The last two years in AI have been disorienting. Every week brought a new model, a new product launch announced on X, and a new reason to reconsider assumptions that felt settled only months before. For startups trying to keep up, the experience has been less like riding a wave and more like trying to read a map while the terrain keeps shifting beneath you.<\/p>\n\n\n\n
It was something I discussed on another episode of The Spiro Circle only a few weeks ago. But this time, according to Tom Findling, Co-founder and CEO of Conifers, the disorientation was merely a prologue for what\u2019s to come.<\/p>\n\n\n\n
\u201cThe last two years were a promo to what\u2019s going on,\u201d he said. \u201cI think starting in 2026, you really see the maturity, both on the model side and the agent side, to come and really disrupt the enterprise.\u201d<\/p>\n\n\n\n
If the explosion of GPT-4, autonomous agents, Claude\u2019s product launches, and multimodal models were just the opening act, what exactly is coming next?<\/p>\n\n\n\n
Findling claims the answer is in a convergence that is only now becoming visible. For most of the AI boom so far, the two core components of enterprise deployment (the models themselves and the agent frameworks built around them) were developing on separate tracks. Models were getting smarter, and agents were getting more capable – but the combination wasn\u2019t yet reliable enough to deploy seriously inside a large organization. \u201cI think we got to the point that agents got to a certain level of maturity when you can deploy them in the enterprise, you can monitor them, you can scale them, and the models themselves have become extremely smart.\u201d<\/p>\n\n\n\n
SOC teams face alert volumes that no human workforce can realistically process. Analyst shortages are chronic, and Findling estimates that fewer than one percent of applicants for analyst roles can actually do the job. Previous attempts to close the gap were built on rule-based logic that couldn\u2019t adapt to the context-dependent, environment-specific nature of real incident investigation. \u201cWe tried for the last decade to solve that problem,\u201d Findling said. \u201cUnfortunately, it didn\u2019t go that well. That\u2019s why we still have business.\u201d<\/p>\n\n\n\n