# Where AI is going in 2026 Source: https://www.youtube.com/watch?v=ZtQjILC-m3I Recap page: https://rapidrecap.app/video/ZtQjILC-m3I Generated: 2026-01-04T16:08:29.249+00:00 --- ## Quick Overview The speaker predicts that by 2026, AI tools like those from Anthropic (Claude) and OpenAI (GPT) will be so capable that 90% of coders' work will be AI-generated, leading to a societal shift where human focus moves away from basic coding/research toward higher-level tasks and problem articulation, fundamentally altering the nature of work, similar to how cars superseded horses. **Key Points:** - The speaker predicts that by 2026, 90% of code written by developers will be generated by AI like Claude and GPT. - The speaker notes that current AI models already demonstrate a significant ability to perform tasks previously requiring human effort, such as generating reports and code. - The speaker cites a tweet from an economist suggesting that AI's ability to generate novel thought is difficult, but the speaker disagrees, pointing out that this is where human intuition is still required. - The speaker contrasts this with the current state, where AI is already writing 90% of code for some users, and notes that the bottleneck is shifting from execution to defining the correct problem. - The speaker refers to the 'normalcy bias'—the human tendency to stick with what is known—as a barrier to accepting the rapid advancements in AI capabilities. - The speaker concludes that the future of work will involve humans acting more as supervisors or directors, offloading cognitive heavy lifting to AI, similar to the transition from horses to automobiles. ![Screenshot at 00:56: The speaker discusses recent papers, noting that DeepSeek has repeatedly demonstrated state-of-the-art performance with models significantly smaller than competitors, suggesting efficiency gains are a key area of progress.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-00-56.jpg) **Context:** The speaker is discussing the accelerating pace of Artificial Intelligence development, specifically focusing on Large Language Models (LLMs) like Anthropic's Claude and OpenAI's GPT. The discussion centers on the speaker's prediction for 2026, where AI will dominate code generation, forcing a shift in human roles toward problem definition and high-level oversight, drawing analogies to historical technological displacements like the invention of the automobile. ## Detailed Analysis The speaker is conducting an experiment using only completely unedited video recordings to discuss predictions about the future of AI in 2026. The core argument revolves around the increasing capability of models like Claude and GPT, which the speaker believes will lead to a massive shift in developer roles. The speaker states that by 2026, 90% of a developer's code will be AI-generated, citing figures suggesting that 90% of Microsoft's code was AI-written by the end of the previous year. This capability is already leading to a 'spillover effect' where tools like Notebook LM and Claude can handle complex tasks like generating reports, infographics, and charts, freeing up human mental energy. The speaker refutes the idea (attributed to an economist and referencing phantom limb syndrome) that humans cannot generate truly novel thought, arguing that the hard part is shifting from execution to correct problem articulation and validation. The speaker concludes that this trend mirrors historical technological shifts, like the transition from horses to cars, where the infrastructure adapts to the new superior technology, making previous methods (like manually writing all code) obsolete. ### AI Progress and Size - DeepSeek repeatedly demonstrated state-of-the-art performance with models much smaller (e.g., low millions of parameters) than competitors like GPT-4/5.2 - This efficiency is key, allowing for broader adoption and computational offloading. ### The Role of Imagination vs. Execution - The speaker contrasts the human tendency to avoid 'novel thought' due to 'normalcy bias' with the AI's ability to generate novel ideas; however, the current bottleneck is defining the correct problem, not execution. ### Future of Work in AI - By 2026, the speaker predicts 90% of code will be AI-written, turning developers into supervisors or managers who direct the AI rather than writing boilerplate code. ### Spillover Effects and Tool Utility - Tools like Notebook LM and Claude are already versatile, handling tasks from report generation to data retrieval, which forces a shift in value away from low-level coding toward high-level problem framing and validation. ![Screenshot at 00:00: The speaker begins the unedited recording, addressing the audience directly into the microphone.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-00-00.jpg) ![Screenshot at 00:48: The speaker uses a hand gesture to emphasize the positive aspect of AI acceleration: the ability to offload cognitive work.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-00-48.jpg) ![Screenshot at 01:58: The speaker mentions that there have been many papers discussing the trend, highlighting the volume of research in the field.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-01-58.jpg) ![Screenshot at 04:06: The speaker points out that the Unix design philosophy \(commands do one thing and do it well\) contrasts with the current trend of expecting a single tool to handle everything.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-04-06.jpg) ![Screenshot at 09:08: The speaker uses a finger gesture near their temple to illustrate the human instinct to doubt AI's capability when it produces something novel.](https://ss.rapidrecap.app/screens/ZtQjILC-m3I/00-09-08.jpg)