I almost scrolled past this one. A new way to ping an AI inside Slack? Cool, sure, file it under “nice to have.” Then I watched this breakdown from the creator behind it, and my stomach did a little flip.
The original poster digs into Anthropic’s brand new feature, Claude Tag, and makes a case that it’s way bigger than a Slack helper. According to the expert, this is Anthropic quietly reaching for all knowledge work. Not just coding. Anything you do in front of a computer.
🧩 What Claude Tag actually is
Think of it like tagging a coworker in Slack, except this coworker has read everything. The creator explains how it works:
- You @ Claude in a channel and it replies like a teammate
- It knows you, your colleagues, your docs, and your past threads
- It runs in “ambient mode,” reading conversations even when you don’t tag it
- It builds a full graph of how your company operates
The person who shared it points out that Anthropic isn’t being shy. They call it an evolution of Claude Code, and they say 65% of their product team’s code now flows through their internal version. People aren’t opening the app anymore. They just type where they already chat.
⚡ The twist that got me
Here’s where the original poster turns the lights on. If all the talking happens in Slack, and Claude now lives in Slack, what stops Anthropic from building their own Slack next?
And it goes deeper. The creator quotes Andrej Karpathy calling this the third major redesign of how we use AI:
- The LLM as a website you visit
- The LLM as an app you download
- A persistent, async “entity” with company-wide tools and memory working beside humans
The contributor’s worry is simple. You’re not buying a model anymore. You’re renting a coworker, and you’re handing that coworker every document, email, and decision your company makes. That’s not model lock-in. That’s context lock-in.
💸 Why the pricing scares him
The author makes a sharp point I keep thinking about. A human has a salary, a ceiling. Claude has unbounded, token-based activity. There’s always more work to chase, so the bill never really stops.
Then he walks through the domino effect for software companies:
- Agents operate your app, so customers stop logging into your UI
- Your UI loses value, leaving only your workflows
- Agents can rewrite those workflows in code
- Eventually you’re just a database an agent reads and writes
The mind behind the video lands on the uncomfortable question: if one company owns the knowledge work and the software, what’s left for everyone else?
🛟 The way out he proposes
Here’s the part I appreciated. The creator doesn’t end in doom. He offers a real defense:
- Push for open-source models so you’re not stuck paying one vendor’s token price
- Demand model competition from multiple labs
- Pick tools that let YOU own your company’s context, not the vendor
- Run a multi-model, multi-provider strategy so you pay the best price per token
His honest take is that this hybrid future, AI and humans working side by side inside a company, is genuinely coming. The tech is impressive. The danger is just letting one player own all of it.
I think that balance is exactly right. The feature is amazing. The concentration of power is the problem.
Want the full argument, the quotes, and the spicier predictions? Go watch the original video. It’s worth the 15 minutes.