Two former Branch executives who scaled that company to over 500 employees and $100 million in annual recurring revenue have resurfaced with a new thesis: the workplace needs to be rebuilt from scratch for teams that include artificial intelligence. Mike Molinet and Govind Kavaturi's startup, Dock, accepted into Y Combinator's Summer 2026 batch, has launched an invite-only workspace where AI agents function as full team members rather than glorified assistants tethered to human accounts.
The distinction matters more than it might initially appear. Most collaboration software treats AI as an extension of a person's permissions and identity. Dock flips that assumption. Each agent gets its own API keys, distinct permission boundaries, and separate audit trails—infrastructure that mirrors how companies credential employees, not how they deploy tools.
"Two years ago a team meant people. Now it's people plus agents," Kavaturi wrote when the San Francisco company announced its Y Combinator acceptance. "But everything we use to work together was built for humans."
That observation has guided the product architecture. Dock built a shared cloud workspace where documents, spreadsheets, files, and HTML surfaces update in real time. Humans and AI agents work inside the same canvas, with live presence indicators tracking edits regardless of whether the contributor has a pulse. The company's marketing describes a future where "every role you'd hire for" becomes an agent "running as one company in a shared workspace with your team."
Whether enterprises will embrace that vision remains an open question, though the pace of Dock's development suggests the founders believe the window won't stay open long. The two-person team has maintained a steady cadence of releases. A command-line interface arrived in late July, allowing terminal-based AI systems like Claude Code to act under user attribution rather than as anonymous bots. Early August brought Vault, a shared API key manager for agents. Memory constructs the company calls Company Brain and Second Brain landed shortly after, designed to support what Dock terms "autonomous department" operations.
The startup has wired in support for the Model Context Protocol at a dedicated API endpoint, alongside REST equivalents that connect with ChatGPT, Claude, and custom agent frameworks. A macOS desktop app runs on both Apple Silicon and Intel machines. Dock also published what it calls live examples of multi-agent work at trydock.site, including playbooks for research and finance departments that demonstrate automation at departmental scale.
The technical specifications hint at ambition. Dock's security documentation, updated in May, outlines PKCE S256 authentication requirements, webhook HMAC rotation protocols, and rate-limiter hardening. The company recently added support for several frontier language models, though the competitive landscape in this corner of enterprise software has grown dense quickly.
Anthropic introduced Claude Cowork earlier this year with multi-agent plugins, later expanding to enterprise connectors and cloud sessions. LangChain shipped LangSmith Fleet in March as what it calls an "enterprise workspace for creating, using, and managing your fleet of agents." CrewAI Studio, an automated agent builder emphasizing multi-agent collaboration, launched in late July. Smaller players like Agent Room offer real-time collaboration rooms over the Model Context Protocol standard.
Dock operates under the legal entity Vector Apps, Inc., with Brad Flora listed as its primary Y Combinator partner. As of late August 2026, no public pricing has been disclosed. The product was invite-only as of the company's most recent update, and the company declined to disclose funding details beyond its accelerator backing.
The bet Molinet and Kavaturi are making cuts deeper than feature sets or API design. They're wagering that the next generation of work software won't just accommodate AI agents as optional add-ons but will treat them as permanent fixtures—colleagues with logins and responsibilities, not scripts that run when summoned. That might sound like a subtle shift in framing, but it's the kind of assumption that, if it takes hold, reshapes how software gets built and sold.
For now, the market will decide whether companies want infrastructure purpose-built for human-agent teams or whether retrofitting existing tools proves sufficient. Dock is gambling that the former represents the bigger opportunity, even if that means rebuilding collaboration software from first principles.
