Slack Veterans Launch Scuba With $34 Million to Give Human Teams and AI Agents a Shared Memory

A New York startup called Scuba launched on Monday with $34 million in funding and a pitch aimed at a gap that has opened up as AI agents move into the workplace: the shared context that both human employees and the agents they work with need, but that currently lives in screenshots, PDFs, browser tabs, and conversation threads.

The company is led by Noah Desai Weiss, formerly chief product officer at Slack, and Dolapo Falola, who led engineering on collaboration products at Slack and later ran engineering at The Browser Company. The two first worked together at Foursquare in 2011 and have been building together on and off since. Scuba was previously operating under the name Flashback.

The financing is actually two rounds disclosed together: a pre-seed led by Accel and a seed co-led by Accel and IVP. The angel list reads like a reunion of the collaboration-software era — Slack co-founders Stewart Butterfield and Cal Henderson, Stripe CEO Patrick Collison, Instagram co-founder Mike Krieger, LinkedIn co-founder Reid Hoffman and former CEO Jeff Weiner, Dropbox co-founder Drew Houston, and Salesforce co-founder Parker Harris, alongside Vanta CEO Christina Cacioppo, Harvey co-founder Winston Weinberg, and Abridge CEO Shiv Rao.

The product rests on two primitives. Captures let a user save articles, videos, web pages, screenshots, and call transcripts through a browser extension or iOS app. Collections organize those captures into shared, searchable sets that a team can query, summarize, and reuse — with the company's internal example being two private Collections holding everything related to product strategy and to how employees describe the company externally.

Where it gets interesting for the agent era is the plumbing around them. Scuba connects to ChatGPT, Claude, and Slack through Model Context Protocol connectors, so a team can pull context out of a Collection into an AI tool, or push AI-generated assessment documents and QA test results back into it. The company is explicit that it is not trying to be another note-taking app or an assistant; the positioning is the context supply layer that AI workflows read from and write to.

"Shared context is usually much messier than you'd think," Falola said. "It's screenshots, PDFs, web pages and ad-hoc conversations. So Scuba gives you a way to make shared memory available on demand."

The funding structure is a statement in itself: rather than dribbling out a pre-seed and a seed, the company disclosed both at once and said the capital will let early users keep the full feature set free while it settles on pricing.

What it means: memory is becoming the axis of competition in agentic software — Meta just shipped persistent Active Memory in its newest models, and agent-memory benchmarks are turning into their own subfield. Scuba is betting the durable layer is not inside the model but around it: a shared, governable store of what a team knows, which humans and agents both read. Whether active saving becomes a habit, rather than another place information goes to die, is the open question.