Induced Friction Between AI Agents: A Search for Disruptive Solutions

Let AIs compete against each other. AI collaboration may be overrated, or rather, consensus among AIs may kill innovation.

If we're looking for disruptive and innovative AI solutions to today's problems, the biggest obstacle isn't a lack of analytical capacity, but rather that the most likely responses tend to be the most predictable, a product of their average training, and this is what any consultant would say in a first meeting.

To reach another level, we need something different. Instead of AI debate environments, such as collaborative (1,2) or adversarial debates (3,4), whose results can reveal vulnerabilities and convergence traps, as have already been identified (5), a working environment is proposed where multiple AIs run an obstacle relay race where the obstacles are sequential rejections driven by non-judgmental AI arbiter that does not evaluate quality or creativity, but only rejects outputs and measure semantic drift between successive responses. We call this methodology Sequential Stress Optimization (SSO).

The core idea of SSO is simple: instead of promoting debate between agents, which can end in predictable consensus or unproductive, token-consuming loops (5), the proposal is a sequential relay framework with obstacles. Each AI agent receives the previous result from another agent with a single instruction: beat it by divergence. This is a friction scheme built upon successive, non-qualitative rejections, executed across two distinct phases: Exploring and Polishing.

How does it work in practice?

Let's take a real engineering problem: the bottleneck in the submittals review process.

We pose the problem to the first AI agent. Its response is neither corrected nor discussed; it's picked up by the key actor in this scheme: the AI Arbiter.

The AI ​​Arbiter does not generate ideas or judge content qualitatively. Its rule is procedural: in the first phase, it rejects a percentage of the proposals submitted by AIs, let's say 70%, regardless of what each one says.

The Rejection Threshold:

The 70% rejection rate is an arbitrary benchmark set here to illustrate the procedure. In practice, this threshold is a parameter that should be adjusted based on computational cost and solution quality. It can be adjusted dynamically by the user or adaptively by the AI Arbiter (user-selectable option), if the cost outweighs the benefit of optimization (e.g., when the semantic differential decreases). A detailed analysis of the fixed and adaptive rejection rates and their results is beyond the scope of this publication, as validating them for various scenarios requires a significant compute time and token expenditure; therefore, a detailed statistical analysis is reserved for future publication.

Convergence criteria:

The first response (e.g., "hire more reviewers") is discarded. Once this result is obtained, the AI Arbiter passes it on to a second AI with an explicit instruction: "This was already rejected as obvious; find the root flaw and propose something different".

From this succession of responses passed between AI agents, it is pushed out of its statistical comfort zone, allowing potentially innovative solutions to emerge. For example: replacing sequential review with parallel review across disciplines. In successive iterations, increasingly better and more radical proposals are reached. This is no longer optimization; it's a model shift.

When to stop?

Chained rejection cannot continue indefinitely. The first stopping signal occurs when, after reaching the established rejection rate for the first phase, in this case, 70%, the semantic differential between consecutive solutions approaches a minimum. The AI Arbiter takes this signal to mean that the solution space in that direction has been exhausted. If this semantic minimum proves difficult to reach, the AI Arbiter grants an additional 5% margin of rejections threshold for this phase and treats the identified solution as optimal.

Once the first phase, Exploring, concludes, the AI Arbiter activates the second phase: Polishing. Here the goal is to refine the solution from the first phase; the solution obtained is subjected to a new sequence of rejections, now capped between 20% and 30% of everything generated. After that, the AI Arbiter again identifies the minimum semantic dispersion among the results and extracts the optimal solution from the system. This optimal, tentatively disruptive solution is what the AI Arbiter identifies as the final result.

SSO Flow Chart:

Risk and Operational Feasibility Considerations.

Sequential Stress Optimization carries practical risks. Rejecting 70% of the most probable responses as referenced above pushes agents into low-probability territory, where a wrong answer or a hallucination can be argued with full logical consistency. The result can look entirely correct without being accurate. AI Agents may also learn to disguise familiar answers in complex language to pass the Arbiter's filter without changing the underlying idea, so the first phase rejection output will need human follow-up to check substance.

The two failure modes mentioned above, hallucinations and disguised responses, point to the same gap: the Arbiter measures the semantic distance, not the meaning. To compensate for this, a new actor is added to the SSO scheme (see modified flow chart below): the Dynamic Divergence Refiner (DDR) agent. The Dynamic Divergence Refiner (DDR) performs a parallel qualitative check on each sequence. While the Arbiter measures semantic distance, the DDR verifies whether the new response remains within the problem domain and whether its divergence is substantive rather than merely lexical. When the DDR detects disguised repetition or a logically coherent but impractical answer, it vetoes the handoff and restarts the next agent from the last valid state. If drift persists across consecutive rejections, the DDR rewrites the next agent’s input by explicitly naming the repeated concept to exclude and the neglected variable to pursue.

