A panel at MIT’s Imagination in Action conference last month laid out the unsolved problems that stand between personal AI agents and broad consumer adoption, according to a Forbes writeup published May 26 by MIT Senior Fellow John Werner.

The panelists converged on three blockers: consent infrastructure for vulnerable users, hosting and inference costs that limit access, and hardware constraints that prevent agents from running outside cloud environments.

OpenClaw as Inflection Point

Maria Gorskikh, an MIT researcher working on project NANDA, identified OpenClaw as the catalyst for mass personal agent adoption.

“What happened that was really important this year was OpenClaw,” she told the panel, according to Forbes. “OpenClaw was specifically important because it kind of kicked off the mass adoption of personal AI agents and it went viral: and now non-tech people and tech people together are excited about the world of agents and the agentic web.”

Gorskikh later identified the two cost barriers she considers most urgent: hosting costs and inference costs. “I think we really need to solve these two barriers in order for everyone to have an agent that is cheap, reliable, secure, and accessible,” she said.

Uzma Farheen, Director of Keep AI Safe, focused on a problem most agent frameworks ignore entirely: users who cannot consent to agent actions on their behalf.

“We are really looking into making a very specific use case, that is for the people who do not have the agency to give consent, because personal agents need consent,” Farheen said, per Forbes. Her organization is building platforms where agents interact with each other in simulated environments to “see where the agents and the models break.”

Farheen’s point touches a gap in current agent architectures. OpenClaw, Claude Code, Hermes Agent, and most frameworks assume a technically competent user who can configure permissions, approve actions, and understand what their agent is doing. Populations that include elderly users, children, or people with cognitive disabilities have no consent framework designed for their needs.

“When innovation happens, safety kind of goes on the back burner,” Farheen said.

Edge Computing and Hardware Minimalism

Jordan Tian, co-founder of ZeroClawLabs, argued that personal agents should be deployable on hardware as small as a Raspberry Pi. “The idea was, we wanted to go all the way down and make something that could be small and deployed on any hardware,” he told the panel, as reported by Forbes. “At the time, everyone’s looking at sandboxing and getting their own Mac Mini, and we were, like, what if you could make it smaller?”

The edge computing angle connects to Gorskikh’s cost argument. Cloud-hosted agents generate ongoing hosting and inference bills that create a floor on monthly operating cost. Agents running on local hardware with lightweight models eliminate the hosting component, though inference quality degrades with smaller models.

Jenny Lay-Flurrie, who leads Microsoft’s Trusted Technology Group established in early 2025, raised the equity dimension in a separate CNBC interview referenced by the Forbes piece. “How do we make sure that we build it right?” she said. “Society is not always the most inclusive place, so there are instances where we have to insert data to train it.”

The Attention Economy Reversal

Greg Raiz, general partner at Founder’s Edge, offered what may be the panel’s most counterintuitive framing. He argued personal agents could reverse some of the attention fragmentation that earlier AI tools created.

“AI has actually split our attention in multitudes of directions,” Raiz said, per Forbes. “I think agentic tools start to be a success where they’re giving our attention back, because they are surfacing the things that are actually important to us.”

Moderator Gunjan Sinha sharpened the point: “Is it possible that the agents are here to help fix our ADHD that they created in the first place for us?”

The panel did not resolve the question. But the framing captures the tension at the center of personal agent design: these tools promise to act autonomously on behalf of users, but the infrastructure for ensuring they act correctly, affordably, and equitably does not yet exist.