We Built Microsoft Teams in 23 Minutes (And You Can Use It) & GPT 5.4 Impressions - EP99.37

Chris and the co-host discuss the recent release of OpenAI's GPT 5.4 and its significant features, including a million-token context window and enhanced reasoning capabilities. Initial impressions indicate that GPT 5.4 is a substantial improvement over its predecessor, with a reasonable pricing structure. The co-host shares their experience using the model, noting its methodical task completion and adjustable reasoning levels for improved speed.

The conversation explores the implications of GPT 5.4 for knowledge work, particularly in Microsoft Office applications like PowerPoint and Excel, and highlights a new ChatGPT add-in for Excel. The speaker reflects on building a plugin for Sim Theory, emphasizing AI's potential to enhance productivity, especially given dissatisfaction with existing tools like Copilot. They discuss advancements in AI's ability to manage context and perform tasks autonomously, reducing reliance on manual processes.

The speaker summarizes their thoughts on GPT 5.4, noting its role in increasing market competition and driving down prices. They introduce a "tool search" feature for efficient tool usage and highlight the importance of context management. While acknowledging a downside in processing speed compared to previous models, they suggest that GPT 5.4's cost-effectiveness may justify this trade-off.

Financial implications of using AI models are discussed, with costs of $30 per million input tokens and $180 per million output tokens. Skepticism arises regarding the value of these expenses, particularly for agentic processes where output tokens can accumulate quickly. The conversation acknowledges recent advancements in AI, such as recreating a fully functional version of Minecraft, and critiques models like Gemini 3.1 for falling behind competitors.

The speaker reflects on the current state of AI models, emphasizing the absence of a competitive Google model and the need for innovation in agentic models. They discuss a project where a Trello clone was created due to dissatisfaction with existing features, highlighting its user-friendly design and operational capabilities.

A shift in work dynamics is emphasized, advocating for a community-driven approach to software development that allows users to install and maintain apps easily. The speaker notes the successful recreation of a version of Microsoft Teams, which supports multiple video chats and various communication channels, achieved through simple prompts.

The potential for companies to replace costly software contracts with customizable solutions is highlighted, predicting a significant shift in operational practices as AI becomes central to daily tasks. The concept of "agent apps" is introduced, which are AI-driven applications capable of managing tasks and communicating with other agents. The speaker stresses the importance of focusing on productivity and actual tool usage.

To enhance existing software, two methods are proposed: using a browser plugin to add features and developing new applications that consume data from existing systems. The speaker envisions a future where AI-enhanced versions of existing software run in parallel, phasing out original systems, and emphasizes the importance of effectively communicating with AI models.

Concerns about the relevance of older programming languages and the potential redundancy of skills are raised, but the speaker believes there are ample opportunities for building new systems. They discuss the complexities and potential risks associated with advanced technology, particularly AI agents that can operate on behalf of users, transforming content creation and facilitating easier experimentation.

Safety and security measures are emphasized as crucial in developing new features, advocating for a curated mix of skills and models to enable effective use of AI agents. The efficiency of AI in handling tasks that typically require manual input is underscored, with examples of AI managing communications effectively. The speakers stress the importance of human involvement in documenting project outcomes and communicating them professionally.

Recent developments in AI, including Microsoft Teams, are discussed, highlighting their capabilities and affordability. The speaker expresses excitement about AI's role as a "delegation machine" that boosts productivity, particularly in managing administrative tasks. They also mention plans to create YouTube videos showcasing their workflows and emphasize the effective organization of their workspace.

A new channel named "Tralow Feedback" will be established on Microsoft Teams to gather user feedback. The speaker acknowledges the pressure of meeting a self-imposed deadline for a new release and humorously invites audience suggestions for potential punishments for missing this deadline. They express hope that listeners will share their enthusiasm and mention being available on Microsoft Teams for further engagement.