Weekly deep dives on the most interesting dev, ai and product releases, research updates and emerging trends in the AI engineering, agent development and software industry.
Ep24 - 5.2, gdpeval and arc smash, joint embedding models deep dive
Good to be back!
Show notes to follow
15 Nov 2025TRANSCRIPT
EP 22 - DGX update, Kimi k2 thinking as new workhorse, deep agents Langchain, and Code Mode for MCP
In this conversation, Tom Spencer and Cameron Rohn explore various themes surrounding AI, including nostalgic references to early AI tools, the evolution of product management with AI, and the innovative capabilities of deep agents and open-source models. They discuss the importance of context in AI communication, the iterative learning process of AI agents, and the challenges of managing expectations with AI outputs. The conversation also delves into the role of long-term memory in AI, the emergence of Kimi K2 Thinking, and the intricacies of fine-tuning AI models for specific tasks. In this conversation, Cameron Rohn and Tom Spencer delve into the intricacies of fine-tuning AI models, discussing the challenges and opportunities that arise in this domain. They explore the innovative applications of fine-tuning, particularly in the context of Nvidia's ecosystem and the implications of dynamic worker loaders and code execution. The discussion also touches on the future of AI applications and user experience, emphasizing the need for better interfaces and the potential for new business models in the AI landscape.Chapters00:00 Nostalgia and AI Beginnings04:28 Exploring AI Agents and Tools09:21 Deep Agents and Workflow Automation29:20 Exploring the DeepAgent CLI32:33 Understanding the Deep Agents UI37:29 Creating Tailored Co-Pilots for Specialists42:30 Integrating Long-Term Memory in Deep Agents46:39 The Future of Agent Management in Organizations54:44 Kimi k2 Thinking and Open source59:04 Fine-Tuning Models for Custom Needs01:16:23 Code Execution and MCP Innovations - Demo01:28:11 Exploring DeepAgents and Code Sandbox Support01:32:56 Creating Multi-Step Workflows with MCP Tools01:37:56 User Experience and Control in AI Tools01:42:54 The Future of AI Applications and Business Models
2 Nov 2025TRANSCRIPT
EP 20 - Claude Skills, LangChain Agent builder, market updates and Options MCP
Notes to come...
buildaipod.com
Inside a recent episode
Ep24 - 5.2, gdpeval and arc smash, joint embedding models deep dive
Published 13 Dec 2025 · Transcript excerpt
[…] And they're like, really? I didn't. Apparently, that's what. Reminem out of. That's what has been reported. And the evidence to suggest that is the knowledge cutoff times right. Because it comes out. And then it comes from goes from instruct to chat to a chat model. Right. And then they can then they put a wrapper around it. Which is like effectively the fine tune or not fine tunes the wrong word where they basically constrain it and put the safety guard rails on top. The example we talked about it a while back. Anthropic actually publishes their evaluation set where they constrain the model as opposed to an unhinged model. […]
Pod Engine is an independent podcast discovery and analytics service and is not affiliated with or endorsed by this podcast. Artwork and show content belong to their owners. Full legal notice.
Explore this show Podcast research with Pod Engine
