Elon Musk Launched an AI That Makes Calls for You. What Comes Next? | A New Era of Phone Calls
What will change when bots start making calls for people — will companies answer them with bots, who can you trust with your own call — and what does Anthropic's listing mean for the price of AI?
XAI's Grok Bot already calls on a person's behalf — to five insurers to compare terms, or to an internet provider to find out what went wrong yesterday — while Anthropic and OpenAI have no such scenario and Gemini is testing it behind closed doors. Anna checked: the feature doesn't work in Europe, because a person has to be told they're talking to AI, and in the US it doesn't work in Washington and Illinois. In a company's place Ilnar would put a classifier at the entrance and transfer the bot to a bot, Alexander predicts a market for manipulating bots, and the hosts see a viable scheme in your own Dot that sorts incoming messages on your side and decides by your rules what to pass on and how. Anna hangs up when she hears a bot and is willing to talk to one only about support and reminders; on hiring, the argument runs between a thousand CVs no recruiter could ever call and ten candidates for a CEO post. In the second part Ilnar reads the S-1 as showing a conservative Anthropic with enough money of its own for its obligations — and that is why it can't hand out limit resets the way OpenAI does; Alexander objects that the obligations come to almost a trillion, heavy users are loss-making, and today's level of models will soon cost tens of thousands of dollars. Then come Boris Cherny's proposal to erase and rewrite Claude Code, Dot on the iPhone button, the limit-reset trap at OpenAI, bans and KYC at Anthropic, and Claude adapting an American's Russian-language product to Russian law.
What to watch for
Key takeaways
Anna tested ChatGPT's new GIF generation: technically it's brilliant and very fast, but from a simple request like "my wife spends too much money" the result is primitive. Ilnar advises giving a reference: ask ChatGPT to describe the style of an image you like — exposure, colour grading — and add your own meaning to that prompt.
Grok has released a caller: you can tell Grok Bot to call five insurance companies and agree terms for car insurance, or to get through to your internet provider and find out why there was a problem yesterday — to hang on the phone instead of you. It came out just a couple of days earlier, there was a huge wave of reposts on X, and Elon Musk keeps retweeting them.
The idea is trivial, but neither Anthropic nor OpenAI lets you set a scenario in which a bot calls, comes back to you mid-conversation with an extra question and checks it with you. Gemini has such a feature, but it's being tested in closed mode with a limited group — perhaps that's why it hasn't launched.
Anna believes such a feature should be banned in Europe: by law a person who hears AI must be told it is AI. She quickly checked: the feature doesn't work in Europe, in the US it doesn't work in Washington and Illinois, and it went live on the fourth of November.
In a company's place Ilnar would simply transfer a calling bot to a bot: let them talk to each other. You need a classifier at the entrance that recognises a bot by intonation and technical signals and sends it to a specially trained "Vasya" who knows how to con ChatGPT, or to a more advanced version of ChatGPT so that it upsells. Alexander sees a market for manipulating bots in this.
The Grok caller reminded Ilnar of Sergei Lukyanenko's story "My Dad Is an Antibiotic": auto-diallers appear in it, and two friends agree never to auto-call each other because it's unethical. With Grok Bot, he believes, such agreements will have to be put into practice: he wouldn't want someone's Grok Bot getting through to him.
Alexander doesn't mind another person's bot calling about something reasonable — the iron, being late for the episode, his mum's trip. Ilnar thinks a more viable scheme is to leave the message with your Dot, which on your side decides by your rules how and when to pass it on: then control is yours. Alexander takes it to an incoming-call filter that puts Ilnar through but not "Vasya, Ilnar's Grok Bot".
Anna always hangs up when she realises it isn't a human talking to her, and doesn't answer contractors on LinkedIn if a bot wrote the message — but she will always answer a human recruiter. She divides the scenarios: technical support and reminders are acceptable, while partnership, choosing among twenty options or selling a service require human contact.
Ilnar doubts a bot is needed to hire a CEO: about ten people are considered per post, so it's easier to call HR. Alexander objects that the funnel is hundreds and thousands of CVs and you can't pick a director from a two-page CV; no recruiter will call a thousand people, and a system would let you pick ten more broadly and seriously. Anna notes that a system has its own algorithm built in too.
Ilnar calls the Anthropic IPO a huge plus: the published S-1 form shows profitability, expenses and investment in compute — about half a trillion up to 2030, as he recalls. The company planned conservatively for buying compute, while its users multiplied, which is why OpenAI hands out limit resets and Anthropic can't.
By Ilnar's estimate, Anthropic has raised more than a hundred billion, will raise another hundred to a hundred and fifty at the IPO, has about a hundred in its accounts, and with current profit that's enough for its obligations. Alexander objects: the obligations come to almost a trillion, and they'll float some ten percent, maybe four or five. Revenue, he says, was five billion in the first quarter of 2026 and sixteen for the half-year.
Ilnar believes inference is cheap: a two-hundred- or five-hundred-dollar subscription pays for the servers, and what's costly is new models and research. Alexander is sure of the opposite: all heavy users are loss-making, a company can't stop developing, and today's level of models, once it becomes a commodity, will cost tens of thousands of dollars.
Boris Cherny, one of the main developers of Claude Code, said Claude Code is outdated and has to be erased and completely rewritten. Alexander recently said the same on the board of a company with a huge infrastructure: rewriting everything used to be impossible in big corporations, and now it's the only salvation.
The Dots update that Tibo wrote about lets you assign Dot to the iPhone's extra button and call it up directly, without scrolling through ChatGPT. Claude Duet made no impression on Alexander: from Anthropic it looks like half-baked copying of Dots, and he stopped using it.
