On This Week in Startups, Jason Calacanis described OpenClaw Ultron, an internal AI agent that acts as "the one canonical employee of the organization."
Listeners heard details of an AI that autonomously researches guests, builds its own CRM, sends emails, and maintains a work log, all without being prompted.
If you are searching for OpenClaw Ultron this article will cover what it is, the technical breakthroughs that made it possible, and how you can get the same capabilities for your business right now.
What is OpenClaw Ultron?
OpenClaw Ultron is Jason Calacanis's custom deployment of OpenClaw, the open-source AI agent framework, configured for his team's internal workflows.
On his podcast Calacanis demoed OpenClaw instances he called "virtual employees" with full access to Gmail, Notion, Slack, and WhatsApp, handling podcast production tasks autonomously. He named the project OpenClaw Ultron and described it as the single AI worker that knows everything about his organization.
You cannot use OpenClaw Ultron, because it is an internal tool of Calacanis’s media company and venture firm. However, the capabilities Calacanis described are available in other products, and the technology behind them has a clear lineage.
Does AI actually work for work? A short history.
Most employees’ experience with AI at work starts with using ChatGPT for drafting emails.
That was a helpful, but a limited us of AI. What Calacanis demonstrated is generationally different. OpenClaw Ultron is an AI agent that autonomously operates across business tools, makes decisions, and completes multi-step workflows.
That did not become possible until a specific sequence of breakthroughs over the past two years.
AI models became agentic
Before mid-2023, large language models could only generate text. Then in June of 2023, OpenAI shipped function calling, and for the first time a model could take action in the real world.
Function or “tool calling” enabled the model to invoke an external API, query a database, send a message, or create a record. Later, Anthropic, Google, and open-source projects released their own agentic models. By late 2024, tool calling was generally reliable and production-grade.
This was the first unlock toward AI that could be an intelligent tool, not just a chatbot.
For a deeper breakdown of how model capabilities have progressed, see our "AI for Startup Leaders" guide.
Pi proved an agent could be autonomous
In November 2025, Mario Zechner released Pi, a coding agent with only four tools: read, write, edit, and bash.
Mario’s thesis was that AI models have become great at reasoning on their own. So, give them access to low level computer controls and get out of the way.
Pi is built to be minimal, but a key innovation was self-extension. Instead of downloading plugins, you ask the agent to build its own tools at runtime. This proved that AI can do sustained, autonomous work, not just answer questions when prompted.
OpenClaw turned Pi into a personal AI employee
Peter Steinberger built OpenClaw on top of Pi's engine, and it became the fastest-growing open-source project ever, with over 180,000 GitHub stars.
He added persistent memory, 13+ messaging channels, browser automation, cron scheduling, and a community skills marketplace.
Developers rushed to use this AI agent that doesn’t wait for a human prompt, can run while you sleep, proactively takes action, and learns from its own work.
We wrote a detailed breakdown of how OpenClaw works, and how it compares to enterprise alternatives here.
Jason Calacanis builds OpenClaw Ultron
Calacanis took OpenClaw and configured it for his companies. He demo’d the agent doing his team's shlep work, including sending outreach emails that recipients could not distinguish from a human. Calacanis predicted a future where every company has access to a tool like OpenClaw Ultron.
That future is closer than he imagined.
The gap between OpenClaw Ultron and your business
OpenClaw Ultron proves the concept. But Jason built it for a small team with a high risk tolerance. For most companies, running an open-source agent with unguarded access to production systems is not viable.
The core issues with the project includes no role-based access controls, no audit trail, no compliance posture.
CrowdStrike flagged hundreds of exposed OpenClaw instances leaking API keys and tokens. You own every integration, every auth token, and every failure. When an API changes at 2am, it is your problem.
We covered these security and architecture tradeoffs in depth in our OpenClaw breakdown.
You can get a business-ready OpenClaw Ultron today
Adapt is the business-ready version of what Jason Calacanis described.
It’s the same thesis - one intelligence layer that connects all your tools and autonomously runs workflows. However, Adapt is built to be used safely by any business team.
OpenClaw Ultron | What Adapt does today |
|---|---|
Pulls together Slack, Notion, email, institutional knowledge | Connects to Slack, Notion, Linear, HubSpot, GitHub, Google Workspace, Stripe, and more |
One canonical employee of the organization | AI computer for your business: questions, answers, and actions across every connected system |
Autonomous virtual employees running 24/7 | Scheduled tasks, proactive monitoring, and automated recurring workflows |
What this looks like in practice
Automated pre-call briefings. Every morning, Adapt can pull from your CRM, Slack threads, and past meeting notes and deliver a single briefing to the rep before each call.
Cross-system answers. Ask "What is our pipeline by stage?" and Adapt can pull from tools such as HubSpot, Stripe, and your database and deliver one answer with charts.
Recurring workflows. You can set recurring tasks in Adapt. Prompt, "Every Monday, pull open Linear issues, cross-reference with GitHub PRs, and post a status update to #engineering." Set this once, and it runs until you stop it.
Org-wide knowledge access. Engineering can ask about customer data within Salesforce without an additional license. Support can ask about the codebase without GitHub access.
Adapt also provides enterprise security (SOC 2 in progress), multi-user role-based access, managed connectors, and cloud hosting. You do not have to buy dedicated hardware, or maintain integrations.
Get early access to Adapt.
FAQ
What is OpenClaw Ultron?
OpenClaw Ultron is Jason Calacanis's internal AI agent built on OpenClaw, connecting his team's Slack, Notion, email, and institutional knowledge into one autonomous tool.
How can I use OpenClaw Ultron?
You can't. It is an internal tool for Calacanis’s team. OpenClaw itself is open-source and free, but requires technical setup and has no enterprise security controls.
What is the business alternative to OpenClaw Ultron?
Adapt provides the same autonomous agent capabilities with enterprise security, managed integrations, and multi-user support.
Is OpenClaw safe for business use?
OpenClaw has significant security risks for workplace use, including broad system access, no RBAC, and no audit logging. See our full breakdown.
