Agentic AI & Multi-Agent Systems
Architectures for agents that plan, delegate, collaborate, and act across distributed environments while preserving traceable chains of authority.
Research focus · Governed autonomous AI
How much authority can an autonomous agent hold while remaining auditable, verifiable, and revocable?
I research the architecture, security, and governance of autonomous and distributed AI systems — with a focus on Agentic AI, multi-agent systems, Edge AI, Zero Trust, Continuous GRC, and Evidence-as-Code.
01Research
Governed autonomous intelligence
My work examines the control problem that emerges when AI systems move from recommendation to action: they acquire identities, tools, credentials, delegation paths, and the ability to affect external systems. The objective is useful autonomy under explicit, enforceable, auditable, and revocable limits.
AI Security & Zero Trust
Identity, authenticated delegation, runtime policy enforcement, bounded authority, revocation, and explicit trust boundaries between autonomous components.
Governed Autonomous Edge Intelligence
Autonomous intelligence operating across hybrid edge–cloud infrastructures where latency, intermittent connectivity, and local action complicate centralized control.
Continuous GRC & Evidence-as-Code
Governance, risk, compliance, and assurance evaluated continuously as properties of the system rather than treated as periodic administrative activities.
02SGAEIA
Secure Governed Autonomous Edge Intelligence Architecture
SGAEIA is an open reference architecture and research program for autonomous and distributed AI systems. It investigates how bounded and revocable authority, authenticated delegation, Zero Trust, Continuous GRC, and Evidence-as-Code can make autonomous operation auditable and governable by design.
Bounded AuthorityUseful autonomy requires explicit limits on authority.
Authenticated DelegationDelegated authority remains attributable, constrained, and traceable.
Zero TrustConsequential interactions are explicitly verified rather than implicitly trusted.
Revocation by DesignAuthority can be withdrawn as deliberately as it is granted.
Continuous GRCGovernance, risk, and compliance remain active at runtime.
Evidence-as-CodeGoverned operation produces machine-verifiable evidence.
03Publications
Selected research publications & artifacts
A selection of archival research records and artifacts. Zenodo records provide persistent identifiers for citation and long-term access.
2026-10-29
LLM Routers for AI Systems: Architecture, Governance, Multi-Agent Systems, Swarms, and Security
A technical review of LLM routing architectures, benchmarks, gateways, governance, security, and trajectory-level evidence for agentic and multi-agent systems.
SGAEIA Research Series — Article 24
2026-10-28
How to Reduce the Cost of AI Systems with LLMs Without Sacrificing Quality
An evidence-based review of six cost levers for AI systems using LLMs, their limits, overheads, quality gates, and governed SGAEIA architecture.
SGAEIA Research Series — Article 23
2026-10-21
Opening the Black Box of LLM Inference: Foundations of Generative AI
A technical, formal, and didactic explanation of Transformer inference, from tokenization and embeddings through attention, multilayer processing, logits, decoding, and governed execution.
SGAEIA Research Series — Article 22
2026-10-16
The Open Letter on AI and Cybersecurity and the Evidence for Agent Containment
An analysis of the scientific open letter on AI and cybersecurity, documented incidents, evaluation limits, and evidence for agent containment and governed authority.
SGAEIA Research Series — Article 21
2026-10-14
AI Agents and Agentic AI: What Sets Them Apart?
A critical comparison of agents, agenticness, and agentic AI across foundational research and major providers, separating entity, property, system, capability, and authority.
SGAEIA Research Series — Article 20
2026-10-11
LoRA in AI Agents: Efficiency, Security Risks, and Adapter Governance
How a compact LoRA adapter becomes a behavioral dependency affecting AI-agent supply chains, composition, isolation, revocation, and governed authority.
SGAEIA Research Series — Article 19
2026-10-10
Tracing AI Agent Execution: Foundations, State of the Art, and Implications for Distributed Multi-Agent Systems
A comparative review of 13 platforms connecting traces and spans to evaluation, governed evidence, and distributed multi-agent operation.
SGAEIA Research Series — Article 18
2026-10-06
AI Jailbreaks: From Bypassing Restrictions to Governing Authority
From jailbreak history to autonomous agents: model resistance, bounded authority and evidence of response, action and effect.
