Why autonomous AI “agents” will be so important to your business

As AI agents evolve from tools into teammates—fulfilling tasks with minimal supervision—they portend a shift in how firms should invest.

In brief

• Autonomous AI agents are evolving into strategic business partners, capable of autonomous task execution.
• Agents' perception, interaction, and autonomy will revolutionize software creation and business operations.
• Investing in AI now requires a focus on building agents with access to quality information and context.

Last year, a team of Stanford-led researchers gave The Sims an upgrade, creating a game named “Smallville” stocked with 25 generative AIs (GenAI). Powered by a large language model (LLM) similar to ChatGPT, the characters were given brief backstories and the ability to remember, reflect, plan, and execute tasks without being given explicit instructions. When one was initially assigned to host a Valentine’s Day party, for instance, the character invited friends, decorated the game’s café, and even played matchmaker—all of its own accord. Party planning was just the beginning, however. Within weeks of the game’s source code being freely available, another team put the town to work—this time making video games.

Smallville’s residents are examples of “agents”—software that can perceive, act, react, and interact with both humans and other agents alike in achieving its goals. While their complexity and autonomy may vary, agents represent a fundamental shift in how applications are built and how organizations are run. Hallucinating chatbots may soon be replaced by fleets of AIs fact-checking each other, thus vastly increasing their reliability. Programs will evolve from tools into teammates, creatively fulfilling tasks with minimal supervision. Agents have already demonstrated their skills in trading stocks, designing products, and dozens of other use cases (at least in a lab setting).

In doing so, they also portend a change in how firms should approach and invest in artificial intelligence (AI). “The most powerful agents are not going to be the smartest, but the ones with access to the best information,” says Rakesh Malhotra, principal, Digital & Emerging Technologies, Ernst & Young LLP, and former cofounder of the AI consulting firm Nuvalence. Creating agents that can be trusted with mission-critical roles will not only require building a new technical stack that is safe and secure with the appropriate guardrails in place, but also personal and business data with more context than any chatbot could muster. “I don’t need a model with the maximum number of parameters,” Malhotra adds. “I just need the one that knows me best.”

Perception, interaction, autonomy

How will they know what’s happening? This is where context is crucial. A chatbot trained on a trillion parameters scraped from the web is ultimately less useful than a trusted agent with access to your most intimate data. Multimodal AI will not only spawn agents able to see and hear, but also wearable devices expressly designed to give them eyes and ears. As Malhotra explains, “The agent that knows the most about me doesn’t just have my personal data, but also my current context: Where am I? What am I doing? What time of day is it? Having all that information at its disposal helps it to make the best decision.”

Taken together, these traits—perception, interaction, and autonomy—will transform how software is made and how firms operate. Like the researchers who set Smallville’s residents to work making video games, agents will be able to write code on the fly, leading to a new generation of precision software tailored to each user rather than one-size-fits-all. And that, in turn, will lead to a dramatic decrease in switching costs between applications as agents handle the translation.

Likewise, they will lower the switching costs of people within firms as well. As the tacit knowledge of how work gets done is transferred to agents, onboarding new talent and augmenting others will become progressively easier and less costly for all involved. “The fluidity of communication and interaction will be significantly improved,” Malhotra says.

If that sounds like reason to celebrate, great—agents already know how to throw a party.

This article was originally published on Fast Company.com.