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

“Application companies can act as the outsourced AI team for enterprises. Instead of building a generic horizontal tool, these vertical startups are becoming the specialized partners that legacy businesses need to navigate the transition, effectively abstracting away the complexity of model volatility.”

“We are seeing that 'skills' may be the minimal viable packaging format for agents. Infrastructure companies have had to reinvent themselves every year, while application companies focusing on specific skills have had an easier time surviving the constant shifts in the underlying model landscape.”

“The agent lab playbook starts with frontier models and specializes for your domain. Eventually, you train your own models once you have enough data, workload, and user behavior to justify the cost and latency savings, moving from general intelligence to highly efficient domain specialization.”

“I've turned much more bullish on open source lately. Every 10x speedup in inference and the ability to run these models on custom chips or alternative infrastructure is unlocking product experiences we couldn't imagine a year ago, allowing for more localized and private control.”

“Today’s models mostly reward frequency of mentions, but in the future, we expect product choice to be shaped much more by personalized memory systems. Memory and personalization are becoming the next big wedge for consumer AI, creating deeper moats for the companies that get it right.”

“Coding has become one of the largest and fastest-growing categories in AI. Between Anthropic, OpenAI, Cursor, and Cognition, we are seeing a massive wave of innovation that still has significantly more room to run as we move from basic assistance to full agentic software engineering.”

“Non-NVIDIA hardware is suddenly getting real attention from the engineering community. As the industry moves from pure capability exploration to efficiency, the focus on alternative inference infrastructure and custom silicon has become a top priority for scaling production-ready agents.”