Re: ANP automating the orchestra*tion

My problem with agentic network protocols is summed up by this "AI Review"
in the Google search "limitations of agentic AI systematic review":

"*Limitations of agentic AI in systematic reviews include a lack of deep
causal reasoning, which can lead to identifying correlations instead of
true cause-and-effect, and potential for hallucinations inherited from the
underlying language models. Other challenges include the "black box" nature
of AI, making transparency and accountability difficult, and the limited
ability to handle complex, multi-stage tasks that require extensive memory
and replanning. Additionally, the rapidly evolving nature of the field, the
lack of standardized architectures, and risks from security vulnerabilities
and misinformation are significant hurdles. *

*Reasoning and reliability*
*Lack of causal understanding: AI agents can identify correlations but
struggle with understanding the true cause-and-effect relationships in
data, potentially leading to flawed conclusions.*

*Hallucinations: The large language models (LLMs) that power agentic AI can
generate incorrect or fabricated information.*

*Limited context and memory: Agentic AI often relies on simple, stateless
cycles and has difficulty maintaining long-term context, making complex,
multi-stage tasks challenging.*

*Methodological heterogeneity: A systematic review of agentic AI studies
finds that variations in evaluation metrics and methodologies make direct
comparisons difficult.*

*Transparency and accountability*
*The "black box" problem: The internal decision-making processes of complex
AI systems are often opaque, making it hard to understand how a conclusion
was reached and who is accountable for errors.*

*Difficulty with accountability: When an autonomous AI makes a mistake, it
is unclear who is responsible—the developer, the user, or the organization.*

*Lack of architectural transparency: Many agentic AI solutions are
proprietary, lacking the transparent, standardized designs needed for
widespread adoption and reliable review.*

*Security and risk*
*Vulnerability to manipulation: Attackers can inject false information into
an AI agent's memory, causing it to make incorrect decisions.*

*Amplified risks: Increased autonomy means AI agents can take actions with
more significant implications, and risks like misinformation and errors are
amplified without continuous human oversight.*

*Security vulnerabilities: Multi-agent systems introduce complex security
challenges, such as chained vulnerabilities where an error in one agent can
cascade and amplify across the system.*

*Implementation and integration*
*Lack of standardization: There is a shortage of standardized APIs and
interoperable designs for multi-agent coordination, hindering the
development and deployment of agentic AI systems.*

*Complexity and integration hurdles: Integrating agentic AI with existing
systems is a significant challenge, and many projects fail due to high
costs and unclear returns on investment.*

*Risk and governance gaps: Many projects suffer from a lack of adequate
risk and governance frameworks, which is a major barrier to successful
implementation*."

The European Union has a GDPR, EU AI Act, Digital Markets Act and Digital
Services Act in place, all of which put technical constraints in place that
will effectively kill such protocols because of how the abovementioned
issues impact these four pieces of legislation.

And complex adaptive systems (of systems) are process oriented and are a
key area of AI in Bio and Life Sciences, chemistry, physics, biology and
environmental science.

Authentication, transparency, accountability,  trustworthiness, safety,
explainability and ethical compliance are crucial for agentic AI and IMHO
and that of the European Union far from resolved and standardized.

Milton Ponson
Rainbow Warriors Core Foundation
CIAMSD Institute-ICT4D Program
+2977459312
PO Box 1154, Oranjestad
Aruba, Dutch Caribbean

On Mon, Nov 10, 2025, 06:57 Paola Di Maio <paola.dimaio@gmail.com> wrote:

> Daniel, on the internal mailing list you shared your AI output
> orchestration method,  plus a whole load of other material
> which looks like food for thought but I have not had the resources to
> process yet
>
> But on a related note, assuing we can automate the orchestration
> somethwere along the line
>  I wonder if you could explain in no more than 10 lines of code
> how does your method relates to ANP if at all
> https://github.com/agent-network-protocol/AgentNetworkProtocol
>
> Thank you
>
> Paola
>