Composite secures NFDG backing for cross-browser agent

Composite has raised $5.6 million in seed funding to scale a cross-browser AI agent designed for work, securing lead backing from NFDG — the venture firm co-founded by Nat Friedman of GitHub and Microsoft fame and Daniel Gross of Apple and Y Combinator.

Menlo Ventures and Anthropic’s Anthology Fund also joined, indicating increasing investor conviction behind browser-native agents that operate where users already are.

Instead of asking people to use a special “AI browser,” Composite follows automation where it goes by extending support across the browsers that teams are already using, converting authenticated tabs and popular SaaS tools into a surface area for agent-driven interaction (clicking, typing, navigating) in the cloud.

The company’s initial concentration is dead-on work use cases — structured, repeatable tasks within products like Jira, applicant tracking systems, analytics dashboards, and ticketing tools — where tiny, reliable actions compound into real-time savings.

Why cross-browser agents matter for enterprise work

Enterprise users are not living in a single browser. For many companies, the default will be to standardize on Chrome for internal apps, retain Edge for Microsoft 365, and Safari on mobile — a mixed environment where AI browser lock-in becomes a nonstarter. StatCounter puts Chrome at around two-thirds global share on desktop, with Edge and Safari making up most of the rest — evidence that tooling has to straddle multiple engines in order to make sense.

Composite’s bet is that this proper abstraction is, in fact, the browser session itself. Because the agent is working inside browsers where people are already logged in, it can carry out actions without getting tripped up by a mess of third-party connectors. That loose coupling avoids integration friction, bypasses brittle API dependencies, and maps well to enterprise identity protections such as single sign-on and conditional access.

How Composite works across browsers to automate tasks

Composite works with any macOS or Windows browser, using the same lightweight extensions to allow its agent to interact with your favorite web apps. Users issue commands in natural language, and the agent performs step-by-step actions — clicking on UI elements, entering data into forms, scraping information, and posting back results — within the user’s authenticated context.

Sample workflows include first responders triaging a Jira backlog with the pertinent documentation handy while commenting on high-priority issues and closing duplicates; recruiters combing through various candidate platforms while composing specialized outreach; security teams transitioning alerts into prepopulated vulnerability tickets; or marketers consolidating ad metrics with web analytics and CRM data to report succinct status notes.

The product is already recommending tasks to perform based on common patterns, and Composite plans to make more “ready-to-run” automation available in this way and also apply scheduling as a part of recurring jobs. The system focuses on trustworthy atomic behaviors, in the form of accurate mouse clicks and keystrokes, rather than sprawling, conjectural autonomy — lessons learned from robotic process automation repurposed for the web app and LLM environment.

Funding and backers behind Composite’s seed round

NFDG’s participation brings operator experience in high-signal domains: Friedman led GitHub through its developer-focused renaissance, and Gross has early wins in AI and infrastructure investments. Menlo Ventures offers enterprise go-to-market knowledge, and the Anthology Fund from Anthropic indicates how critical it is for responsible, controllable model behavior in agents working within sensitive work contexts.

Composite was started by its two founders, Yang Fan Yun, who used to be a product manager at Uber, and Charlie Deane, who had previously built a business centered on proxy servers. The thesis came from seeing knowledge workers spending significant amounts of the day on repetitive browser tasks that are structured enough that an agent can run safely with guardrails.

Enterprise readiness and governance for safe agents

For organizations, the selling points are practical: users retain their existing browsers; admins can limit tools and domains; tasks run locally to reduce data egress; and teams can whitelist or blacklist sites directly.

And yet those controls are the governance patterns CISOs want — least-privilege access, clear traceability of actions, and the way to ring-fence sensitive assets.

If Composite continues to evolve along enterprise lines — adding audit logs, granular role-based control, and alignment with compliance frameworks already frequently called out — it could serve buyers where they are: hungry for demonstrable productivity gains without having to add new data paths into ones they already have to vet.

A crowded but early landscape for browser-based agents

The agent space is becoming crowded. New AI browsers try to reimagine the interface; productivity suites get in your face with platform-centric agents that have highly enriched internal context. Others like Highlight play with the whole desktop as a context, and niche tools focus on specific artifacts (examples are spreadsheets). Composite’s cross-browser philosophy is different: it takes the web as it is and then makes sure that it works across a wide range of apps.

The problem for every player is to do that in a way that proves durable efficiency gains. Buyers will want to see shrinkage in time-to-resolution, diminished manual switching overhead, and consistent task success rates — not just cool demos. Get past the froth. Analysts have observed a transition from “prompting” to “doing” in enterprise AI; the winners will quantify that shift with defensible metrics.

What to watch next as Composite scales its agent

Immediate goals include wider support among browsers and tasks, stronger task discovery, and dependable scheduling. For the longer term, integrations with identity providers and audit exports, along with team-level administration, may determine whether Composite becomes a department-wide standard or more of a power-user tool.

Founded by investors who understand developer and enterprise adoption cycles, Composite is laying the groundwork to test an attractive hypothesis: the fastest road to agentic work isn’t a new browser; it’s smarter automation inside the ones professionals trust today.