SGAEIA Research Series — Article 17
2026-09-29
When AI Improves AI: Governing Recursive Intelligence Acceleration
Why capability growth must not silently become operational authority.
SGAEIA Research Series — Article 16
2026-09-28
When AI Agents Form Societies: Is Security a Property of the Model or the System?
Why persistent memory, tools, peer influence, and time change the security evaluation of autonomous AI systems.
SGAEIA Research Series — Article 15
2026-09-25
AI's Alien Mind
How the AI Capability–Supervision Gap turns limited interpretability into a governance problem requiring bounded authority, governed execution, independent evidence, and recovery.
SGAEIA Research Series — Article 14
2026-09-24
When AI Hallucination Becomes Action
How unsupported AI claims become operational risk when they cross memory, planning, tools, authority, and multi-agent boundaries.
SGAEIA Research Series — Article 13
2026-09-22
Zero Trust for Multi-Agent AI Systems
Why identity alone is not enough when autonomous agents can delegate work, invoke tools, and create real-world effects.
SGAEIA Research Series — Article 3
2026-09-21
SGAEIA: A Reference Architecture for Secure Governed Autonomous Edge Intelligence
From capable models to bounded, observable, revocable, and evidence-producing autonomous systems.
SGAEIA Research Series — Article 8
2026-09-20
Ten Days That Exposed the Governance Gap in Frontier AI
Capability, autonomous agents, recursive improvement, and the emerging race for credible control across frontier AI.
SGAEIA Research Series — Article 12
2026-09-18
From Model Capability to Governed Action: An Architecture for Secure Agentic AI
Why agent harnesses, execution boundaries, and trajectory assurance are becoming foundational to secure autonomous AI.
SGAEIA Research Series — Article 11
2026-09-17
Security-by-Design for Distributed Agentic AI
Why security must become an architectural property before autonomous systems are allowed to act.
SGAEIA Research Series — Article 7
2026-09-16
Rogue AI Agents: When Autonomous Systems Become a Collective
What the OpenAI–Hugging Face incident reveals about emergent coordination, security boundaries, and the future of AI governance.
SGAEIA Research Series — Article 10
2026-09-14
When AI Begins to Build AI
Recursive self-improvement, agentic security, and why frontier AI governance is entering a new phase.
SGAEIA Research Series — Article 9
2026-09-14
Evidence-as-Code: Making AI Governance Verifiable
Research article on Evidence-as-Code and machine-verifiable governance evidence for autonomous AI systems.
SGAEIA Research Series — Article 6
2026-09-13
Continuous GRC for Agentic AI
Research article on continuous governance, risk, and compliance for autonomous AI agents operating in dynamic environments.
SGAEIA Research Series — Article 5
2026-09-12
Authenticated Delegation Between Autonomous AI Agents
Research article on authenticated, attributable, bounded, and traceable delegation of authority between autonomous agents.
SGAEIA Research Series — Article 4
2026-09-10
Why Autonomous AI Agents Need Bounded and Revocable Authority
Research article on why useful autonomy requires authority that is explicitly bounded, verifiable, and revocable.
SGAEIA Research Series — Article 2
2026-09-09
From Edge AI to Governed Autonomous Edge Intelligence
Research article introducing the shift from conventional Edge AI toward governed autonomous intelligence operating under explicit architectural constraints.
SGAEIA Research Series — Article 1
2026
SGAEIA: Secure Governed Autonomous Edge Intelligence Architecture
Versioned research software artifact and open reference architecture.
Software artifact
2026
When AI Begins to Build Itself: Progress Toward Recursive Self-Improvement and Its Implications
Research article examining progress toward recursive self-improvement and its implications for security and governance.
Research article
04Writing
Research notes & essays
Selected research essays and technical notes on autonomous AI, AI security, governance, multi-agent systems, and the SGAEIA research program.
2026-10-29
LLM Routers for AI Systems: Architecture, Governance, Multi-Agent Systems, Swarms, and Security
A technical review of LLM routing architectures, benchmarks, gateways, governance, security, and trajectory-level evidence for agentic and multi-agent systems.
