AI Talk - How to build your AI-powered content system for social media
Most "AI content" advice stops at the prompt. This session is about everything after it: the pipeline that takes an idea from a real customer call all the way to a scheduled post, and the review gates that keep it from becoming an AI-slop engine.
Pauline Bertry and Alexander Mikhailov run that pipeline at Polar Bear, and they're walking us through their own system end to end. Pauline leads Polar Bear and owns the go-to-market side, so she decides what gets published, in whose voice, and whether a draft is good enough to go out. Alexander builds the machinery: the agents, the MCPs, the automations. You get the editorial half and the engineering half in the same session.
What We'll Cover
The five steps of their system
Start from real conversations: a daily agent pulls call transcripts and mines them for post ideas, so content comes from real work instead of a blank page
Package each voice as a skill: every author's voice and every format, post, caption or carousel, lives in a reusable AI skill so drafts sound like the person every time
Draft with agents, decide as humans: agents deliver proposals to Telegram or a web UI, and every draft is accepted, edited or rejected
Generate visuals inside the brand system: carousels, insight cards and infographics produced automatically from a codified visual identity, with no designer in the loop
Schedule through one calendar: approved posts flow to Buffer on a weekly multi-voice template, fixed slots, several voices, no copy-paste
Open Discussion
Your questions about where to put the humans, what to automate, and what breaks
You'll Leave With
A concrete picture of what a full content pipeline looks like, not just the drafting step
The idea of a voice as a reusable skill, and why that beats prompting from scratch every time
Where to put review gates so quality survives the automation
An honest read on which parts of this are worth copying and which depend on their setup
Who Should Come
Founders and marketers publishing regularly and running out of hours
Product and GTM people who want content to come out of real customer conversations
Anyone who has tried AI drafting and got something that sounded like nobody
Builders curious about the agent, MCP and automation plumbing underneath
Guests
Pauline Bertry: ex-McKinsey design director, now leading Polar Bear, where her focus is on go-to-market, including its content system and its AI agent team.
Alexander Mikhailov: AI product manager and serial AI builder with five tools on the market. He engineers the agents, MCPs and automations that run the pipeline end to end.