I recently sat down with Ksenia from Turing Post TV to talk about what I believe is one of the biggest challenges and opportunities as enterprises move from experimenting with AI to deploying agents at scale. The model is only one part of the equation. For agents to do real work inside an enterprise, they need trusted context, the right identity and permissions, access to business logic and actions, strong security and governance, and the ability to use the right model for the right outcome at the right price/performance. Most AI harnesses focus only on the agent. Salesforce’s Enterprise AI Harness focuses on making the entire enterprise ready for all agents, providing the trusted context, models, agency, actions, security, governance, and control that any agent needs to do real work. But there is an even deeper challenge. Much of what makes a business actually work has never been fully captured in software. The reasoning behind a decision. The context shared in a meeting or 1:1. Why an exception was made. The tradeoffs a team considered. Much of this critical business context lives across conversations, in the minds of a few key people, and other unstructured forums, rather than being captured in a structured, reusable form. Closing that gap and turning this institutional knowledge into trusted enterprise context that agents can reason over may be one of the most important problems we have to solve. We covered this and much more. Agent identity, security, model routing, context, economics, what it takes to move agents from interesting pilots to meaningful production systems, and even some advice for students and people early in their careers navigating this extraordinary period of change. Thank you, Ksenia, for the thoughtful conversation. Watch the interview here: youtube.com/watch?v=ewm5Ubzb… One recommendation: watch it at 1.25x speed.

Dubbed Koa, the new model is designed to power sales and service agents. cio.com/article/4222320/sale…

Salesforce debuts Koa, a specialized model built to reason about CRM data siliconangle.com/2026/09/15/…

At Dreamforce, we introduced Koa, Salesforce’s first CRM reasoning model, built on NVIDIA Nemotron. AI is getting smarter at an extraordinary pace. But the Agentic Enterprise needs more than intelligence. It needs AI that understands how businesses actually work, their context, workflows, policies, tools, and decisions. That’s why we built Koa. Koa brings nearly three decades of Salesforce CRM expertise into a purpose-built reasoning model for Agentforce. It is post-trained entirely on synthetic data across the customer lifecycle, runs inside the Salesforce trust boundary, and uses no customer data for training. And we’re already seeing what specialized intelligence can unlock: greater precision, reliability, and context retention across complex CRM tasks. Koa isn’t about one model replacing every other model. I believe the future is much more exciting than that. The future of enterprise AI is open, with the right model for the right job, surrounded by trusted context, actions, security, and control. Koa is an important step toward that future. A huge thank you to our teams at Salesforce and NVIDIA for bringing it to life. I can’t wait to see what our customers build with it. lnkd.in/gCFU6Qcd