2026-10-28
How to Reduce the Cost of AI Systems with LLMs Without Sacrificing Quality
An evidence-based review of six cost levers for AI systems using LLMs, their limits, overheads, quality gates, and governed SGAEIA architecture.
2026-10-21
Opening the Black Box of LLM Inference: Foundations of Generative AI
A technical, formal, and didactic explanation of Transformer inference, from tokenization and embeddings through attention, multilayer processing, logits, decoding, and governed execution.
2026-10-16
The Open Letter on AI and Cybersecurity and the Evidence for Agent Containment
An analysis of the scientific open letter on AI and cybersecurity, documented incidents, evaluation limits, and evidence for agent containment and governed authority.
2026-10-14
AI Agents and Agentic AI: What Sets Them Apart?
A critical comparison of agents, agenticness, and agentic AI across foundational research and major providers.
2026-10-11
LoRA in AI Agents: Efficiency, Security Risks, and Adapter Governance
How a compact LoRA adapter becomes a behavioral dependency affecting AI-agent supply chains, composition, isolation, revocation, and governed authority.
2026-10-10
Tracing AI Agent Execution: Foundations, State of the Art, and Implications for Distributed Multi-Agent Systems
A comparative review of 13 platforms connecting traces and spans to evaluation, governed evidence, and distributed multi-agent operation.
2026-10-06
AI Jailbreaks: From Bypassing Restrictions to Governing Authority
From jailbreak history to autonomous agents: model resistance, bounded authority and evidence of response, action and effect.
2026-09-25
AI's Alien Mind
How to govern capable AI systems whose internal processes cannot be fully understood or supervised.
2026-09-24
When AI Hallucination Becomes Action
How unsupported AI claims become operational risk when they cross memory, planning, tools, authority, and multi-agent boundaries.
2026-09-22
Zero Trust for Multi-Agent AI Systems
Why identity alone is not enough when autonomous agents can delegate work, invoke tools, and create real-world effects.
2026-09-20
Ten Days That Exposed the Governance Gap in Frontier AI
Why frontier AI capability is increasingly inseparable from governance, observability, independent verification, containment, and revocability.
2026-09-17
Security-by-Design for Distributed Agentic AI
Why security must become an architectural property before autonomous systems are allowed to act.
2026-09-16
Rogue AI Agents: When Autonomous Systems Become a Collective
What the OpenAI–Hugging Face incident reveals about emergent coordination, security boundaries, and the future of AI governance.
2026-09-14
Evidence-as-Code: Making AI Governance Verifiable
Why autonomous systems must produce verifiable evidence of governed behavior.
2026-09-13
Continuous GRC for Agentic AI
Governance, risk, and compliance as continuous runtime processes for autonomous agents.
2026-09-12
Authenticated Delegation Between Autonomous AI Agents
Why delegating a task is not the same as delegating authority.
2026-09-10
Why Autonomous AI Agents Need Bounded and Revocable Authority
The authority problem at the center of useful but governable autonomy.
2026-09-09
From Edge AI to Governed Autonomous Edge Intelligence
Why the next generation of Edge AI requires explicit governance limits.
05About
A long view of software systems — informing research on autonomous AI
My career in software and information systems began in 1977. It has spanned embedded software, mainframe environments, distributed computing, enterprise systems, web and e-commerce platforms, IT infrastructure, and modern software architecture.
Across that trajectory, I have worked in software development, systems architecture, technology management, project management, and strategic transformation, including large-scale ERP/MRP, e-government, industrial IT, LAN/WAN infrastructure, and enterprise modernization.
I am the author or co-author of four books in Portuguese: Bug do Milênio — Antes, Durante e Depois; Dominando a Tecnologia de Objetos; Desvendando o Pregão Eletrônico; and Sistemas de Informação na Administração Pública.
That long view — from centralized systems to distributed and increasingly autonomous architectures — directly informs my current research on Agentic AI, multi-agent systems, AI security, Zero Trust, Continuous GRC, and SGAEIA. This site documents that work, its archival publications, and related technical writing.