Killing Slack was the only way to make AI accurate

When we started to solve the problem of achieving AI that can work accurately on internal data and be fully trusted, we took 3 big bets that we bet the farm on:

Bet 1: Coding will be human-grade reliable

AI writing code is the only way to achieve the degree of reliability, repeatability and user-facing meta cognition that is trustworthy.

Instead of RAG, or tool-calling style approaches, we bet on making AI write code to solve problems.

Bet 2: Multiplayer AI is the path to shared context

Accuracy and trust when humans work with each other on intelligent tasks, is all about shared context. And capturing that context is impossible, and hence cannot be embedded into an AI agent by any one person or team.

We cannot separate the act of building context for an AI agent from the act of using the AI agent. Context is fundamentally spread and split across multiple people. Multiplayer AI is the only way AI will be reliable.

We bet that only a Wikipedia style operating model can maintain the kind of accurate and up-to-date information required for shared context.

Bet 3: Democratizing data with AI is only a security problem

As an industry we've never actually democratized access to all data and all systems before. It was never possible or feasible to do so before. So there's never been a technology platform to manage granular access control at scale to heterogeneous resources: databases, SaaS tools, internal APIs

We bet on RLS + ABAC as the simplest yet most scalable form of managing internal access control and authz, and built a virtual data layer that enforces those policies on anything you connect without having to move that data.

Solving this, we ended up with something that's given me goosebumps.

It looks like a Slack but connected to every single person and to every single data source, API & SaaS tool in the company.

You're probably already doing that today. Slack + Claude code web + Glean + custom agents + tools. So, no big deal right? 🤷

Except that it turns out that something just fundamentally changes when the workspace itself is an AI. People just start talking to the AI about anything and everything.

Shahidh found a box of encyclopedias just lying outside the elevator in our office, so he just asked PromtpQL:

Most AI tools today feel like they're bolted on top of "legacy" software (eg: SaaS tools) and ways of working (eg: messaging, meetings). And to all of us who're AI-pilled, these legacy foundations are feeling increasingly shaky and likely ngmi in a post-AI world.

It's no surprise that AI tooling is always in a state that sort of works, and then every few weeks we seem to have a "this changes everything" moment when LLMs improve, but we never quite reach that flow state with AI.

So we did All The Things that we thought from first principles were necessary to truly unleash AI on ourselves and our systems. And then, we pulled the plug on Slack.

And now, it suddenly feels like we're a completely different team.

Describing the Vibe Shift

Conversations just naturally become the work.

PromptQL spins up your Claude Code / Codex, PR comes through, team mates review together, you ship, PromptQL monitors prod. It's a good time.

Bug -> Solve collaboratively -> Review -> Ship -> Monitor in prod

Goodbye DMs

You stop DMing people and you stop waiting for them, because PromptQL already knows.

Ping a human and wait, vs ping AI and know instantly.

That one tribe feeling…actually happens.

Territorial behavior is an artifact of information asymmetry. When nobody shows up uninformed, because anybody can understand the full context in seconds, "stay in your lane" stops making sense.

Mike leads our sales team and Zack runs revenue marketing, both of them quite far from product, but now they're just plugged in.

Everyone's job description is to just get to the shared outcome.

No 👏more 👏 coordination 👏theatre 👏

Doing anything involving multiple people is hard, especially if its something new.

Now, No meetings about work. No issue tracking. No "alignment" discussions. You don't have to explain why you're doing something.

Not having meetings where there's inevitable posturing, and instead just getting things done together, actually brings people closer together.

All The Things we did

We built 3 things that allowed everything to come together:

1. A Multiplayer AI surface for web/desktop/mobile

2. An AI assisted wiki

3. A permissions system that makes complex access control on disparate systems easy

It's still in preview, but we're getting ready to start to share it with people who're ready to do push AI to see how fast they can really go.

See you on the app!