Slack alumni launch Scuba to kill off AI agents that know nothing about your company

Scuba emerges from stealth with $34 million raised to develop a shared memory between collaborators and AI agents, compatible with ChatGPT and Claude.

AI agents can generate code, analyze documents, or prepare presentations, but rarely know the decisions, preferences, and discussions that actually shape a team's work. Scuba wants to fill this gap with a collective memory accessible to team members and their assistants alike. The startup is emerging from stealth mode with $34 million raised and one ambition: enabling AI systems to work from what the company already knows, rather than starting from scratch with every request. Context, the new problem for AI agents

For Noah Desai Weiss and Dolapo Falola, the two co-founders, the challenge no longer lies solely in model capabilities. It also stems from the information available to them to accomplish a task.

Within a company, knowledge is scattered across Slack conversations, working documents, meetings, screenshots, and informal exchanges. A decision might be made in a thread, only to become unfindable a few weeks later. AI agents face the same struggle: without access to this history, their responses can ignore already established constraints or choices.

Scuba therefore proposes building a shared memory that retains not only information, but also the context in which it was created. A memory that organizes itself without manual effort

How it works relies on Captures, which allow saving articles, PDF documents, videos, messages, images, or voice notes. The application notably supports audio recordings up to 90 minutes, complete with a transcript.

These items are automatically analyzed and grouped into Collections, organized by project, client, or topic. Team members can share them to find important information without manually maintaining internal documentation.

Scuba also offers natural language search, personalized summaries, and daily briefings capable of flagging developments, trends, or contradictions between different pieces of information.

The goal is to maintain up-to-date knowledge as work happens, rather than an accumulation of quickly outdated documents. ChatGPT, Claude, and Cursor can leverage this memory

Scuba isn't looking to replace the AI assistants companies already use. Thanks to the Model Context Protocol (MCP), the platform can make its Captures available to tools like Claude, ChatGPT, and Cursor.

A coding agent can thus consult previously made technical decisions, while an assistant tasked with preparing a sales proposal can retrieve a client's preferences or takeaways from past meetings.

A Slack integration also allows prompting @Scuba to find information or save a discussion directly within a channel.

Every response generated in Scuba relies on the saved information, complete with interactive references to trace back to their source. The platform can also perform web searches when more recent data is needed. $34 million and founders hailing from Slack

Behind Scuba are Noah Desai Weiss, former Chief Product Officer at Slack, and Dolapo Falola, previously VP of Engineering at The Browser Company. Both have also worked at Google and Foursquare.

The company announced it raised $34 million across two funding rounds. Accel led the pre-seed before co-leading the seed round with IVP, alongside BoxGroup, Conviction, Homebrew, and Scribble Ventures.

Several entrepreneurs are also participating as individual investors, including founders of Slack, Instagram, Stripe, LinkedIn, Salesforce, and Dropbox.

Scuba claims to have tested its product with hundreds of users during a six-month private beta. Its team of ten explains using this shared memory to coordinate its projects with fewer than two hours of weekly meetings. However, these results stem from their own internal experience. A limited beta for a broader challenge

Sign-ups for the Scuba beta are open, but teams are being onboarded gradually. The company has not yet shared public pricing plans or a timeline for general availability.

The question of privacy also remains: centralizing discussions, internal documents, and strategic decisions to make them accessible to multiple agents requires precise management of permissions and sensitive data.

Scuba is part of an evolving landscape in how professionals use AI. As models become capable of executing more tasks, the quality of the context they receive becomes as important as their own capabilities. The goal is no longer just to generate a relevant answer, but to produce the one that truly aligns with a team's history, constraints, and choices.