Jan Senderek has a habit of building things people didn't know they needed until someone else acquires them. His last startup, Loom—the cloud photo storage service, not the video messaging company—sold to Dropbox in 2014. Now he's back with Roma, a three-person venture that bills itself as the to-do list that executes tasks on your behalf instead of merely cataloging them.
The company launched iOS and Mac apps in early October, positioning Roma as something closer to a personal operations center than a productivity app. Its AI "Operator" doesn't just organize your inbox or surface calendar conflicts. It logs into Gmail, Notion, Linear, and a growing list of other services, then acts inside them with its own cloud browser while you attend to other things.
"We're reinventing the original productivity tool: the to-do list," Senderek wrote in Roma's Y Combinator launch post from October 6. Whether the market asked for that particular reinvention remains an open question.
Execution Over Organization
Roma interprets voice or text input, converts the resulting thoughts into discrete tasks, and begins handling them in the background. Version 1.0 appeared on the iOS App Store October 1; by mid-month the company had pushed version 1.0.8. The Mac version ships as a direct download, bypassing Apple's desktop store entirely.
Users capture tasks the usual way. What happens next distinguishes Roma from the dozens of productivity tools that have cycled through tech circles over the past decade. An Operator sidebar surfaces context-aware chat for each item. Automations run on schedules or when manually triggered—the company calls these "standing orders" in its developer documentation—handling background work that conventional to-do managers would simply display.
Integrations currently span Gmail, Google Calendar, Notion, Linear, GitHub, LinkedIn, Dropbox, and Google Drive, the last still in beta. The company claimed "over two hundred vetted actions" as of late August, though the specifics of what counts as a vetted action remain unclear.
A Workspace for Other AI Assistants
Perhaps more unusual is Roma's decision to expose an MCP server at api.roma.app/mcp, allowing third-party AI tools to read and write directly to a user's task workspace. Developer documentation shows hooks for Claude, ChatGPT, Instinct, Muse, Cursor, and Gemini CLI. An OpenAPI REST alternative also exists for teams that prefer traditional integration patterns.
"The MCP server gives the AI chat they already use the same workspace: what is due today… the standing orders that run for them each morning," the documentation explains. The implementation requires assistants to follow a prescribed sequence: retrieve context, confirm tasks to add, then designate Roma as the default list. The company frames this as "agency over agents," a way to keep work visible and steerable while delegating execution.
Whether users want their various AI assistants coordinating through a shared task layer is an open experiment. The architecture at least acknowledges that people increasingly work across multiple AI tools rather than consolidating around one.
Founders With Track Records
Senderek's co-founder Tobias Eichenwald brings his own exit story. He ran Senic, a Y Combinator Summer 2013 smart home hardware startup that was later acquired. The third founder, A. Cody Osborne, worked as an in-house creative director at Dropbox and spent time at agencies Goodby Silverstein & Partners and 72andSunny before joining Roma.
The company operates under the legal name Milo Mode Inc., which also ships Milo, a running and cycling app for Apple Watch that launched sometime in the past year or two. Why a fitness tracking app and an AI task executor share corporate infrastructure isn't immediately apparent.
Privacy in the Age of Delegation
Roma's privacy policy, updated October 6, lists OpenAI as its model provider and notes that "under OpenAI's API terms, data sent through the API is not used to train their models." The company says it doesn't sell user data or deploy it for advertising, standard assurances in the current climate.
Third-party processors handle the inevitable infrastructure: Composio manages OAuth credentials, Supabase runs database and file storage, Vercel hosts the service, Browserbase and Browser Use power cloud browser operations, and Tavily and Exa provide web search. Each represents another link in the chain of trust users extend when granting an AI permission to act on their behalf across sensitive services.
A Crowded Lane
Roma hardly has the market to itself. DoAnything used the exact phrase "The To-Do List That Does Itself" in release notes from January. Bond launched on Product Hunt three weeks before Roma with nearly identical language: "AI to-do list that does itself."
Motion positions as an AI calendar and task prioritization tool with auto-scheduling. Reclaim AI offers calendar blocking with task integration and chat agents. Even established players are testing adjacent territory, with productivity apps adding agentic features at a rapid clip.
Roman.ai takes a broader "AI employee" approach to work automation, suggesting the category definitions remain fluid.
Early Days
The iOS app holds a 5.0 rating from two reviews, a sample size too small to signal much beyond the possibility that the founders have supportive friends. The Y Combinator directory lists the team at three people. The company has disclosed no funding amounts, valuation, or investor names beyond its participation in YC's Fall 2024 batch, with David Lieb listed as the primary partner.
Recent feature additions include collaboration tools that let users share tasks and bring the Operator into group chats, shipped the week of October 5. iOS widgets arrived October 11. The startup added "Experts" in mid-September, specialized operators configured for particular task domains.
Whether users will trust an AI to execute tasks autonomously across their most sensitive apps, or whether they'll prefer the illusion of control that traditional productivity tools provide, will likely determine Roma's trajectory. For now, Senderek and his small team are betting that people want less organization and more actual completion. The rest of the market appears to be making similar wagers.