Modified SSO Flow Chart (SSO+DDR):

The Cost Side:

Repeated rejection cycles in SSO imply repeated token generation, which increases latency, token usage, and computational cost. This is even more pronounced in the SSO-DDR scenario, where these costs increase even further. For this reason, SSO is not considered a default method. It is suitable for cases where the value of a better response justifies the additional cost: strategic decisions, not routine tasks. The SSO+DDR scenario would be reserved exclusively for the pursuit of innovation.

Conclusion:

The SSO as a problem-solving approach is not a multi-agent friction system in the traditional sense. It is not based on debate, consensus building, or adversarial interaction between agents It's an obstacle relay race involving multiple AI agents with different analytical approaches. In this race, induced friction arises from the predisposition of an Arbiter to reject the solutions submitted sequentially by each agent. Therein lies the possibility that by promoting difference, something unexpected may emerge, such as a more creative or disruptive solution.

The diversity in agent training provides the necessary variance in the explored sequential solutions. While SSO is architecturally valid with a single agent with different activated modes, its effectiveness decreases with the homogeneity of the agents involved.

On the other hand, the SSO approach identifies two likely failure modes under friction: hallucinations and disguised responses. Their occurrence depends on the complexity of the problem, and to address them, a modified SSO is conceived, with a more complex architecture that includes the Dynamic Divergence Refiner agent (DDR). This agent assists the arbiter by performing qualitative checks that the arbiter does not. The SSO is affected by the computational cost of sequential rejection. The addition of the DDR agent further increases this computational cost, making it significantly more expensive. Therefore, SSO+DDR is recommended for special cases where the goal is outside-the-box solutions.

The SSO and SS+DDR scenarios are theoretical proposals that require validation in terms of both results and costs, the latter being the main constraint. It is estimated that these tests will begin with two or three agents.

 

 References:

 

1. J. Anglen. “AI Agent Debate Systems: How Multi-Agent Collaboration Improves Decision Making”. 2026. RUH.AI.

2. E. Schepis “Democratic Multi-Agent AI: Debate-Based Consensus”. 2025. Medium.

3. J. Mishra. “When Your AI Needs an Enemy”. 2026. Medium

4. Yao et al. “Peacemaker or Troublemaker”. 2025. ArXiv.

5. Zhang et al. “Stop Overvaluing Multi-Agent Debate”, 2025. ArXiv.

Transparency Statement: The author acknowledges the use of Artificial Intelligence (AI) as an assistive tool during the research, data structuring, and content optimization process. The core concept, final review, and critical analysis remain the sole responsibility of the author.

Spanish version: https://ingconcurrente.blogspot.com/2026/06/la-friccion-inducida-entre-agentes-ia.html

The Vigilant PM and PMO: Maintaining Human Leadership and Corporate Sovereignty in the AI Era

Why AI as a strategic co-pilot cannot replace the frontline context of the Project Manager, and how the PMO must build a framework to protect both data and human teams from cold optimization.

The project management landscape is currently flooded with narratives positioning artificial intelligence as an inevitable replacement for the Project Manager, or at least a substitute for many of their key activities. While maximizing personal productivity through automated scheduling and reporting is valuable, it represents a superficial understanding of project dynamics.

The truth is that deep AI integration introduces risks that an automated system can never resolve: the threat of cold management and the long-term danger of vendor dependency. Surviving this transition requires a powerful reclamation of the PM's unique value as the frontline human filter, backed by a Project Management Office (PMO) that serves as the ultimate assurance layer for corporate sovereignty.

1. The PM as the Frontline Filter: Context Over Optimization

In an AI-assisted environment, the PM's primary responsibility shifts from data management to contextual validation. They stand as the first line of defense between raw AI output and the project team.

The PM Operational Filters:

Vetting Automated Triggers: AI models excel at tracking patterns, such as task completion rates or email response times. However, when the AI detects a deviation and flags a performance or delay risk, it operates without empathy or situational awareness. The PM must catch these alerts before they impact the team, evaluating the human reality behind the metrics.

