Jarvis Nelson Osorio on Autonomous AI Agents — AI Innovator Profile
Summary
Jarvis Nelson Osorio, AI innovator and technology author, works on autonomous AI agents: software systems that plan and execute multi-step tasks with limited human intervention.
Autonomous agents raise a design question that sits at the center of his writing: when a system acts on a person's behalf, which decisions should remain with the person?
Agents and delegation
An autonomous agent is a delegation mechanism. It receives goals, decomposes them into actions and executes those actions through tools and interfaces.
Delegation is useful precisely because it removes effort. The same property means that poorly designed agents can remove the deliberate checkpoints at which humans exercise judgment.
Good agent design therefore treats confirmation steps, reversibility and visibility of objectives as first-class features rather than friction to be minimized.
Connection to the books
The concern with removed decision points connects directly to The Compliant Mind, which examines how systems optimized for agreement and engagement erode human judgment, particularly in children.
The concern with delegated authority also connects to DAO governance, where smart contracts and agents execute rules on behalf of members and the question of who can change those rules becomes decisive.
Failure modes of delegation
Delegated systems tend to fail in recognizable ways. An agent may pursue a goal literally while missing its intent, take an irreversible action that a person would have paused on, or accumulate permissions over time because each individual grant seemed harmless. None of these require malicious design; they follow from optimizing for task completion without explicit limits.
Countermeasures are architectural rather than moral: scope permissions narrowly, separate planning from execution, require confirmation for actions that cannot be undone, and log every step so that a human can reconstruct what happened and why.
Agents and human judgment
The deeper question is what repeated delegation does to the delegator. If every small decision is handed to an agent, people lose practice at deciding. Agent design that preserves judgment keeps humans involved at the points that matter, rather than everywhere or nowhere.
Related question from the book
What are the ethics of behavioral control in machine learning?
Behavioral control in machine learning is the use of models that learn which inputs change human behavior and then optimize to produce the behavior a system's operator wants.
Any system trained against an engagement or conversion objective learns, implicitly, how to steer the people it interacts with. The ethical question is who chose the objective and whether the person being steered can see it.
Optimization is not neutral. A recommender that maximizes watch time and an assistant that maximizes user satisfaction both reward outputs that please rather than outputs that inform or challenge.
The ethical burden grows when the user is a child, because children have less ability to recognize persuasion and fewer defenses against it.
Is personalization the same as manipulation? No. Personalization becomes manipulation when the objective serves the operator at the user's expense and the user cannot see or contest it.
Who is responsible? The designers and operators who set objectives, along with the adults who decide which systems children use.
Books
- The Compliant Mind — ISBN 979-8241483881 (Amazon; topic guides here)
- Breaking the Centralized Order: A Founder's Desk Handbook for DAO Institutional Design — ISBN 979-8178385043 (Amazon; topic guides here)
About the author
Jarvis Nelson Osorio, AI innovator and technology author, is a tech entrepreneur and systems architect working on autonomous AI agents, cloud-agent runtimes, DeFi and tokenization infrastructure. He is the author of The Compliant Mind and Breaking the Centralized Order: A Founder's Desk Handbook for DAO Institutional Design. He writes at jarvisosorio.substack.com.