AIE Europe Debrief + Agent Labs Thesis: Unsupervised Learning x Latent Space Crossover Special (2026)

In this special crossover episode of 'Latent Space: The AI Engineer Podcast' with Swix from Leighton Space, the hosts dive deep into the state of the AI coding wars in 2026, framing it as a pivotal year where coding agents are breaking containment to expand into all domains. The conversation explores the maturation of AI infrastructure, with a focus on stability in 'harness engineering' and 'context engineering' after years of rapid reinvention. Key themes include the rise of domain-specific models, the strategic shift toward in-house model training for cost and latency savings, and the growing importance of alternative hardware like Cerebrus and Talos. The hosts debate whether foundation models are eating into startup categories, concluding that while midsize startups face pressure, the real opportunity lies in AI-native application layers. A major insight is that the future of AI is not just about better models, but about new paradigms—like zero human review in software development and the emergence of world models that enable true spatial and physical understanding. The episode ends on a note of optimism, emphasizing that the most successful builders in 2026 will be those who embrace volatility, experiment relentlessly, and focus on compounding advantages through personalization and memory systems.