In September 2025, farm tech®
As we move towards 2025 and 2026; farm tech® interviewed Bryant as a supplementary video to these articles. This article will also be published in three parts. In this first article, Bryant distinguishes between autonomy and accountability for AI agents and explains how these characteristics must work together. Bryant emphasizes that “autonomy is not a binary” when it comes to implementing agent AI in the pharmaceutical industry.
“You often see parades in that sense,” Bryant said. “In reality, autonomy exists in many different realms, and in regulated realms like ours, autonomy with accountability is always a design choice. By the way, autonomy and accountability, this sounds contradictory, but autonomy without accountability is really chaos. But the opposite is also true: accountability without autonomy can cause paralysis.”
Bryant then predicts what will happen in 2027 as well as 2026.
“Our design goal for the next two years is to increase the level of autonomy,” Bryant says. “So the agent AI can plan, it can adapt, and it will adapt as it runs. You can see the approach change. You can reflect on the results. The AI can learn from the results, and it can learn in real time and refine its approach while collaborating with other agents.”
The first part of Bryant's interview can be seen above.
3 articles written by Bryant are available
transcript
Editor's note: This transcript is a lightly edited version of the original audio/video content. It may contain errors, informal language, or omissions spoken in the original recording.
I'm Jason Bryant. I work at Alice Global. At ArisGlobal, I'm the Senior Vice President of Product Management for AI. I'm also the general manager of their flagship GenAI product called NavaX.
OpenAI recently announced Sora 2, a video generation model. And I was struck by the way the company's CEO, Sam Altman, essentially framed the purpose of AI. AI is not just about productivity, he said, but about new possibilities. We are witnessing that change in the pharmaceutical industry. We now have intelligent autonomy, a term that can actually enable goal-driven reinvention. It's not just that today's versions are faster and higher quality. We're currently building for the second wave of agent AI. And agent AI here unlocks decision-making intelligence. This is true insight. This is unlike anything we've seen with traditional business intelligence techniques and tools.
Maybe if you take a step back and give a really simple definition of agent AI, it's really about one goal orientation. In other words, these are agents working toward a goal, and they achieve this through their second characteristic: autonomy. And autonomy here introduces controllable degrees of freedom within safe boundaries, which is why agent AI enables this new possibility. As you say, it's a reinvention. It's not just about productivity.
Here are two examples from a productivity perspective. The next wave might consider using agent AI to reduce resources. We are thinking about resource optimization. Therefore, it's not just about reducing headcount, it's about how those resources are deployed. Alternatively, the aim is not only to reduce cycle time, as is often the case with efficiency goals, but also to reduce decision time. When it comes to processes, agent AI means that workflows can be dynamically adapted. This means these workflows can be re-prioritized and escalated based on scenarios. It can provide a challenge. Can suggest alternatives. And this is where process reinvention comes into play. AI is not prescriptively taught how to work. In reality, we adapt our reasoning, approaches, and actions based on how best to achieve that goal.
Autonomy is not a dualistic thing, and we often see autonomy in that sense exaggerated. In fact, autonomy exists in many different realms, even in regulated realms like ours. The design choice is always to have autonomy with responsibility. Now, autonomy and responsibility, that sounds contradictory. Autonomy without responsibility is indeed chaos, but the reverse is also true. Responsibility without autonomy can cause paralysis. So we think in terms of bounded autonomy, which effectively defines an envelope, defines constraints, but allows us to move freely within it. So these are about scope limits, certainly guardrails, conditions that trigger escalation, and that's primarily against humans, so when you think about the scope of autonomy, think about the lower end of the scope. Once again, those days are gone. Very valuable, still valuable, and increasingly valuable, but with relatively limited automation locked in and powered by AI.
And the current reality of agent AI is what I would call a fairly narrow range of agents. Therefore, they can interpret intentions and can do some things, but their actions are very strictly limited. That's what I mean by narrow range. For example, medical writing in our industry. But looking to the future, our design goal for the next two years is to increase the level of autonomy. This means agent AI can plan, it can adapt, it can adapt during execution, and you can see its approach change. This can be reflected in the results. You can learn from them. You can learn from them in real time and refine your approach while collaborating with other agents.