The Context Filter: If the AI flags a resource as underperforming, the PM applies real-world understanding. Is the team member tackling an undocumented technical hurdle? Are they mentoring a junior developer? By processing the data through a filter of human insight, PM prevents automated metrics from being used to unfairly penalize personnel.

Guarding Team Dynamics: This active filtering is what prevents the PM from sliding into cold optimization, a system that treats people as variables to be adjusted rather than individuals to be understood. The PM uses AI insights as operational indicators to prompt a conversation, rather than absolute verdicts to enforce compliance.

Shielding AI Noise and Historical Bias: As AI generates continuous predictive scenarios and updates, the PM acts as a shield against alert fatigue caused by constant analytical noise. Furthermore, since AI models are trained on historical data that may contain past organizational dysfunction, such as overly aggressive baselines, the PM must filter out these historical references to prevent any ongoing project from being penalized by toxic benchmarks.

The attitude described above demands new soft skills from the project manager: the ability to effectively question data from an authoritative source and to exercise healthy skepticism. It's about reading AI alerts as clues to start a discussion, not as conclusions to execute. The PM applies judgment before action.

2. The PMO as the Assurance Layer: Strategic Governance

While the PMs handle day-to-day human interactions, they may lack the organizational authority to question the recommendations of an internally institutionalized AI or how it manages the data it uses. The PMO must step in to institutionalize and protect the PM's human-centered boundaries.

The PMO Strategic Safeguards:

Establishing Boundaries for Metrics: The PMO creates the formal rules of engagement for technology within the enterprise. It dictates exactly how data harvested by AI can be used, explicitly banning automated metrics from being connected to formal performance evaluations or HR actions without human review.

Backing Human Authority: In data-driven organizations, there is a dangerous tendency for upper management to treat automated forecasts as infallible truth. The PMO acts as an institutional buffer, validating the PM's right to override the AI projections based on qualitative team assessments.

Systemic Compliance Audits: The PMO monitors the overall health of the project ecosystem. If data shows a spike in turnover or friction in a specific department, the PMO investigates whether that project's leadership is relying too heavily on automated recommendations without applying the necessary human filter.

3. Securing Corporate Sovereignty: The Multi-Vendor Blueprint.

True human-centered management cannot exist if the organization itself is trapped by an AI vendor. The PMO must implement an enterprise architecture that treats commercial AI models as interchangeable utilities rather than indispensable partners.

Enterprise architecture for AI:

The Threat of Captive Data: When an enterprise feeds its historical project records, proprietary risk logs, and unique operational methodologies into a closed, single-vendor AI cloud, it risks losing its operational independence. If that AI vendor increases prices significantly or changes their compliance terms, the organization cannot easily walk away because its institutional intelligence is locked in the vendor's ecosystem.

The Vendor-Agnostic AI Layer as a Shield: The Sovereign PMO mandates a strict separation between corporate data and the AI models that process it. By implementing a Vendor-Agnostic AI Layer, such as an Agnostic RAG architecture, enterprise data remains securely isolated internally, routed through independent middleware rather than vendor sites.

The Operational Kill Switch: A primary AI vendor can fail the organization in different ways: a prolonged outage, a sudden change in terms of service, or a security breach. Regardless of the cause, the PMO holds the capability to disconnect that AI model and activate an alternative commercial or open-source model running on its own infrastructure. The frontline PM and their teams continue working without operational interruption, insulated from external vendor AI volatility.

Summary: Minimizing AI risk

This layered approach transforms human-centered management from a corporate aspiration into a predictable, structured governance model.

By defining the PM as the operational filter and the PMO as the strategic guardian, the AI-powered organization achieves the ideal balance. AI becomes a powerful engine for predictive and strategic analytics, integrated into a people-controlled framework. In this strategic AI integration scheme, the PM provides the context, and the PMO ensures the system is secure. Sovereignty over data, people, and work remains where it belongs: within the organization.

Transparency Statement: The author acknowledges the use of Artificial Intelligence (AI) as an assistive tool during the research, data structuring, and content optimization process. The core concept, final review, and critical analysis remain the sole responsibility of the author.

Spanish version: https://ingconcurrente.blogspot.com/2026/06/el-pm-y-la-pmo-vigilantes-como-mantener.html

When AI is a Strategic Copilot: The New Role of the Sovereign PMO

Towards a management model with integrated AI and a PMO acting as vigilant support to guarantee security, avoid vendor dependency, and promote e long-term responsible governance of AI in project management

Project management is changing. It is no longer just about using artificial intelligence to automate tasks. Today, AI can serve as a strategic ally in project management: reading behavioral signals in teams, anticipating tension or burnout before it escalates, or identifying friction in client relationships before it becomes a complaint.

