iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Jenning Ho built opposition into his AI council because he did not want one reasoning context to frame a task, choose how to challenge its own framing, authorize action, and judge the result. His design assigns distinct review responsibilities to separate roles, then asks a human to weigh the evidence and decide what happens next.
That is Ho’s design rationale and account of experience—not a controlled demonstration that multi-agent systems outperform single-agent workflows. His central lesson is narrower: a proposal should face deliberate challenge, and agreement or disagreement should be evaluated rather than counted.
Why Ho wanted a proposal challenged
In his DEV Community article, Ho describes a risk he observed in self-review: a model may find implementation-level problems while leaving the premise that shaped its solution largely unexamined. If the original interpretation is wrong, polishing the implementation can still miss the actual need.
Ho compares this gap to the difference between development and quality assurance. A developer concentrates on intended behavior and implementation; QA asks what happens when assumptions fail and what the experience is like for the user. His council applies that contrast by assigning different responsibilities to reviewers rather than simply asking several agents to repeat the same review.
What each council role is meant to challenge
Ho’s first council round uses four lenses. Each is intended to examine a different aspect of a proposal:
- System Thinker: considers wider effects on the system over time.
- Critical Reviewer: searches for failure modes, including cases where a proposal seems successful but does not achieve the real goal.
- Simplifier: asks whether the solution adds machinery the problem does not require.
- Alternatives Explorer: develops a materially different direction in case the preferred approach is mistaken.
The point is not that four names guarantee four independent views. Ho argues that responsibilities and contexts matter more than labels: reviewers need distinct work to do, and their objections need to be considered against evidence.
How a change in framing can change the task
Ho illustrates the value of challenging the initial interpretation with an example from his own work. A task appeared to call for improving an unreliable mechanism. After inspecting the source, he says he found that an expected field or wiring had never been populated. The problem therefore looked less like a faulty mechanism and more like a missing connection that needed to be built or restored.
Recommended Free Tools
This is Ho’s account of that project, not an independently verified diagnosis. Its relevance is the distinction it makes: a reviewer should be able to question whether the proposed repair addresses the right problem, not only whether it is implemented well.
How to handle disagreement without turning it into a vote
Ho does not describe the council as a majority-vote system. Several agents agreeing does not automatically outweigh one objection that is well supported, and disagreement by itself is not evidence that a proposal is wrong. The decision depends on what the evidence supports and what is at stake.
He identifies source evidence, impact, severity, reversibility, and human judgment as factors to weigh. A serious objection may matter especially when a decision has substantial impact or is difficult to reverse; a disagreement without supporting reasons should not be treated as decisive merely because it is forceful.
What the synthesizer must not smooth away
Collecting reviews is not enough if synthesis erases the useful tension between them. Ho warns that a summary can turn sharp objections into mild caveats, diminish a minority concern, or present incompatible recommendations as though both can be followed at once.
Free tools Windows power users keep installed
One-click scans. No signup required.
His approach calls for the synthesizer to check whether an objection has actually been resolved, whether a minority view points to a different underlying problem, and whether conflicting recommendations still require a choice. If a trade-off remains, the summary should preserve it for a decision rather than imply consensus.
Where human oversight belongs
For Ho, human involvement is not a claim that people are neutral or always interpret evidence correctly. People can anchor on an early explanation or misread what they see. The human role is to make trade-offs consciously and determine when consequential execution is authorized.
Rank #4
He applies that separation to the workflow itself: the Architect should not freely execute its own proposal, and an executor should not approve its own work. He also notes that a corrective goal can still produce a bad remediation plan; even a sound audit does not make every proposed fix sound.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What opposition can—and cannot—establish
Ho’s article presents a design rationale and personal observations, not controlled performance results or a general empirical comparison of single-agent and multi-agent systems. He also cautions that agents can share training patterns, contextual anchors, and blind spots. Assigning different names does not make them independent institutions.
Opposition therefore acts as a pressure on a proposal, not a guarantee of correctness. As Ho puts it, “Opposition creates pressure. Evidence decides what survives.” The practical aim is to make it harder for one plausible interpretation to become authoritative simply because the workflow gave no one a defined responsibility to challenge it.
Quick Recap
Read Jenning Ho’s article on DEV Community.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
