Are Agents the Future?
AI agents have captured the imagination of technologists and business leaders alike. The dream is tantalizing—a world where agents work tirelessly in the background, handling complex tasks and making life easier, from automating mundane business processes to discovering new drugs. But how close are we to realizing that future, and what should we watch out for along the way?
To help frame the conversation, we asked a few people on our team for their thoughts. Here’s what they had to say about the opportunities, challenges, and realities of agentic systems.
The Vision: Agents That Work for You
"The rosy future we imagined is one where you can set an agent to work and check in later to see results," says Ryman Stringer, envisioning scenarios like drug discovery or automated financial reminders. For instance, imagine a personal assistant that says, “You haven’t charged your roommate for groceries this month like you usually do. Want me to check what you've spent since you last charged them and send a request?” Simple tasks like this seem well within reach.
However, other workflows are harder to automate. Take calendar management. As Ryman notes, scheduling isn’t just about picking an open time slot—it’s about the unspoken nuances we hold in our heads. “I know I have a happy hour the night before, so I don’t want a morning meeting the next day.” Or, “I was hoping to go to the gym at lunch, so I’d prefer a meeting at 2 rather than 1, unless it's an important one in which case I can schedule it, but only I can determine if it's important enough.” These small considerations are nearly impossible for an agent to predict or prioritize without explicit input.
And then there’s the chaos of relying on agents to navigate the web. “Having an agent click around and do tasks for you on a browser sounds like a hellscape,” Ryman warns. As Reed Coke aptly puts it, “If you ever have too much faith in your fellow human, simply look at the HTML source code of 10 different websites.”
The Reality: Hype vs. Practicality
Ben Szuhaj, AI Strategist, offers a pragmatic take: “Agents are often used as a catch-all term for process automation,” with some tasks still better suited for robotic process automation (RPA) or APIs. For more individual or flexible workflows, agents show promise—but only if you’re willing to accept some level of error or risk.
Agents are also an easy way for companies to talk about large-scale AI-driven transformation. “They inspire confidence and help juice stock prices by making it easy to imagine job automation and measurable ROI,” Ben notes. The buzz might be useful for storytelling, but how often does it translate into lasting value?
It’s worth noting that many of the core components we envision for agents are already emerging through other advances. For instance, reasoning models are proving to be increasingly adept at problem-solving, while the growing embeddedness of AI solutions is reducing the friction of computing experiences. In some cases, these advances may come together in the form of powerful agents, but in other cases, agents may not be the best form factor.
Stay Grounded: Practical Tips for Evaluating Agents
Michael Wharton, VP of Engineering, stresses the importance of an evaluation-driven mindset. “Agentic approaches can unlock exciting outcomes, but it’s important to keep a level head,” he explains. Before diving headfirst into agents, he recommends these guiding principles:
- Focus on business value: Keep your eye on the big picture, including long-term costs, system complexity, and legal risks.
- Be rigorous in evaluation: “One impressive demo might not scale. Think about how this solution would perform on a batch of 100 random samples.”
- Consider simpler solutions first: Sometimes, process changes can deliver more value than complex AI systems.
We’re still in the early days of using large language models (LLMs) for agents, and best practices are evolving daily. As Michael puts it, “AI agents are production software and should be treated as such.”
The Future: Agents, Tools, and Alignment
Reed draws a parallel between agents and the iPhone’s evolution into a platform with distributed app developers. In the future, we could see a similar dynamic, with some companies providing agents and others building specialized tools or APIs for those agents.
But this is complicated by a long-running challenge for AI: the alignment problem. “Just because you have an agent doesn’t mean you can ignore alignment,” Reed warns. “If tools and agents are developed separately, the alignment challenge will be even greater.” While LLMs offer some hope in solving this issue, it’s far from a done deal.
Reed also points out that agentic systems aren’t new. “Agents have been around since the ’90s,” he says. So if anyone is pitching you an agent solution and claiming it’s a revolutionary concept, it might be time to ask some tough questions.
Conclusion: Are Agents the Future?
Agents are undoubtedly exciting, but they’re not a silver bullet. As Michael reminds us, they’re just one tool in a broader AI toolbox. Some problems can be better solved with simpler automation solutions or process changes. Others will benefit from agents—if approached with rigor and a focus on business outcomes.
We’re still figuring out what the future holds, but one thing is clear: it’s not about chasing the latest buzzword. It’s about staying informed, grounded, and ready to adapt as the technology evolves.
If you want to explore what agents could mean for your business, reach out to one of our experts. We’d love to help.
