Scuba Raises $34M to Solve AI Agents' Context Gap
Scuba, the shared context engine for teams and their AI agents, emerged from stealth on October 13, 2026, with $34 million in total funding led by Accel and IVP. Founded by Noah Desai Weiss, former Chief Product Officer at Slack, and Dolapo Falola, former Head of Engineering at The Browser Company, Scuba targets the most common bottleneck in enterprise AI deployments today: agents that can execute tasks quickly but have no institutional memory of how your team thinks, what your company has already decided, or why you made the choices you made.
The Context Gap Holding Enterprise AI Back
Every serious enterprise AI deployment runs into the same wall eventually. Your agents are fast, your models are capable, and your workflows are automated. But the answers your agents produce feel generic. The decisions they make miss institutional nuance. The code they write misses your team’s architectural principles. The emails they draft sound like they were written by someone who has never spoken to your customers.
That gap has a name: context. Specifically, the shared, proprietary, continuously evolving institutional knowledge that lives inside teams rather than on the public internet. Your ICP definition built from 200 discovery calls. Your competitive positioning refined through 50 loss reviews. Your architectural decisions documented in a Slack thread from nine months ago. Your go-to-market calendar built around your biggest customer’s budget cycle. None of that is in any frontier model’s training data, and none of it gets automatically loaded into every agent you run.
Teams have tried to patch this with elaborate prompt templates, agent handbooks, and Notion dumps. These approaches break at scale. They require constant manual maintenance, surface stale information, and treat context as a one-time input rather than a continuously evolving team asset.
Scuba’s thesis: this is not a prompting problem. It is a memory layer problem. And they are building that layer.
How Scuba Works: Captures, Collections, and MCP
The product centers on two core concepts.
A Capture is any artifact a team member decides is worth preserving: a 96-page analyst report, a screenshot of a competitor’s pricing page, a voice memo from a board meeting, a web page, an iMessage thread, or a sprint retrospective write-up. Users save Captures with a single keyboard shortcut (Cmd+Shift+S on Mac). Scuba extracts the full content, summarizes it, and connects it to everything else the team has saved. No manual tagging, no categorization.
A Collection is the team-level knowledge base that Captures flow into. Collections are self-maintaining: as the team adds Captures, Scuba updates the relevant Collections to reflect current decisions and direction. Collections can be shared across the entire team, and specific Collections can be locked so only designated owners can edit the canonical version.
The AI integration works through an MCP (Model Context Protocol) server. Teams connect Scuba to Claude, ChatGPT, Cursor, and Slack, so agents can read and create Captures directly. When a coding agent opens a pull request in Cursor, it can pull in the team’s architectural review principles without anyone pasting them into the prompt. When a sales agent drafts a follow-up, it can reference the team’s actual ICP definition and competitive positioning, not a generic version.
In Slack, @Scuba answers questions from teammates and automatically captures Slack threads into the relevant Collections without anyone leaving the channel.
The platform is available on Mac, iPhone, and Windows, with browser extensions. It launched in limited access with a waitlist on October 13.
Why the Founding Team Matters
The credibility here is unusually high for a seed-stage company. Noah Desai Weiss spent his career solving exactly this problem: getting the right information to the right person at the right moment. He did it at Google, then at Foursquare (where he led product and first met Falola), then at Slack (where he served as Chief Product Officer). Dolapo Falola was an engineering leader at Slack during the same period before becoming Head of Engineering at The Browser Company.
This is the fourth time Weiss and Falola have built together. Nine of Scuba’s ten-person team have shipped products with them before. The company, officially registered as Flashback AI Inc., completed a six-month private beta with hundreds of teams running product launches, investment research, and GTM campaigns on the platform.
The investor list is exceptional. Accel’s Andrew Braccia, who led the firm’s investment in Slack and led both of Scuba’s funding rounds, frames the company as the next mandatory layer in enterprise software: “Slack did it for team communication with channels. Cursor did it for writing code with AI. Now Scuba is doing it for team alignment with shared context.”
