MatIntellect

⚡️ Durov finally shipped natively what we'd been hacking together through a user account. On May 7 Telegram opened up Chat Automation - you connect a bot to your profile, and it answers messages for you

What Chat Automation is:
You go to Settings, open the Chat Automation section, pick a bot - and from then on it works as a personal secretary in your name. You decide where it's allowed to reply: exclude all chats with contacts, allow only new chats or give access to specific groups. Granular permissions down to each individual chat - not "turn it on everywhere and pray", but real control. For developers Telegram described a "secretary bot" standard in the Bot API - meaning anyone can build their own AI assistant and distribute it via @username, with no workarounds and no user-account magic

What else shipped on May 7 in the same release:
✨ Guest Bots - you mention an AI bot by @username in any chat, even if it's not there - a quick fact-check or generation right in the group
✨ Bot-to-Bot Communication - bots can now message each other with mutual opt-in, laying the groundwork for autonomous multi-agent scenarios
✨ Streaming Text - the bot types its answer gradually like ChatGPT, you don't wait for the whole answer at once
✨ AI writing styles - you write a prompt with your style, share a link, anyone plugs in your "tone of voice" with one tap

The line has been visible since late March - that's when they rolled out Managed Bots, a system where one manager bot creates personal agents for each user in two taps. Now they've added a layer on top: an autonomous mode that works for you, bot-to-bot communication, guest access. Telegram is clearly building infrastructure for full-blown AI agent workflows inside the messenger, not for "an FAQ chatbot"

What this means in practice:
🟢 The end of user-account hacks - no more running a secretary on Telethon at the risk of a ban, a native mechanism with permissions and audit
🟢 A marketplace of secretary bots - TelyAI, Tanka and Chat Data already build these, now they get a standard interface and a distribution channel
🟢 Bot-to-bot opt-in - a deliberate design against spam chains, both sides have to explicitly turn the mode on, otherwise nothing happens

💼 Telegram is opening the Bot API to autonomous AI agent scenarios for the first time - until now everything was built on user-account workarounds

What my secretary can do now

My personal agent Jes has lived in Telegram for 134 days. I write to it or send a voice note, and it handles 10 scheduled jobs on its own and hands tasks to 7 other bots: 165 tasks have gone out this way since July. It comes to me only for a choice or a sign-off. Anything risky, paid or public waits for my yes: deleting data, spending money, messaging clients and publishing

The most useful part of a secretary isn't smart answers but the small things that need attention rather than expertise: remind a client, check a status, follow up three days later. Each takes a minute, but together they eat the day. For me that part is covered by an agent working with a plain program: every morning I get a day plan from my calendar, tasks and reminders, 127 mornings in a row, and 240 reminders have gone out in that time

What to give a secretary first

Start with reminders and summaries: they have a clear input and a clear result, and a mistake is easy to spot. Leave messaging clients in your name for later, and only with confirmation. A good secretary saves time rather than creating new things for you to check

💭 I spent half a year building my own version of this kind of secretary via a user account and Telethon - it works, but the ban risk was always hanging over it. Now Telegram says "here's the standard, do it via the Bot API, we'll even open a store". My self-hosted builds instantly become less unique, but more people will try AI agents as a category - and that's more interesting than any single feature in my stack

Share:

No comments yet

Leave a Comment

Fields marked with an asterisk (*) are required