A Return After 13 Years: From Machine Learning Clusters to AI Conversations

Thirteen years is a long time in technology. When I last hit "publish" here in 2013, Bitcoin was a novel curiosity and building a 24-core cluster into an IKEA Helmer cabinet was a fun weekend project. The world has changed, and so have my projects—but my fascination with making powerful technology accessible and practical has not.

Back then, my friend Alex Nugent and I were obsessed with raw computing power, as seen in our 2008 post about that 24-core cluster. Today, Alex and I are building something different: AI Receptionist, a practical tool born from the real frustration of modern business communication. (More on that shortly).

Some things do remain constant. I've always believed in building lasting tools. Case in point: our open-source Java library, XChange, is still actively maintained on GitHub after all these years. It's a testament to building with care and for the community.

The AI Tool Landscape: Power Demands Responsibility

The tech world today is buzzing with agentic AI—tools that don't just answer questions but take actions. One project, Moltbot (formerly Clawdbot), recently captured this hype, exploding on GitHub. Its promise is compelling: a personal, self-hosted AI assistant.

However, its rapid rise was marred by serious security oversights, from exposed API keys to critical vulnerabilities in its architecture. This pattern is unsettlingly familiar. It mirrors the core security dilemmas we've written about before, like the risks of building production systems on platforms such as n8n (which we detailed here).

These incidents underscore a critical rule: any tool granted high-level access to your digital life must be treated with corresponding seriousness. Whether it's a local LLM agent, an MCP server from an open-source project, or an automation tool from Zapier, you must be vigilant about the permissions you grant. Security isn't a feature; it's the foundation. The power to act is also the power to cause harm if that access is not meticulously controlled and isolated.

Many guides assume you're running it on a dedicated local machine. But you don't need a Mac Mini (a nod to our ancient 500GB MacBook upgrade guide for those who remember!). A well-configured VPS can work, provided you prioritize isolation, strict firewalls, and treat its access tokens with the gravity of root passwords.

Our New Chapter: AI Receptionist

So, what have Alex and I been building with all these security and reliability principles in mind?

AI-Receptionist.com solves a simple, painful problem: small businesses are drowning in robocalls and spam, missing real customers in the noise. We've combined decades of experience in AI, distributed systems, and yes, even low-level hardware (hello, Helmer cluster) to build an intelligent, 24/7 phone agent.

It's more than an answering machine. It understands context, filters spam intelligently, and ensures no genuine call is missed—all at a fraction of the cost of a human receptionist. Crucially, it's built as a secure, fault-tolerant service from the ground up. It's the applied, reliable implementation of the AI principles that much of the current hype is about, designed to operate in the real world where security and privacy are non-negotiable.

A Question for the Past

Writing this feels like sending a message in a bottle to a different internet era. To those who commented on those old posts about Linux clusters and Bitcoin—are any of you still out there? If you are, drop a note. I'm curious what you're building now and what you think of this wild AI landscape.

The goal of this blog was always "obscured clarity"—cutting through noise to the practical heart of technology. That mission continues. Let's see where the next conversation takes us.

– Tim