IVP General Partner Shravan Narayen co-led the seed and offered the sharpest framing of why this matters for agents specifically: “Without understanding a team’s shared judgment, agents can never achieve their full potential.”
Angels include Slack co-founders Stewart Butterfield and Cal Henderson, Stripe CEO Patrick Collison, Instagram co-founder Mike Krieger, LinkedIn co-founder Reid Hoffman, Dropbox co-founder Drew Houston, and Salesforce co-founder Parker Harris, among others.
The Context Layer Is the New Enterprise AI Infrastructure Category
The broader wave Scuba is riding goes beyond one startup. Enterprise AI teams are discovering that frontier model access is no longer the competitive differentiator. What differentiates teams is what context those models operate with.
This is reshaping how enterprises think about their AI infrastructure stack. As the table below shows, the categories that were once differentiated (compute, models) are now near-commoditized. The emerging differentiation layer is context.
| Stack Layer | Examples | Current Status |
|---|---|---|
| Compute | AWS, Azure, GCP, CoreWeave | Commodity |
| Models | GPT-6, Claude, Gemini 4, Meta Muse | Near-commodity |
| Application | Salesforce, ServiceNow, Workday | Maturing |
| Orchestration | LangChain, AWS Strands, Temporal | Standardizing |
| Context | Scuba, Pinecone, custom RAG | Emerging category |
Scuba’s approach differs meaningfully from traditional RAG (retrieval-augmented generation) implementations. RAG builds indexes over existing documents. Scuba captures the context that never makes it into documents in the first place: the judgment calls, gut checks, lessons learned, competitive reads, and interpersonal history that defines how a team actually operates.
As Weiss explained to SiliconAngle: teams “have their own point of view, their own taste, their own hard-fought lessons learned. And currently, almost none of that is being leveraged when they’re using AI.”
Implications for GTM and Operations Teams
For AI-native enterprise leaders and GTM teams, Scuba directly addresses a pain point visible in almost every AI deployment: teams invest in capable models and well-designed agents, then spend disproportionate time doing “context injection,” manually writing prompt preambles that explain ICP, competitive landscape, brand voice, and historical decisions every time they run an agent workflow.
Manual context injection is fragile. It is inconsistent across team members. It surfaces outdated information. And it scales poorly as teams grow and agent workflows multiply.
A maintained context layer turns context injection from a manual task into ambient infrastructure. Your outbound agent already knows your ICP without being told. Your competitive analysis agent already knows your current positioning. Your quarterly planning agent already knows the strategic decisions from the last board review.
This is also why the MCP integration is significant beyond product convenience. MCP has become the standard protocol for giving AI agents structured access to external context. Scuba’s MCP support means it drops into existing agent workflows without requiring teams to rebuild their tooling. The context layer plugs in rather than replaces.
For teams operating multi-agent systems running coordinated agents across outbound, research, content, and customer success workflows, the context layer is arguably more important than the model selection. Agents running on shared, current, proprietary context will consistently outperform agents running on generic prompts, regardless of which frontier model powers them.
The same logic applies to AI content systems at scale. Content agents operating with Scuba-level context about brand voice, audience segments, and competitive differentiation produce qualitatively different output than agents working from a static system prompt.
What Comes Next
Pricing has not been announced. Weiss has said the additional funding is less about burn rate (the team has “many years of runway”) and more about letting early users explore the full feature set before determining what pricing structure makes sense.
For enterprises that want to move ahead of the curve on context infrastructure before it becomes a standard line item in every AI stack, Scuba is worth tracking early. The founding team has built at scale before. The investor syndicate is among the strongest in enterprise software. And the problem they are solving is one that every serious AI deployment encounters.
If the context layer thesis is right, the teams that build institutional memory infrastructure now will have a compounding advantage over those that wait. Every Capture saved, every Collection maintained, every decision logged is context that future agents can build on. Teams that start early will have a richer, more accurate, and more current context layer than teams that start later.
That compounding advantage is exactly what Enera helps enterprise teams architect: from agent infrastructure and GTM automation to the context and knowledge systems that make agents actually useful at scale. Book a call with the Enera team to talk through how a context layer strategy fits into your AI transformation roadmap.