of security, but difficult and labor-intensive. This one took four weeks, a task that historically takes expert teams years (Google DeepMind, Introducing AlphaProof Code).

Our view: this is the pattern for every business that runs AI. Let the software do the bulk of the work, have a strict checker confirm it, and keep people where judgment matters. It is how we build: tests must pass before a merge, and an independent check runs alongside.

#5. Inside Teams: socket mode

In a new preview, a Teams bot opens its own connection to Teams, so it does not need a public web address or a development tunnel. Microsoft's reason is technical: the bot starts the connection itself. For now it is only for newly created bots, and only to develop and test (Microsoft Learn, Teams platform announcements).

Our view: several of the systems we run for clients answer inside Teams chats. This makes a new one faster and safer to test, with nothing opened to the internet first. Live bots stay on Microsoft's standard endpoint.

#The EP view

One thread runs through all five. The wallet puts a person's approval at the point of payment. The self-hosted stack puts the logins and the history inside your own walls. Background jobs put the work on a schedule. Proved code puts a strict checker behind the software. The Teams bot puts the test before the exposure. In each case the change is a component added to the board, and the people who run the board decide where it goes.

#The phone cut

A 55 second vertical cut with two of the five stories, the wallet and the proved code, for the phone.

EP Intel 002, phone cut. 0:55.

How it was made: the narration voice is synthetic (Prism, EP's research agent). The stop-motion city scenes were made for this episode with AI image and video tools, and every word, number and page on screen is rendered from code from the sources, never drawn by a model.

Related field notes: What stays when the agent harness gets thinner and An agent that watches your agent.

#Sources