It still cannot ring the plumber.

That gap is bigger than it sounds. A huge slice of small business still only happens by phone: getting three quotes, chasing an unpaid invoice, booking a job, checking whether something is actually in stock, getting a human at a utility company. Nobody automated it, because agents had no way to pick up a phone.

Two weeks ago that changed. I connected it on Sunday, gave it one instruction, and listened to the transcript of my AI ringing a supplier and getting a price.

Setup took under two minutes. It is currently costing me nothing. Here it is.

The 20-second version: DialMCP is a connector you add to Claude in about two minutes. After that, you can say "call this number and find out X" and your AI places a real call from your own verified number, talks to a human, and comes back with a transcript, a recording, and a plain summary of what was agreed. It launched July 29 and costs nothing for US and Canadian numbers during launch.

Time: two minutes to connect. You need: Claude (or ChatGPT, Cursor, Codex) and your phone for one SMS code.

The old way: you need three quotes. That is three calls, two voicemails, one hold queue, and a follow-up you forget. So you get one quote and accept it, or you do not bother at all.

The replacement: you hand the job over. It presses through the phone menus, waits on hold without complaining, and if it hits voicemail it hangs up rather than leaving a message (DialMCP).

Here is the part I did not expect. It is not hiding what it is. The first thing anyone hears is that it is an AI assistant calling on your behalf and the call is recorded (DialMCP). The calls go out from your real number, so the person can ring you straight back.

That number becomes your caller ID. No API keys, nothing to configure (DialMCP). The same URL works in ChatGPT developer-mode connectors, Cursor, and Codex.

There is no charge during launch for calls to US and Canadian numbers. Outside those two countries, this one is a look-ahead for now.

Try something low stakes first so you can hear how it handles a real person.

What you get back: a full transcript, a link to the recording, and a structured result: what happened, what was agreed, and anything needing follow-up (DialMCP).

Read the transcript. That is the moment it clicks, because you can see exactly how it handled the questions it was not briefed on.

One thing worth knowing: the objective is doing all the work. "Call and ask about prices" gets you a vague call. The specific version above gets you a usable answer. That is the whole skill, and it is what most of the Pro section is about.

One call is a novelty. The value is in knowing which jobs are worth giving it, and how to brief each one so you get an answer instead of a shrug.

That is this week's Pro section: eight calls that are worth handing over, the brief template that makes each one work, the guardrails, and the other connectors that let AI act in the physical world, including one that posts actual letters.

🔒 Inside the Pro section:

The Phone Agent Pack. Connecting takes two minutes. Knowing what to hand over is the part that saves you a morning a week:

Also in your library: the Skills Starter Pack, the Selling System, the Computer Use Playbook, the Unattended Job Pack, and 110+ up to date workflows. Every workflow from every issue, permanently.

AI images have had one embarrassing flaw since the beginning: they could not spell. Any sign, label, or logo came out as convincing-looking gibberish, which made them useless for anything real.

Meta's Muse Image fixes it. It renders readable, correct text inside a picture, and it can even generate working QR codes, because it writes and runs code during generation rather than guessing at letter shapes (Meta).

That sounds like a small thing. It is the difference between AI images being a toy and being usable for a menu, a poster, a price list, an infographic, or an ad with your actual offer written on it. It is in the Meta AI app at no cost.

⚖️ Europe's AI disclosure rules went live on August 2. Chatbots and voice agents must now tell people they are AI, and AI-generated content needs labels and machine-readable marks, with fines up to 15 million euros or 3% of worldwide turnover.

🔌 The plumbing behind every AI connector got rebuilt. The Model Context Protocol shipped its biggest revision since launch on July 28, dropping sessions entirely so connectors run on ordinary web infrastructure. One of its co-inventors called it the most important release since remote connectors first shipped.

📈 The clearest sign yet that AI spending is real, not hype. Palantir reported on August 3 that revenue rose 92.8% year over year, with its US commercial business up 149%. Businesses have moved from testing AI to paying serious money to run it.

📅 Two ChatGPT models retire this month, and a pricing change lands on the 31st. A dated tracker of model events lists an endpoint shutdown on August 17, two ChatGPT retirements on the 26th and 30th, and a price change on the 31st. If any saved prompt, automation, or connector of yours points at a specific model version, check it before the end of the month.

Quick version: On August 2 the EU switched on rules requiring AI systems to tell people they are AI, and AI-generated content to be labelled. Fines reach 15 million euros or 3% of global turnover. It applies to anyone whose AI output reaches people in the EU, so plenty of businesses outside Europe are now in scope without realising it. There is a significant carve-out for anything a human reviews and takes responsibility for.

The timing is almost funny. The same fortnight AI agents got the ability to phone real people, Europe made it law that AI has to say it is AI.

From August 2, the European Commission's AI Office and national regulators began enforcing the transparency rules in Article 50 of the AI Act. Chatbots and other interactive AI must tell people they are dealing with a machine. Deepfakes have to be labelled. AI-generated or altered content has to carry machine-readable marks so it can be detected automatically.

The penalties are not symbolic: up to 15 million euros or 3% of worldwide annual turnover, whichever is higher (Cooley).

Here is the part most people outside Europe have missed. The rules apply to providers and deployers placing AI on the EU market, or whose AI outputs are used inside the EU. Forbes put it plainly: this reaches US companies serving EU customers or employees. If you have European customers, subscribers, or staff, and you use AI in anything they see, you are closer to this than you think.

There is also good news buried in it, and almost nobody is reporting that part properly. The rules are narrower than the headlines suggest, and there is a carve-out that exempts a lot of ordinary business use.

Worth noticing what this says about where things are heading. The phone connector in this week's workflow already discloses at the start of every call, before any law told it to. The tools that will survive this are the ones that treated disclosure as the default rather than a compliance chore.

🔒 The Pro breakdown: the disclosure lines written out so you can paste them straight in, a decision tree that gets you to your answer in about a minute, and the carve-out that exempts most AI-assisted writing.

Workflow Library Updates:

NEW Resources:

Each resource lives permanently in your Pro account. Use them whenever you need them.

There is a list of phone calls you have been meaning to make. The quotes you never chased. The invoice you keep not ringing about. The supplier you have been meaning to check with.

Give one of them away this week and listen to the transcript. Two minutes to set up, and the first call is the one that changes how you think about the rest of the list.

👇 What’s behind the paywall:

Pro members deploy AI in their work an average of 5-8x more often than free readers (based on reply data from past issues). The difference is having the exact setup, not the concept.