Izzo: Your daily standup just turned into three Slack threads, two action items got lost in the noise, and your AI agent is asking you to re-explain the same codebase conventions for the fifth time this week.
Izzo: You're listening to Exploring Next, episode two-sixty-two. I'm Izzo, and with me is Boone. Today we're digging into PromptQL — a tool that's trying to turn all that coordination theater into actual work.
Boone: And when I say actual work, I mean your Slack conversation about a checkout bug automatically delegates to Claude Code, which inherits your team's context, pushes a fix, and updates your internal wiki. No re-explaining required.
Izzo: Okay, that sounds too good to be true, but there's a real problem here. A Hacker News thread went viral in February asking OpenAI to build their own Slack specifically for AI agents. Three hundred twenty-seven comments of pure frustration.
Boone: Right, because current chat tools are fundamentally designed for human-to-human coordination, not human-agent or agent-agent workflows. The context just... evaporates.
Izzo: So PromptQL is tackling this head-on. They're a Hasura spin-off that pivoted from AI data analysis into what they call an AI-native workspace. Boone, walk me through what makes this different from just adding a chatbot to Slack.
Boone: The core innovation is their Shared Wiki architecture. Traditional LLMs forget previous interactions or hallucinate from stale training data. But here, every conversation is teaching a living, internal Wikipedia that accumulates context organically.
Izzo: No more documentation sprints?
Boone: Exactly. When your engineer fixes a bug or your marketer defines what a recycled lead means, they're not typing into the void. They're building institutional memory that any agent can inherit later.
Izzo: But how do you prevent the AI from learning junk? Like someone's doctor's appointment reminder from 2024?
Boone: Human-in-the-loop verification. Users have to explicitly hit 'Add to Wiki' to canonize facts. So the system learns from intentional knowledge sharing, not random chatter.
Izzo: Smart. And they're using a virtual SQL layer instead of copying data around?
Boone: Yeah, this is clever. Instead of replicating data from Snowflake, Postgres, Stripe, wherever — they query it in place. Nothing gets extracted or cached. Your data stays exactly where it lives.
Izzo: That's huge for enterprises. McDonald's and Cisco aren't exactly thrilled about the idea of 'just connect your data' if it means copying everything to some vendor's servers.
Boone: And they've got attribute-based access control at the infrastructure level. If a regional manager asks for vendor rates across all regions, the AI redacts columns they can't see — even if the underlying model 'knows' the answer.
Izzo: Hold on, let's get concrete here. What does this actually look like in practice?
Boone: Their demo shows an engineer posting about a failing checkout in #eng-bugs. Instead of tagging a human SRE, they delegate to Claude Code via PromptQL. The agent knows 'EU payments switched to Adyen on Jan 15' because that was wikified weeks ago.
Izzo: So it's inheriting team context automatically.
Boone: Right. Within minutes, it identifies a currency mismatch, pushes a fix, opens a PR, and updates the wiki for next time. The context compounds.
Izzo: I'm giving this a solid A-minus just for solving the 'explain our deployment process for the hundredth time' problem. But what about the business model? Are they doing per-seat like everyone else?
Boone: Nope, consumption-based. They call them Operational Language Units — OLUs. CEO Tanmai Gopal says charging per seat penalizes companies for connecting their whole team.
Izzo: That's... actually really smart positioning. Most tools want you to limit access to control costs. This incentivizes the opposite.
Boone: And for enterprise storage, customers get their own VPC. Any data the AI saves goes to your S3 bucket using Iceberg format. Total data sovereignty.
Izzo: Boone, be honest with me. Is this actually going to replace Slack and Teams, or is this just another integration that dies in the enterprise software graveyard?
Boone: Well, Gopal claims they've shut down their internal Slack entirely. And honestly? If you can turn every conversation into actionable work instead of coordination theater, that's a pretty compelling value prop.
Izzo: The timing feels right too. Companies are realizing that chatting with PDFs isn't enough. They need AI that can act, but they can't afford unsupervised agents running wild through their systems.
Boone: It's that middle ground — AI that learns like a teammate and executes like an intern, but with enterprise guardrails. High-stakes actions still require human approval.
Izzo: Alright, I'm convinced this is worth exploring. What should people actually go build or try? First, check out their virtual SQL approach — that's applicable even if you're not using PromptQL. Look into Apache Iceberg for data sovereignty patterns in your own agent workflows. Second, experiment with human-in-the-loop knowledge capture. Even in your current Slack, try maintaining a team wiki that agents can reference. Make it a habit to canonize useful context. And third, play