Alex and Pete dissect whether AI agents can actually replace knowledge workers, discussing why language models excel at testable tasks like coding, but fare less well at more general business activities that often require navigating the messy political realities of the corporate landscape.

Alex and Pete discuss how AI agents are collapsing the software development lifecycle (SDLC), eliminating manual coding steps whilst creating new requirements for testing, monitoring, and managing increasingly complex automated workflows at unprecedented scale.

Alex and Pete discuss our problematic tendency to project human characteristic on LLMs, and use anthropomorphic language when discussing them.

Alex and Pete grapple with the ethics of using generative AI in creative work—balancing utility against copyright theft, environmental impact, and career pressure whilst acknowledging there’s no easy resolution.

Alex and Pete examine whether AI will replace jobs within 18 months, discussing automation history, current limitations, and Microsoft’s predictions regarding workplace displacement.

Alex and Pete discuss AI coding agents’ limitations after Alex’s LLM client build revealed extensive bugs and missing features. They explore testing challenges, agent reliability metrics, and corporate liability concerns preventing enterprise adoption. This episode will make a lot more sense if you listen to the previous episode first.

Alex and Pete discuss Alex’s rebuild of his perma-regenerating LLM chat client. This time he did it properly! No messing, just hardcore, disciplined, systematic, agent-coding goodness. Join them as he walks Pete through the process in excruciating, step-by-step detail. The Vanishing Gradients episode Alex mentions.

Alex and Pete examine OpenAI’s comprehensive report on ChatGPT usage patterns, analysing classification methodologies and user behaviour across consumer and workplace contexts. They discuss the shift from advice-seeking to task execution, the dominance of writing and information-seeking functions, and implications for future AI adoption and market opportunities.

Pete and Alex examine AI model evaluation methodologies, comparing traditional machine learning metrics with the qualitative assessment challenges of large language models. They discuss the collaborative requirements between technical and business teams to establish evaluation criteria for generative AI systems, highlighting the subjective nature of testing conversational outputs versus binary classification tasks. With the help […]

Pete and Alex dig into cybersecurity risks with AI agents and generative AI systems. They cover two main problems: people coding dodgy applications without security knowledge, and hackers directly exploiting AI agents that have access to tools and data. Despite the scary possibilities, they reckon most vulnerabilities are manageable with decent security practices. Practical AI […]