Evolving our calendar assistant Reclaim to be AI-native without starting over

Reclaim, the Dropbox-owned calendar assistant, uses AI to find and adjust time for tasks, habits, and meetings. That works when inputs and expected outcomes are clear, but people also rely on it for scheduling that isn't straightforward. As natural-language interfaces became more capable, a new possibility appeared: users describing a scheduling goal in their own words, even when it maps to no existing setting or command. Words alone can't determine what happens on a real calendar, though. A request only has meaning alongside what's already scheduled, when the user is available, and which preferences and commitments constrain any change.

Bridging the two demanded more than adding an AI interface to the product. It required rethinking how conversational requests become calendar changes while keeping the existing scheduling system intact. Calendar changes carry real weight — time is precious and events involve other people — so any AI capability had to extend Reclaim naturally and reuse its scheduling logic.

Three actors, one scheduling path

Before the redesign, Reclaim handled calendar changes in two ways. Users made specific changes themselves, such as moving an event or updating a setting. Separately, the automated scheduler ran in the background, weighing each person's preferences and availability to find time for tasks, focus work, and recurring routines like lunch, then adjusting those blocks as schedules shifted.

AI agents introduced a potential third actor: a user describes what they want, and the agent interprets the request, examines the relevant calendar state, and determines how Reclaim can help. But LLMs interpret the same request differently because they generate answers from patterns in training data. An open-ended request such as "make time for this tomorrow" could lead an agent to find an open block, create a new event, or move an existing one — options with different effects on the user's schedule, and sometimes on other people's calendars too.

That variability set agents apart from the two existing paths. Because all three actors reach decisions differently, they still had to rely on the same scheduling logic; a separate AI-only path would have duplicated functionality and made behavior harder to keep consistent across interfaces. The answer was an agent platform built within Reclaim's existing architecture.

Inside the agent platform

The platform supplies a model with relevant calendar details and access to Reclaim's features while controlling what it can see and do. The model interprets what the user says and works out intent. The agent manages the broader process: it provides Reclaim's instructions, the relevant calendar context, and tools that let it ask Reclaim to look up information or request a specific action. When a tool returns information, the agent can pass it back to the model to inform the next step — a repeated exchange called the agent loop.

Rather than exposing everything at once, Reclaim gives the agent only the calendar information and tools relevant to each request, which focuses it on the immediate task and limits it to appropriate actions. For more complex requests, the agent can delegate one specific part to a specialized subagent, while internal checklists and review steps keep the overall request on track.

The team built the agent loop, the tools, and the context-gathering system in-house, connecting directly to model providers instead of adopting a broad agent framework. A framework explored initially lagged behind the provider APIs they wanted and carried more structure than Reclaim required. Owning these components allows the design to follow Reclaim's scheduling needs, makes new provider capabilities faster to adopt, and supports different model providers without rebuilding the rest of the platform — at the cost of more code to maintain, some of which agents themselves help maintain.

The same tool system supports Model Context Protocol in two directions. Within Reclaim's agent loop, an MCP client lets the agent use compatible tools from other services. Separately, Reclaim's MCP server exposes selected Reclaim tools to supported AI clients, including Claude and ChatGPT. Those clients rely on their own models and agent loops but can call the same tools used inside Reclaim.

Schedule Actions and Preview Mode

Each scheduling capability is represented internally as a Schedule Action — a standard operation such as creating or updating an event, changing an RSVP, or finding availability. Whether the request comes from a user, the automated scheduler, or an agent, the same Schedule Action type applies, including the same validation and commit process.

The shared path matters because one calendar change can ripple across several events. Moving a meeting may alter someone's availability or prompt Reclaim to adjust flexible events elsewhere. Separate implementations per actor would force engineers to reproduce those rules and keep every version aligned as the product evolved. With one Schedule Action, an operation changes in a single place instead of three, which keeps behavior consistent across interfaces.

Consistency alone isn't enough: users need to see whether a change's wider effects match their intent. Preview Mode provides a temporary calendar where AI suggestions and chat-requested changes can be reviewed before they touch the real calendar — showing how an event or settings change would affect the rest of the schedule, so the effect on shared events can be checked before updates go to other attendees.

To keep the preview fast, the automated scheduler was reworked as a pure function, so it can calculate a proposed schedule without saving anything to the user's actual calendar. Reclaim repeats that calculation whenever the user adjusts the preview. Because busy calendars and multi-attendee meetings involve a lot of data, a readily accessible copy is kept in Redis, a store built for fast retrieval. Attendee availability is updated inside the preview whenever an event moves, so the next calculation reflects the proposed schedule.

What the rebuild preserves

Reclaim now has a consistent route from natural-language request to reviewed calendar change: the agent platform connects models with relevant calendar information and Reclaim features, Schedule Actions route proposed updates through the product's scheduling logic, and Preview Mode makes their effects visible before anything reaches the calendar or attendees.

The work also clarified what AI needs to be useful inside an active workflow — surrounding information to understand a request, a controlled way to act on it, and a review step before the result moves forward. In Reclaim, a scheduling goal travels from conversation to proposed schedule to approved update without the user translating it into a specific setting or command.

The lasting value is the freedom to evolve without fragmenting the product. As AI changes how people ask for help, new capabilities can arrive through the same scheduling foundation instead of a separate experience each time.