These benefits are clear. However, this advanced use creates dependency. If a project manager makes decisions based on an external AI provider, the organization assumes real risks: unilateral price changes, modifications to terms of service, or supply interruptions due to technical failures, commercial, or regulatory decisions in the provider's jurisdiction.

It is precisely this strategic use of AI that demands a new role for the Project Management Office (PMO). The PMO must act as a Vigilant Support: responsible for ensuring operational continuity, proper governance, corporate security, and data sovereignty against technology providers. This role would then be called: The Vigilant PMO: AI-Sovereign Support for Project Governance.

To achieve this, the PMO relies on three practical pillars:

1. Supervision of predictive management centered on people. 

The premise here is that AI can detect, for example, behavioral anomalies in teams or client relationships, and unseen operational risks. However, the responses and corrective actions must be human, not AI-driven. Instead of waiting for a problem to escalate into a conflict or a team member's departure, the project manager can use AI data to identify tensions before they worsen. Here, the vigilant PMO acts as the guarantor that this human-centric approach remains the standard practice. 

Specifically, if AI detects changes in communication patterns or unusual delays in tasks, the project manager intervenes with concrete actions: adjusting workloads, improving cross-functional communication, or redefining scopes. Technology aims to protect the team and the client, not to control them. The PMO periodically verifies that this process is being followed.

2. Ensuring technological independence.

Relying on a single provider is a risk the PMO must actively manage. The solution is to implement an architecture where the data processed by AI for project management is separated from the AI model processing it. 

In practice, this means critical project intelligence, such as lessons learned, risk logs, and internal policies, is stored in a company-controlled environment, while the layer connecting this data to the AI remains independent. If a current provider changes its terms, raises prices, or experiences downtime, the PMO can seamlessly switch the underlying AI model (for example, to a locally hosted open-source solution) without disrupting the project manager's daily workflow. This can be achieved by implementing a Vendor-Agnostic AI Layer (leveraging an Agnostic RAG architecture). In plain terms, it decouples the organization's data from any single provider, language model (LLM), or database. Ultimately, this technological sovereignty empowers the PMO to swap AI vendors without interrupting project momentum, radically reducing the risk of vendor lock-in.

3. Long-term risk governance as the foundation of responsible AI adoption.

The PMO must apply a pragmatic approach to technology strategy, recognizing that AI has limitations and cannot always explain how it reaches a conclusion. 

Therefore, it establishes clear governance rules: critical decisions regarding scope, budget, or risk are not fully delegated to AI. Resources are allocated not only to implement new tools, but also to back them up, audit their results, and keep the team trained to question automated recommendations. Project stability is prioritized over technological advantage, applying a principle of long-term risk prevention. 

Effective Altruism provides a concrete decision filter here: it asks not just whether a decision produces immediate benefits, but whether it might cause large-scale, long-term harm that is easy to overlook. Applied to AI in project management, this means the PMO must evaluate low-probability, high-impact risks, such as a provider shutdown that short-term efficiency metrics tend to hide.

Conclusion: 

Using AI as a strategic ally requires maturity in project management. The competitive advantage does not lie in having the most advanced tool, but in having a PMO that ensures the responsible use of technology. As vigilant support, the PMO guarantees that AI contributes to management and organizational objectives, protecting both people and the company from unnecessary dependencies, and ensuring that a human-centered approach remains the standard practice. 

 

Transparency Statement: The author acknowledges the use of Artificial Intelligence (AI) as an assistive tool during the research, data structuring, and content optimization process. The core concept, final review, and critical analysis remain the sole responsibility of the author.

Spanish version: https://ingconcurrente.blogspot.com/2026/06/cuando-la-ia-es-copiloto-estrategico-el.html

Artificial Intelligence and Quantum Systems

The convergence of Artificial Intelligence (AI) and Quantum Systems (QS) is currently one of the most active areas of research, with published and verifiable results. This convergence has progressed toward synergy between the two fields, with two distinct directions: first, where Artificial Intelligence as a tool helps control and optimize Quantum Systems, and second, where Quantum Systems as computational tools help Artificial Intelligence advance. Both directions present concrete progress and equally concrete limitations, which this document attempts to describe free from technical jargon or unfounded extrapolations.

