The CEO Who Asked Anthropic to Kill Slack

When the CEO of a $5.6B data infrastructure company publicly begs an AI lab to “make a new Slack,” you know something’s broken.

George Fraser, Fivetran’s co-founder and CEO, published a blog post this week titled simply: “Anthropic, please make a new Slack.” His argument is brutal in its clarity: Slack will be the Waterloo of closed data.

The problem isn’t that Slack is bad at messaging. It’s that Slack has become the operating system of modern work — and all that institutional knowledge is locked behind a paywall and an API that treats your data like it’s not yours.

---

The Data Prison

Here’s the core of Fraser’s critique:

Your company’s Slack is a goldmine. Every decision, every debate, every context-rich conversation — it’s all there. When an employee leaves, their DMs vanish. When you need to train an AI on your company’s institutional knowledge, you’re stuck exporting what Slack lets you export.

Which isn’t much.

Slack’s business model depends on keeping your data inside their walled garden. They sell you search, they sell you retention, they sell you “AI features” that are really just their AI running on your data, inside their infrastructure.

But here’s the thing the AI revolution is exposing: Your data wants to be free. Not free as in beer — free as in “portable enough to train your own models, run your own agents, and own your own infrastructure.”

---

Why Anthropic?

Fraser isn’t asking for a Slack clone from Microsoft (they tried with Teams). He’s asking Anthropic because:

1. They understand agents. The future isn’t chat — it’s autonomous AI assistants that can read your entire company history and act on it.

2. They’re not Google or Microsoft. Big Tech’s incentive is to lock you into their ecosystem. Anthropic’s incentive (right now) is to make Claude indispensable.

3. They have the technical capability. Building a Slack competitor from scratch with AI-native architecture is easier than retrofitting AI onto a decade-old chat platform.

The irony: Slack’s biggest moat is your data. But that moat becomes a liability when the AI revolution demands data liquidity.

---

The Self-Hosted Angle

While we wait for Anthropic to answer Fraser’s plea, there’s a more immediate question: Why don’t we have a good self-hosted Slack alternative?

The options that exist are… fine:

• Mattermost: Enterprise-focused, feature-complete, but feels like it was designed by committee
• Rocket.Chat: Open source, but the UX is stuck in 2018
• Zulip: Unique threading model, but the learning curve is real
• Matrix/Element: Decentralized, but the onboarding is rough

None of these are “AI-native.” None were built with the assumption that your chat history is training data. All of them treat messaging as the product, not the data infrastructure layer.

---

What an AI-Native Slack Would Look Like

If someone wanted to build a true Slack killer — self-hosted, AI-native, data-sovereign — here’s what it would need:

Data-First Architecture

• Every message, file, and reaction stored in a queryable format
• First-class export APIs (not an afterthought)
• Built-in vectorization for semantic search
• Your data, your infrastructure, your rules

Agent-Native Design

• AI agents as first-class citizens (not bolted-on bots)
• Agents can read channels, search history, and take actions
• Permission controls that work for both humans and agents
• Audit logs for every agent action

Modern UX

• Real-time sync that doesn’t suck
• Thread support that actually works
• Mobile apps that don’t feel like afterthoughts
• Integrations that don’t require a PhD to configure

Self-Hostable by Default

• Docker/K8s deployment in 15 minutes
• No “cloud required” features
• BYO LLM (Claude, GPT-4, or local models)
• Usage-based costs, not per-seat ransom

---

The Bounty Case

Slack’s pricing is the classic SaaS trap: per-seat pricing that scales with your success. A 100-person company paying $12.50/user/month is spending $15,000/year on chat. A 500-person company? $75,000.

But the real cost isn’t the subscription. It’s the opportunity cost of your data being trapped.

Every day your institutional knowledge sits in Slack’s cloud is a day you can’t:

• Train custom AI models on your company’s context
• Run autonomous agents that understand your business
• Build internal tools that leverage your conversation history
• Migrate to a better platform without losing everything

This is exactly the kind of bounty Bountymon exists to track. The moat is real (network effects, integrations, UX). But the vulnerability is also real (closed data, per-seat pricing, AI-native competitors).

---

The Bottom Line

Fraser’s right: Slack will be the Waterloo of closed data. The company that builds an AI-native, data-sovereign alternative won’t just win the chat market — they’ll become the infrastructure layer for the entire AI-powered enterprise.

Whether that’s Anthropic, an open-source project, or someone we haven’t heard of yet, the opportunity is massive.

And the bounty? It’s not just about saving money on chat. It’s about owning your own institutional memory in the AI age.

---

Think you can build this? We’re tracking bounties for Slack alternatives that prioritize data sovereignty. Check the bounty board for requirements.