OpenWork

Roadmap & Position in Agent Infrastructure

Open-source desktop agent platform for governed team AI workflows

Company Overview

OpenWork is an open-source agent platform that runs AI workflows across desktop, cloud workers, skills, plugins, MCP servers, and user-chosen models. The buyer is engineering, IT, and operations teams; no named customers public.

What They're Building

The company's public product roadmap & what they're committed to building.

OpenWork Desktop

A local app for running agents over files, browser tasks, skills, plugins, MCP servers, and any supported LLM.

OpenWork Cloud

A team hub for orgs, shared skills, cloud workers, custom LLM providers, and RBAC.

Model Governance

Admin controls let teams allow or block models, providers, and app updates by organizational policy.

Enterprise Deployment

Self-hosting, audit trails, SSO, SAML, and MDM, the adult-table features for security buyers.

Plugin Marketplace

Claude-compatible plugins and Git-based provisioning, pointing the product toward reusable agent packages.

Latest Intelligence

Zeitgeist tracks private signals to determine where the company is heading strategically.

New Intel: OpenWork is wrapping local agents, MCPs, plugins, and BYO models in a team admin layer. That pulls it into Claude Cowork territory while giving security buyers a cheaper path around model lock-in.

Founder and Key Execs

Ben Shafii

Founder (YC P26 founder, built AI automation products used by tens of thousands and sold a prior company)

Omar McAdam

Security contact (open-source and infrastructure background from public profiles)

Founder Force Multiplier

Ben Shafii has built AI automation products used by tens of thousands of people and sold a prior company, which is real shipping experience for a category prone to demoware. Omar McAdam brings open-source and infrastructure depth, which fits a product where local execution and security posture matter to the buyer.

Funding History

2026 | Founded
2026 | Accepted into Y Combinator Spring 2026
2026 | Public funding amount not disclosed

Competitors

Claude Cowork:

Anthropic’s native coworker product has model distribution and polish; OpenWork counters with open source, local execution, and BYO providers.

OpenAI Codex:

OpenAI’s coding agent ties tightly to its own model stack; OpenWork is broader agent infrastructure for teams that want provider choice.

Cursor:

Cursor owns the AI IDE workflow; OpenWork aims at desktop and team agent workflows beyond code editing.

OpenWork

's Moat:

Defensibility is still ahead of the company. The likely path runs through workflow switching costs from shared skills, MCP configurations, approved model routing, and team admin policy accumulating inside customer deployments.

How They're Leveraging AI

Model Routing

OpenWork routes agent tasks across local, customer-owned, and third-party LLMs based on team policy, cost, and model fit.

RAG

OpenWork likely uses local file and workspace context as a retrieval layer so agents can act on private company material without sending it to a central vendor cloud.

Agentic Workflow Automation

OpenWork turns reusable skills, plugins, MCP servers, and workspace context into governed workflow agents for teams.

AI Use Overview:

OpenWork's edge is orchestration, not model training. It routes agent sessions across local or third-party LLMs, MCP tools, plugins, and team-governed skills, treating the model layer as configurable rather than proprietary.

More Similar Companies

Tasklet

Runs cloud AI agents for recurring knowledge-work workflows.

Firebase-grade infra founders are attacking workflow automation as LLMs start to operate software, with early ARR and USV as both customer and investor.

Decagon

AI agents that autonomously handle enterprise customer support across chat, email, voice, and SMS.

Agentic AI for customer support is one of the few GenAI categories with clean ROI math, and Decagon has locked in the blue-chip logo sweep that compounds into category default status.

Manus

Autonomous AI agent platform that executes multi-step business workflows end-to-end

A model-agnostic orchestration layer with real browser and shell execution captured viral enterprise demand faster than any US agent startup, a thesis validated by Meta's $2B acquisition.

Bubble Lab

Gives ops teams an AI super-employee in Slack that automates cross-platform workflows.

Zapier and Make require users to think in workflows. Bubble Lab's Pearl lives in Slack and runs multi-step operations (Jira, Stripe, Notion) from plain English, which means ops teams adopt it without learning a new tool.

Back To All Companies >