Catch Emerges From Stealth With $5M Seed to Build an Autonomous AI Executive Assistant

Catch has emerged from stealth with $5 million in seed funding to expand an agentic AI assistant designed to take over the administrative work that consumes a significant portion of an executive’s day.

The round was co-led by Entrée Capital and Pitango, with participation from Seedcamp and Factorial Capital. Rather than building another AI tool that summarizes emails or recommends what a user should do next, Catch is attempting something more difficult: allowing executives to delegate entire administrative tasks to software and have them completed.

Founded by Nir Sabato and Yoav Ramon, Catch focuses specifically on executive administration, including scheduling, email management, travel, follow-ups, calendar conflicts, reminders and outbound phone calls. The company says its system has already scheduled more than 12,000 meetings, handled over 200,000 emails and completed more than 20,000 delegated tasks.

Moving AI from assistance to delegation

The distinction between assisting with a task and actually completing it has become one of the central challenges facing agentic AI.

Large language models can already summarize an inbox, draft an email, suggest flights or determine when two people are available. But turning those individual capabilities into a system that can reliably coordinate with another person, update a calendar, make a reservation and follow up when circumstances change requires a much deeper execution layer.

Catch is built around that gap.

Rather than asking users to configure workflows or construct their own agents, the system connects to the tools executives already use. Catch’s website shows support across services including Gmail, Outlook, Google Calendar, Slack, WhatsApp, iMessage and phone, allowing users to communicate with the assistant much as they would communicate with a human executive assistant.

Once connected, the system develops an understanding of preferences such as meeting habits, calendar priorities and travel requirements. Catch says it can then act proactively. A canceled meeting, for example, can trigger rescheduling, while an upcoming trip can lead the assistant to identify missing hotel arrangements rather than waiting for an explicit prompt.

This represents an important shift in how AI products are being designed. Instead of creating another interface that executives have to manage, Catch is attempting to disappear into the communication channels and software they already use.

Autonomy does not mean acting without limits

The harder problem with administrative agents is not whether an AI model can understand a request. It is determining when software should act autonomously and when a human should remain involved.

Catch appears to be taking a tiered approach.

For routine administrative work, the goal is to minimize unnecessary approvals. But actions with financial or other meaningful consequences can require confirmation. In the company’s travel workflow, for example, Catch can interpret a travel brief, compare flights and hotels against constraints, and assemble an itinerary, but the user provides explicit approval before a booking is completed.

Similarly, Catch’s inbox product can triage messages and draft replies based on a user’s previous writing style, while its current workflow keeps the user involved before certain messages are sent.

That boundary may prove crucial for agentic AI adoption. Full autonomy sounds compelling, but in practice businesses are likely to adopt agents gradually, expanding their authority as systems demonstrate that they can make reliable decisions within clearly defined domains.

Administrative work could be a particularly useful proving ground because it combines repetitive tasks with enough contextual decision-making to test whether AI systems can move beyond simple automation.

Catch is building specialized AI underneath the assistant

Behind the conversational interface, Catch is also developing its own AI architecture rather than relying exclusively on a general-purpose foundation model.

The company describes a system of narrowly scoped “splinter agents”, with individual agents given access only to the information needed for a specific task. The idea is to reduce the amount of sensitive context exposed during execution while also improving latency and reliability.

Catch says anonymized interaction data can be used to improve its own models through supervised learning and reinforcement learning. At the same time, its security documentation says outside AI providers are contractually prohibited from retaining customer data or using it to train their own models.

Because an executive assistant may have access to calendars, emails, travel plans and potentially sensitive company information, security becomes considerably more consequential once an AI system can take actions rather than simply generate text. Catch says it is SOC 2 Type II compliant, alongside CASA Tier-2 and Google verification, and encrypts stored and transmitted data while using granular permissions for connected services.

Founders with backgrounds in AI and venture investing

The company was founded by two executives who approached the administrative problem from different directions.

CEO Nir Sabato previously spent several years investing at Entrée Capital, where he focused on areas including artificial intelligence and enterprise technology. Before entering venture capital, he held strategy roles at Fiverr and BLEND. Catch says Sabato’s experience working with founders helped expose how much high-value executive time was still being consumed by relatively routine coordination work.

CTO Yoav Ramon, meanwhile, has spent more than a decade working with machine learning systems. His background includes speech and language technology at Hi Auto and serving as CTO of healthcare AI company Nym, where teams deployed large language models in regulated healthcare environments.

That combination is relevant because building an administrative agent requires more than an effective chatbot. The system needs to understand preferences, maintain context across interactions, interact with external software and determine when it has enough confidence to act.

Agentic AI’s next battleground may be ordinary work

Catch is entering a rapidly expanding agentic AI market, but its larger significance may come from how narrowly it has defined the problem.

Many AI companies are attempting to create increasingly general agents capable of performing dozens or hundreds of types of work. Catch is taking the opposite approach: concentrating on a relatively constrained domain and trying to make the system reliable enough that users actually stop performing those tasks themselves.

That could become an important pattern for the next stage of enterprise AI.

The most commercially useful agents may not initially be autonomous digital employees capable of performing any task. They may instead emerge as highly specialized systems that understand one category of work deeply enough to be trusted with progressively greater authority.

Executive administration offers a revealing test. Scheduling meetings, managing travel and coordinating email may sound mundane compared with coding or scientific research, but these workflows require AI to interact with real people, understand incomplete instructions, remember individual preferences and make decisions that have real-world consequences.

If systems such as Catch can reliably handle those workflows, the significance extends beyond saving executives a few hours. It would demonstrate that AI agents are beginning to cross the boundary between generating useful information and assuming responsibility for getting work done.

The new $5 million seed round gives Catch additional capital to test whether that transition can scale.