Arguing about AI with Billy Hollis

So what can numerics do for you? Carl and Richard talk to Jeffrey Sax about his work building Numerics.NET, a collection of general-purpose mathematical and statistical classes built for .NET. Jeffrey explains that numbers in a computer differ from real numbers in the world; for storage and computational efficiency, typical variable types in any programming language have limits on precision and size. How those variables are used can affect the outcome, with potentially disastrous consequences. Numerics.net provides exacting control of precision, as well as complex mathematical functions. It's far from free - but if you need it, you need it!

Ready for a rant? Carl and Richard talk to Billy Hollis about the impact of artificial intelligence on software development. The conversation starts with a listener comment about going all-in on AI to speed up development radically, which raises the question of how much AI is the right amount. What's safe and what is reckless? And how will that position change over time? Are the problems we're having today just growing pains of the tools, or are they systemic to the technology? Lots to debate!

How about a cloud-native AI assistant? That's RockBot! Carl and Richard talk with Rocky Lhotka about RockBot, Rocky's own variant of the OpenClaw/Hermes type personal assistant. Rocky talks about being uncomfortable with the security approaches the other agents have taken, so he went full enterprise architect on the problem and built RockBot as a set of containers behind the Kubernetes orchestrator. Cloud architecture without the requirement of the cloud! The conversation dives into how to give RockBot controlled access to services like calendar, email, etc. And it's all .NET and open source, so if you want to make changes or contribute, go for it!

Whether quantum computers are around the corner or not, the message is clear: the time for post-quantum cryptography is coming fast. Carl and Richard talk to Michael Howard about Microsoft's efforts to make post-quantum cryptography available to everyone by 2029. Howard discusses the quantum computing and cryptography issue, specifically the ability of quantum computers to run Shor's algorithm quickly enough to break RSA encryption. While it isn't possible just yet, the harvest-and-decrypt-later concerns are real. And the solution is pretty straightforward- move to TLS 1.3, and be prepared for a wave of changing encryption algorithms as they roll out and are found wanting. A crazy time is coming soon!

So what can numerics do for you? Carl and Richard talk to Jeffrey Sax about his work building Numerics.NET, a collection of general-purpose mathematical and statistical classes built for .NET. Jeffrey explains that numbers in a computer differ from real numbers in the world; for storage and computational efficiency, typical variable types in any programming language have limits on precision and size. How those variables are used can affect the outcome, with potentially disastrous consequences. Numerics.net provides exacting control of precision, as well as complex mathematical functions. It's far from free - but if you need it, you need it!

How can constraints make software development assistants more efficient? Carl and Richard talk to Don Demcsak about his work on getting LLMs to build higher-quality software with fewer resources. Don talks about how domain-driven language techniques help to define and constrain language around a given application problem space, but that largely hasn't been applied to coding agents so far. And as powerful as domain-driven is, it has challenges when it comes to building applications - and, more importantly, testing them. That's where behavior-driven approaches have advantages, which can also be applied to agents. Applying techniques of the past to tools of the future can accelerate development even further!

Ready for a rant? Carl and Richard talk to Billy Hollis about the impact of artificial intelligence on software development. The conversation starts with a listener comment about going all-in on AI to speed up development radically, which raises the question of how much AI is the right amount. What's safe and what is reckless? And how will that position change over time? Are the problems we're having today just growing pains of the tools, or are they systemic to the technology? Lots to debate!

What's the best way to interact with an LLM in applications? Carl and Richard talk to Chad Michel about the evolving user interface to large language models. Chad recalls Build 2023, where Technical Fellow Steven Batiche talked about LLMs starting out beside your application (like Copilot), but eventually moving inside, like Claude Cowork. Then Steven suggested that the ultimate destination is outside - a separate interface that then orchestrates across applications. The conversation dives into how development has evolved with these tools, and what other applications could look like in the future. 

What can OpenClaw do for you? Carl and Richard talk to David Guttman about his work on OpenClaw, and helping others be successful with it.OpenClaw was one of the first personal agents released and has continued to evolve with the industry. David talks about controlling access to OpenClaw, and controlling what OpenClaw has access to. The challenge is managing prompt injection, and the agent can be manipulated by others to share information inappropriately. But it can also be a time saver for routine and annoying tasks - it takes time to set up properly, but that time is paid back quickly when it gets to work!

What's Shawn been up to lately? Carl and Richard chat with Shawn Wildermuth about his work in the Netherlands. Shawn talks about his last show on being a senior developer - and how much that has changed in the past two years. The role of large language models in software development has changed careers, but they're still fun. The conversation also dives into the role of Aspire to utilize the best of cloud architecture and how LLMs make that easier as well - and enable developers to take on more architectural roles without getting stuck in the details of implementation.