Ashu Garg reflects on AI developments in 2025 and makes predictions for 2026. He acknowledges significant trends, like the rise of reasoning models and compound AI systems that have shifted the focus from single models to integrated systems capable of complex tasks. OpenAI remains a leader but faces competition from newcomers like Gemini and Claude, which are gaining traction. The economic model of AI has evolved, with startups targeting services instead of just software, fundamentally changing how businesses approach AI. Garg highlights the struggle of integrating AI into enterprise workflows, noting that while many companies have experimented with pilots, genuine deployment remains challenging. He anticipates that 2026 will see startups partnering with enterprises to create environments where AI can learn and improve on real workflows. This will likely involve smaller, customized AI models that operate within corporate infrastructure, prioritizing security and efficiency. He argues that the future of decision-making in enterprises will rely on a "context graph," which captures decision histories and allows organizations to automate processes more effectively. Security is a major concern as AI agents take more autonomous actions. Garg predicts that by 2026, AI security will become a board-level priority, with companies adopting zero-trust principles for AI systems. He warns that the risks associated with AI agents include potential data breaches, and he expects at least one high-profile incident that could escalate concerns around AI security. Major SaaS players like Salesforce are responding by integrating AI deeply into their offerings, indicating a shift in how these companies view AI as an essential component rather than an add-on feature.