Your future office might look like a scene from a blockbuster movie, but instead of little yellow creatures causing trouble, you'll be dealing with armies of AI software bots. At an MIT conference held in April, tech leaders gathered to discuss the rise of these 'agentic' systems—AI programs that can think for themselves, complete tasks, and, most importantly, talk to each other without waiting for you to click a button.
Debo Dutta, the Chief AI Officer at Nutanix, sees these agents as a version of the Minions from the popular 'Despicable Me' films. He warns that without the right guardrails, these digital helpers could become a massive social experiment that changes how we live and work. The real challenge is determining who controls the 'gurus' that command these swarms and how we prevent them from going off the rails.
This is my cartoon version. I love cartoons and the ‘Despicable Me’ series, so I think of (agents) as ‘minions.’ So if you don't completely put the right guardrail and governance around the minions... it's going to change our society quite a bit.
Most experts on the panel believe we are heading toward an 'internet of agents.' This means instead of just browsing websites, you'll be interacting with an open ecosystem where programs from different companies trade information and negotiate services. Antonio Figueiredo, a senior director at Salesforce, admits that we are currently in an experimental phase that will likely last another 18 months while firms figure out the basics of their workflows.
Partha Madhira, an engineer at Kyndryl, argues that companies must expose their agents as service providers to make this work. By doing this, they can communicate across company lines, which is essential because the business models for tech firms are shifting from selling static software to selling active, autonomous intelligence. Sam Sharaf, a project lead at Google, notes that while internal 'employee agents' handle basic tasks, the internet of agents will increase the scale and complexity of operations significantly.
Rudina Seseri, the Founder and Managing Partner of Glasswing Ventures, suggests a hybrid future that mirrors how we use cloud computing today. Just as we have public clouds and virtual private clouds, we will likely see a mix of open agent networks and locked-down, private corporate systems. She believes that while the public 'internet' of agents will grow, companies will still need private 'intranet' spaces to keep their most sensitive data secure.
Technical Hurdles to AI Autonomy
Security remains the biggest headache for the engineers building these swarms. Seseri points out that identity is the primary vulnerability in multi-agent systems. Without a way to verify exactly which agent is performing an action or holding specific access rights, governance breaks down completely. If an agent cannot prove its identity or authorization, it becomes impossible to manage the handoff of tasks from one bot to another securely.
Sam Sharaf adds that orchestration and trust are the most important layers in the tech stack. If an agent cannot explain what it did, who it talked to, and the reasoning behind its choices, it isn’t an enterprise tool—it's a liability. He likens poorly tracked agent behavior to 'distributed uncertainty,' warning that businesses need better telemetry data to avoid a situation where they lose control of their own automated processes.
Alex Fleck at Microsoft describes this new blueprint as a system that is AI-operated but human-led. This emphasizes that while the agents do the heavy lifting, human judgment must remain the final filter. The core issue is whether businesses will focus on the 'cost of the fuel,' which is the immediate energy and computing power, or the 'cost of the whole trip,' which includes the potential risks and long-term consequences of leaving AI in charge of complex decision-making.
The timeline for this transition is still fuzzy, but industry insiders are watching closely. Partha Madhira predicts that year one will be messy as organizations struggle to adapt, but by year five, the technology will likely become invisible and integrated into everything we do. Rudina Seseri suggests that within five years, we won't just be talking about agents; we will be discussing Artificial General Intelligence, or AGI, which implies machines that can perform any intellectual task a human can.
For businesses, the move toward multimodality—where agents process voice, vision, and robot-driven physical tasks simultaneously—will be the next big shift. This will create new possibilities but also raises concerns about whether it will lead to a 'winner-takes-most' scenario where a few massive platforms control the entire agent economy. The era of the autonomous swarm is already starting to take root behind the scenes of your favorite apps and services, and we will be seeing significant developments in the coming years.