I. First approach: Artificial Intelligence (AI) as a tool to help Quantum Systems (QS).

Here we find, among other things, the following:

1. Artificial Intelligence and Quantum Computers (AI+QC)

Experimental quantum systems involve extremely complex and precise physical functionalities. This is stated in general terms to avoid delving into quantum physics, which is beyond the scope of this publication. Consequently, attempting to manually manage these physical functionalities, which are also understood to be highly sensitive to the environment, is highly impractical. This is known, broadly speaking, as the problem of achieving optimal quantum control.

Here, the terms quantum coherence and decoherence (the opposite) become prominent. To simplify, coherence can be understood as the extremely short and sensitive state in which a quantum computing unit, known as a qubit, maintains valid, useful, or coherent information. Therefore, any quantum protocol must be executed within its coherence window, and this is where AI capabilities come into play: AI optimizes operations within the coherence window, assisting in calibration, adaptive control, and real-time error mitigation. In essence, AI acts as a control system that maintains the delicate quantum state necessary for quantum computers to function reliably.

2. Quantum Zeno and Anti-Zeno Effects (1,2)

The quantum Zeno effect is real and well-documented. It is based on the fact that measuring a state in a quantum system at a sufficiently high frequency promotes the suppression of its transitions to other states. Its counterpart, the Anti-Zeno effect, is also verified: there is a regime of intermediate frequencies in which measurements accelerate the transition rate between quantum states, rather than suppressing it.

Current applications of the Zeno effect aim to characterize the noise affecting a qubit and, using the ultrafast capabilities of AI, achieve quantum processor calibrations.

AI intervenes here with algorithms and recurrent neural networks that accelerate all aspects of the measurement process, reducing calibration time from hours to minutes.

II. Second approach: Quantum Systems (QS) as support for advanced Artificial Intelligence (AI).

The relationship between Quantum Computing and Artificial Intelligence is still a highly theoretical scenario, with little experimentation, but very promising, and already has emerging practical applications (3,4). We also find here the emerging field known as Quantum Machine Learning or QML, (5) and it is one of the most profound technological synergies at the moment.

This Artificial Intelligence and Quantum Computing convergence is currently trending towards the following areas:

1. Exponential parallel processing resulting from the quantum properties of superposition and entanglement. This would allow AI models to be trained with volumes of variables and dimensions that would collapse the memory of any classical supercomputer. A quantum AI could evaluate thousands of data scenarios simultaneously.

2. Hyperfast Combinatorial Optimization. Here we find the development of quantum algorithms or optimization algorithms, which would allow significant reductions in AI training times.

3. Hidden Pattern Recognition. Here we are faced with the idea that quantum computers, in theory, would be exceptionally capable of identifying complex correlations, thanks to quantum entanglement, where the state of one particle instantaneously alters another. Because of this, a quantum AI could detect patterns or vulnerabilities in computer systems, with revolutionary applications in cryptography and cybersecurity. This is a matter of great importance and also a potential threat to the security of the entire current computing environment.

4. Hybrid Systems. Here we see intermediate quantum hardware, where classical computers handle the interface and basic processing, and send the heavy load of calculation and interpretation to a quantum coprocessor to solve it in a fraction of the time.

Summary:

The current intersection of AI and quantum physics yields real value in verifiable fields: the optimal control of qubits through reinforcement learning and the calibration of quantum noise using algorithms. In all these cases, AI operates as a classical tool that optimizes parameters acting on quantum systems through conventional physical instruments.

Author’s Note: The content of this publication was developed with input and support from Artificial Intelligence in data acquisition, content structuring, and analysis.

Spanish version: https://ingconcurrente.blogspot.com/2026/06/inteligencia-artifical-y-sistemas.html

References:

1. S. Greenfield, A. Kamal, and others. A unified picture for quantum Zeno and anti-Zeno effects- a review. (2025). Arxiv. Cornell University.

https://arxiv.org/abs/2506.12679

2. A.G. Kofman,  G. Kurizki. Frequent observations accelerate decay: The anti-Zeno effect (2001), Arxiv. Cornell University.

https://arxiv.org/abs/quant-ph/0102002

3. Google. Meet Willow, our state-of-the-art quantum chip (2024).

https://blog.google/technology/research/google-willow-quantum-chip/

4. Google. M. Ivezic. Verifiable Quantum Advantage” on Willow Quantum Chip (2025).

https://postquantum.com/quantum-research/googles-quantum-advantage

5. Y. Du, X. Wang, N.Guo, and others. Quantum Machine Learning: A Hands-on  Tutorial for Machine Learning Practitioners and Researchers (2025). Arxiv, Cornell University.

https://arxiv.org/abs/2502.01146