954: Recap of 2025 and Wishing You a Wonderful 2026

Episode Summary

Executive Summary: John Crohn’s year-end episode reflects on 2025 as the year agentic AI became practical, though still overhyped, and urges preparation for greater machine autonomy through data, security, and governance. He also celebrates the podcast’s strong growth from video expansion, thanks listeners, acknowledges global hardships, and previews a 2026 trends episode plus the upcoming 1000th episode.

Main Topics: 2025 as the year of agentic AI: Crohn says the expected shift toward agentic AI did occur in 2025, with increasingly autonomous systems becoming more capable in real-world tasks. Preparing for autonomous AI systems: He stresses that organizations should ready their data, security, and governance now to benefit from rapidly improving LLM capabilities. Podcast highlights and notable guests: He reflects on standout episodes, including guests like Azeem Azhar, Andre Berkhoff, Carl Benedict Frey, and a memorable live recording with Aurelian Geron in New Zealand. Podcast growth through video strategy: The show reworked operations to succeed on video platforms, especially YouTube, resulting in major growth in watch time and subscribers. Gratitude and acknowledgement of hardship: He notes that the year was not without pain, including AI-related job insecurity, war, and personal loss, and expresses gratitude for listeners’ support. Looking ahead to 2026 and episode 1000: Crohn previews a follow-up episode with Sadie St. Lawrence on 2026 trends and says he needs something special for the podcast’s 1000th episode in June.

Key Arguments: Agentic AI is overhyped in the discourse, but genuinely powerful and increasingly relevant. Leading proprietary LLMs can already replace humans on two-hour software development or ML engineering tasks with about 50% accuracy. Because task duration doubles every seven months, organizations should act now to prepare infrastructure and governance for faster AI autonomy. Video-first distribution significantly improved the podcast’s reach and engagement, validating the operational shift to YouTube. Despite industry growth, the year brought real human costs and uncertainty, so gratitude and perspective matter. Listener feedback will shape the show’s next phase, especially as it approaches episode 1000.

Data Points: Task duration handled by leading proprietary LLMs: 2 hours - Crohn says current models can replace humans on software development or ML engineering tasks of this length with 50% accuracy. Accuracy on those tasks: 50% - Estimated success rate for leading proprietary LLMs on two-hour-long software development or ML engineering tasks. Task duration improvement cadence: Doubles every 7 months - Crohn uses this trend to argue that capability gains are accelerating quickly. YouTube watch time growth: 10x - Total YouTube watch time per day increased two months after implementing video changes in August. YouTube subscribers a year ago: 25,000 - Subscriber count before the video-focused overhaul. YouTube subscribers today: Over 140,000 - Subscriber count after the podcast’s video expansion. Upcoming milestone episode: Episode 1000 in June - Crohn notes the show will soon reach its 1000th episode.

Pivotal Quotes: "We now stand at a point where leading proprietary LLMs can replace humans on two-hour-long software development or ML engineering tasks with 50% accuracy." — John Crohn: Used to illustrate how far agentic AI has progressed and why preparation is urgent. "It is overhyped, for sure, but it is also tremendously powerful." — John Crohn: His balanced take on the term “agentic AI.” "We implemented them in August, and merely two months later, our total YouTube watch time per day had increased 10 times over." — John Crohn: Describing the impact of the podcast’s video-platform strategy.

Implications: Listeners and organizations should prepare now for more autonomous AI by strengthening data, security, and governance. For the industry, video-first content and practical AI adoption are becoming key competitive advantages.