In the span of four weeks, an open-source project called MoltBot has accumulated more than 69,000 stars on GitHub. For context, that puts it in the same tier as enterprise-grade projects like Kubernetes within its first year.
MoltBot (rebranded from Clawd Bot following a trademark request from Anthropic) is a self-hosted assistant that connects large language models, typically Claude or GPT-4o, to a user’s local machine. Unlike Microsoft Copilot or Zoom AI Companion, which operate within defined application boundaries, MoltBot sits at the operating system level.
It reads the screen, controls the mouse and keyboard. It executes terminal commands. And, crucially for the collaboration space, it unifies messaging platforms—Slack, Teams, Telegram, WhatsApp—treating them as interchangeable channels for a single intelligence.
The result is an assistant that not only suggests actions but performs them. Users message their MoltBot instance from a phone, and it opens applications on their desktop, navigates interfaces, retrieves files, and sends emails. No API integration required.
But if that level of access sounds like a security team’s nightmare—it should. Are organisations ready for employees deploying autonomous agents with admin-level permissions outside IT oversight?
Why Enterprise UC Tools Haven’t Solved This Problem
The appeal of MoltBot lies in what it exposes: a gap between what enterprise collaboration platforms promise and what users actually need.
Most UC vendors have spent the past 18 months embedding AI into their products. These implementations are powerful within their ecosystems—Microsoft’s Copilot excels at summarizing Teams meetings and drafting Outlook replies; Zoom’s Companion handles transcription and follow-ups effectively.
But workflows don’t stay inside ecosystems. A typical knowledge worker might receive a request via Slack, pull data from a legacy CRM that only has a web interface, update a Jira ticket, and confirm via WhatsApp. The official tools can’t bridge that gap without expensive middleware or custom development.
MoltBot’s approach is blunt but effective: it automates at the GUI level, the same way a human would. It doesn’t wait for vendors to build integrations. It just opens the browser and clicks.
The Return of Shadow IT—With Execution Permissions
For IT leaders, this development is uncomfortably familiar.
The pattern mirrors the “Bring Your Own Device” wave of the early 2010s, when employees bypassed clunky corporate systems by using consumer tools like Dropbox and Slack. The difference now is scale and scope.
Shadow IT was about file sharing and messaging. “Shadow Agents” have admin-level access to machines, persistent visibility into workflows, and the ability to execute code. Employees are granting these permissions because the productivity gains are immediate and significant.
The governance challenge is acute. Unlike a cloud service that can be blocked at the firewall, MoltBot runs locally or on private servers. It operates within the user’s session, making it difficult to detect or control through traditional endpoint management tools.
Security researchers have already flagged the risks. A compromised agent doesn’t just leak data—it can navigate systems, launch applications, and mimic legitimate user behavior to exfiltrate information or escalate privileges.
Yet adoption continues, driven by a simple calculation: the official tools are too slow, too restricted, or too siloed to match what a local agent can deliver.
What This Means for Collaboration Platforms
MoltBot’s rapid uptake is a market signal that the current approach to enterprise AI is leaving a significant portion of the user base unsatisfied.
The issue isn’t capability—vendors have access to better models, more data, and deeper integrations than any open-source project. The issue is scope. Enterprise AI is being designed for safety and governance first, utility second. MoltBot reverses that priority.
For UC vendors, the challenge is clear: users are demonstrating demand for assistants that operate across platforms, not just within them. They want persistent context, cross-application workflows, and execution capabilities that don’t require opening a ticket with IT.
If the industry doesn’t provide a governed path to that functionality, the workforce will continue building their own—security policies notwithstanding.
