Scuba Emerges With $34M to Build Context Engine for AI Agents

Scuba, a New York startup building a shared context engine for teams and the AI agents that work alongside them, has emerged from stealth with $34 million in funding across two rounds. The pre-seed round was led by Accel, while the seed round was co-led by Accel and IVP. BoxGroup, Conviction, Homebrew, and Scribble Ventures also participated, alongside a long list of angel investors drawn from the founders' networks.

The company, formally Flashback AI Inc., was founded in 2025 by Noah Desai Weiss and Dolapo Falola, two product leaders who previously worked together at Slack. According to the company's Business Wire announcement, Scuba describes its product as a context engine: a central place where teams store the decisions, debates, documents, and customer insights that shape how they work, so that both people and AI agents draw on the same shared memory.

Context, not intelligence, is the bottleneck

The core argument behind Scuba is that today's AI models are not limited by their reasoning ability so much as by the information in front of them. As reported by SiliconANGLE, Desai Weiss framed the problem as models being context constrained rather than intelligence bound: give the same model better, team-specific context and the quality of its output rises with it.

That resonates with a pattern agent teams keep running into. Individual contributors move faster with AI assistants, yet few teams report feeling smarter together. Speed doesn't add up when everyone races in different directions: lessons get orphaned, alignment decays, and the monthly all-hands is too slow to catch it. According to Accel's own announcement of the funding, the decisions, debates, and customer insights that matter most are scattered across docs, threads, and people's heads, so everyone ends up prompting alone from scratch without the weight of the organization's earned knowledge.

How the platform works

Scuba's first product component is a capture layer. Through a browser extension and an iPhone app, users save information relevant to their work: news articles, chatbot responses, internal correspondence, screenshots, and videos. A built-in notetaking tool can transcribe audio notes up to 90 minutes long, according to SiliconANGLE's report on the launch.

Captured material is organized into thematic repositories the company calls Collections. A marketing team might keep a Collection aggregating everything about an advertising campaign, and the company says these repositories require no manual editing. Scuba also generates a daily Briefing that summarizes what changed inside a Collection, surfacing patterns such as an unusual spike in bug reports.

For the agent side of the equation, Scuba exposes captured context through an MCP server, so tools including Claude, ChatGPT, Cursor, and Slack can pull the team's shared knowledge into the model's working context. Per the Let's Data Science summary of the launch, that integration is what makes the service a context engine rather than a note-taking app: agents retrieve grounded, team-specific context instead of averaging answers from the public internet.

What the $34M signals for agent tooling

Scuba's emergence fits into a broader funding pattern this month: capital is flowing into the infrastructure layer that makes autonomous agents usable inside enterprises. According to Upstarts' exclusive report on the launch, the founding team's pedigree explains the investor enthusiasm. Desai Weiss served as chief product officer at Slack, while Falola led engineering at The Browser Company after earlier stints at Slack and Google, and the duo first worked together at Foursquare in the early 2010s.

The angel roster reads like a who's who of product-led software: founders of Slack, Instagram, Stripe, LinkedIn, Salesforce, Dropbox, Datadog, Vanta, Harvey, Abridge, Duolingo, Granola, Warp, Zynga, Groupon, Coda, Opendoor, and Flatiron Health all backed the rounds, as listed in the company's announcement. That kind of operator-heavy syndicate tends to signal product-market intuition more than financial engineering.

The company says hundreds of early beta users already rely on Scuba to develop company and product strategy, track customer feedback, and cut down on meetings. The open question is the same one facing every agent-infrastructure bet: whether shared context becomes a durable platform layer or gets absorbed as a feature of the model providers themselves. Either way, the direction is clear. Teams whose agents understand their values, judgment, and history will keep moving smarter together, while teams whose agents only see the public internet will converge on the same average answers as everyone else.

Readers following this beat can see how the pieces connect: ServiceNow's AI Passport is tackling signed identity for enterprise agents, and research shows personal AI agents help networked teams work measurably faster. Scuba adds the third leg: the shared memory those identified, networked agents all need to draw on.