Did Clawdbot Just Show Us the Future of AI Workers? & Kimi K2.5 Dis Track Tested - EP99.32

Chris provided an update on the Still Relevant Tour, focusing on Australia and considering additional locations like the UK and US due to high interest. He introduced Multbot, an AI assistant developed by Peter Steinberger, which has gained attention after a name change prompted by a lawsuit threat. Multbot is locally hosted and capable of interacting with various platforms, but it faces security concerns, including vulnerabilities to prompt injection attacks and excessive token usage due to hijacking incidents.

The appeal of having an AI assistant as a digital coworker was discussed, highlighting the integration of open-source tools into user-friendly formats that enhance productivity. The project boasts around 3,000 skills, allowing users to add more and fostering a collaborative environment for developers. The speakers noted that while many AI use cases are mundane, the concept of a virtual employee is exciting, with AI now contributing actively rather than merely following instructions.

The conversation emphasized the evolving role of AI tools in the workplace, predicting a shift towards "agentic workflows" where personal assistants perform tasks autonomously. The diminishing need for technical expertise was highlighted, as users can now navigate AI tools independently. The importance of coordinating inputs and outputs, rather than just focusing on coding, was stressed, along with the necessity of maintaining a human-centered approach in product design.

Cognitive overload from managing multiple tasks was acknowledged, with a call for human involvement in reviewing AI-generated work. The speakers discussed the need for clarity in goals when using AI outputs and proposed a teaching mode for AI to assist users in effective communication. They also highlighted the importance of being goal-oriented and understanding AI-generated outputs.

The discussion included the current focus on building agents and orchestrating AI systems, alongside uncertainties about AI's future impact. The challenges of accessing off-the-shelf solutions for working with multiple Managed Cloud Providers (MCPs) were noted, as well as the technical expertise often required for configuration. The conversation speculated on how AI could transform the work of traditional knowledge workers, such as doctors and lawyers, by streamlining workflows.

The advantages of local computing solutions over cloud services were discussed, particularly regarding privacy and data security. The introduction of Kimmy K 2.5, a visual agentic intelligence model, showcased its advanced features, including a 256k token context window and the ability to manage up to 100 sub-agents in parallel workflows. The model's open-source nature and new vision capabilities were also noted.

Market dynamics are shifting with the rise of cost-effective agentic models like Malt Bot, appealing to users seeking affordable automation solutions. The value of sub-agent architecture for enhancing efficiency through task breakdown was recognized. The introduction of K to 2.5 Agent Swarm marked a transition to self-directed execution using parallel agent reinforcement learning, with challenges in coordination acknowledged.

The critique of Anthropics' approach pointed to a lack of corrective action opportunities in their handling of tool calls. The conversation also featured commentary on two diss tracks, focusing on performance metrics and cost-effectiveness. Advancements in model capabilities, including the creation of a black hole simulator and a full-fledged game using local code, were discussed, showcasing the seamless integration of generated assets.

The conversation concluded with skepticism about the decline of Software as a Service (SaaS) and potential disruptions in the industry, emphasizing the importance of security compliance in agent settings. Excitement built around an upcoming product version with significant improvements, alongside discussions about the high costs of running inference and competitive pricing. Performance metrics, particularly the speed and efficiency of agent swarm technology, were a key focus, with critiques of competitors regarding pricing and performance. The strength of open-source solutions in democratizing AI was emphasized, along with a commitment to enhancing capabilities and breaking industry norms.