At OpenAI pressing reset pushes the weekly reset back seven days again: reset a day before the scheduled one, and the next moves a week later. Alexander ate more than ten billion tokens on Codex in a month, and for normal work he needs the volume of the Claude and ChatGPT Pro subscriptions every day — in money, hundreds of thousands of dollars a month.
OpenAI officially allows several accounts, Anthropic doesn't and keeps up mass bans around the world. When Anthropic asks whether you're a company or a private person, Alexander advises not to rush: using an ordinary account in your own business, if you're essentially on your own, is allowed. The company does a lot of KYC and fights above all against use against humanity, the US and Anthropic itself — against distillation.
Claude has long known Alexander is American, yet adapts any Russian-language product of his to Russian law straight away, and a Belarusian landing page to Belarusian law. Ilnar explains it by the training dataset, AI advises Anna on everything from Poland's point of view, and Alexander thinks the model primarily sees where you are. You need to know the rules, he concludes, but without paranoia.
What this episode is about
A Sunday ToTheMoon episode in which Alexander Volchek, Ilnar Shafigullin and Anna Volchek meet as a trio — Max Grigoriev isn't there this time. It opens with Anna's test: ChatGPT has learned to make GIFs; technically it's done brilliantly, but from a simple request the result is primitive. Ilnar advises giving a reference instead of a bare prompt: ask ChatGPT to describe the style of an image you like — exposure, colour grading — and add your own meaning to that prompt. The morning of the recording began with a sharp drop in the tech market, but Alexander is sure it will turn around.
The main news is the caller in Grok. You can tell Grok Bot to call five insurance companies and agree terms, or to hang on the phone with an internet provider; the episode plays a recording of such a call to the insurer Progressive. There was a huge wave of reposts on X, and Elon Musk keeps retweeting posts about Grok Bot. At Anthropic and OpenAI, Alexander says, you can't set a scenario in which a bot calls and comes back to you with a question mid-conversation, while Gemini is testing a similar feature in closed mode.
Anna believes this should be banned in Europe: by law a person must know they're talking to AI. She checked — the feature doesn't work in Europe, in the US it doesn't work in Washington and Illinois, and it went live on the fourth of November. Alexander is sure companies will get scripts for when a bot calls, and a market for manipulating bots; in a company's place Ilnar would put a classifier at the entrance and transfer the bot to a bot or to a specially trained "Vasya". Telling a bot by its voice isn't easy: Alexander's ChatGPT Voice talks in a way that made a friend ask what he'd done to it.
Anna recalls the film "Mercy" with its AI judge: a mind that weighs bare arguments without emotion is another level of negotiation. The caller reminded Ilnar of Sergei Lukyanenko's story "My Dad Is an Antibiotic", where friends agreed never to auto-call each other; he wouldn't want someone's Grok Bot getting through to him. Alexander doesn't mind if the matter is reasonable — the iron, being late for the episode, his mum's trip — and Ilnar proposes a scheme in which your own Dot sorts messages and decides by your rules how to pass them on. Sam Altman, Alexander recalls, wants to give everyone their own compute, and our models will talk to each other.
Anna always hangs up when it's not a human talking to her and doesn't answer LinkedIn messages written by a bot; she considers only support and reminders acceptable. All three argue about hiring: Alexander pictures a bot that calls a CEO candidate back exactly forty-seven minutes later and picks ten out of a thousand, Anna calls it idealisation, Ilnar unnecessary with ten candidates per post. Ilnar tells of data-scientist acquaintances who hand their dating-app chats to a system on ChatGPT and Gemini, and Anna of Greenhouse's data: over a year, half of American candidates had an interview with AI. Alexander, who dislikes calls, talks about multichannel service and about Airbnb.
The second part is the Anthropic IPO. The valuation is expected to exceed two trillion; in its SEC filing the company had to write that AI will increasingly create itself, while Sam Altman writes against this backdrop that AI must not be allowed to govern itself. From the S-1 Ilnar sees a conservative company — about half a trillion on compute up to 2030, more than a hundred billion raised, a hundred to a hundred and fifty expected at the IPO, a hundred in its accounts — and explains by this why OpenAI hands out limit resets while Anthropic can't. Alexander objects that the obligations come to almost a trillion, and cites revenue: five billion in the first quarter of 2026, sixteen for the half-year.
Will AI become expensive? Ilnar believes inference pays for itself and only new models are costly; Alexander, that all heavy users are loss-making and that today's level of models will cost tens of thousands of dollars once it becomes a commodity. As examples he cites OpenAI's engine for creating hardware, YouTube, Amazon, WhatsApp and Telegram, which were loss-making for years, and Elon Musk's reply to complaints about tokens in Grok: "What did you expect?" Boris Cherny proposed erasing and rewriting Claude Code; Apple, according to Alexander, cut its AI spending per developer to two hundred dollars a day.
The end is practical. Dot can now be assigned to the iPhone button, while Alexander has dropped Claude Duet as a half-baked copy of Dots. Pressing reset at OpenAI pushes the scheduled reset back a week, and Alexander has eaten more than ten billion tokens on Codex in a month. OpenAI officially allows several accounts, Anthropic doesn't and keeps up mass bans and KYC; Alexander advises not to rush answering whether you're a company or a private person. The episode ends with Claude adapting an American's Russian-language product to Russian law and advising Anna on everything from Poland's point of view: you need to know the rules, but without paranoia.
The episode shows that a bot's call is less a matter of technology than of rules: who may call through AI and when, what a company will answer a bot and how to keep control on your own side — while the argument over the Anthropic IPO is a reminder that today's price of AI is temporary.
Episode transcript
The episode is in Russian; below is an English reading guide to the transcript (the full EN transcript is a machine translation). Voice matching applied to 255 segments: 0 identified, 0 mixed, 255 probable, and 0 unresolved.
Read transcript on a separate page
Loading…