Where Should AI Agent Autonomy End? Our CEO Weighs In for MIT Sloan Management Review

How much should we let an AI agent decide on its own? That question sat at the center of the latest Responsible AI panel from MIT Sloan Management Review and Boston Consulting Group, published this week, and our CEO Richard Benjamins was one of the international experts invited to respond.

The panel put a blunt provocation to more than fifty AI academics and practitioners: "Responsible governance that treats agents as autonomous decision makers will fail." Seventy-two percent agreed or strongly agreed. But the interesting part of the piece is not the headline number; it is what the panelists mean by autonomy. Several point out that today's agents are autonomous only in an operational sense: they plan, call tools and execute steps within limits someone else has set. Technical independence, as Stanford's Riyanka Roy Choudhury and others argue, does not translate into moral or legal accountability. An agent cannot be sued, cannot lose a license and cannot own a consequence. Treating it as though it can, several panelists warn, simply gives humans and institutions a way to launder responsibility through the machine.

Richard's answer sits in the more nuanced middle of the debate. As he put it in the article, "impactful decisions should not be fully autonomously taken by AI agents," while "trivial decisions can be." In other words, the right level of autonomy is not a fixed setting; it is a judgment that has to track the stakes of the decision itself, reassessed as those stakes change.

That is precisely the problem RAIGHT.ai was built to help organizations solve. Calibrating autonomy to risk only works if a company can first see, classify and document what each of its AI systems is actually doing, and can point to a named, accountable owner for every consequential decision an agent takes. Governance that lives only in a policy document rarely survives contact with a fast-moving agentic system; it has to be built into how that system is scoped, permissioned and monitored from the start.

You can read the full panel discussion, including all the expert responses, on the MIT Sloan Management Review site: Responsible AI Means Knowing the Limits of Agent Autonomy.  https://sloanreview.mit.edu/article/responsible-ai-means-knowing-the-limits-of-agent-autonomy/

Transparency note

This post was drafted with the help of AI, based on a carefully constructed prompt, and reviewed, reformulated and/or rewritten by the author before publishing.