"Agents, unlike workflows, dynamically create their own plan to fulfill a task - they select their tools, pick execution paths, and control how they accomplish tasks. Unlike workflows, an agent has a built-in capacity to figure out how best to accomplish a task without predefined static implementation. That means the agent can decide on the fly when to perform a calculation, consult a database, or otherwise adapt its plan." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"An agent is a program powered by LLMs that can independently make decisions, plan iteratively, and execute tasks to achieve complex goals." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"An agent’s memory draws from multiple sources: the native knowledge encoded in its LLM weights, the transient information provided in its immediate context, and external repositories accessed through retrieval techniques such as retrieval-augmented generation (RAG). Together, these form a dynamic hierarchy of recall, reasoning, and adaptation that defines how an agent perceives, interprets, and acts in the world." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"An agentic mesh is an interconnected ecosystem that makes it easy for agents to find each other, collaborate, interact, and transact." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"As agent autonomy and sophistication grow, we believe that agents need to become enterprise grade. They must integrate easily into an enterprise’s technology and application landscape. They must adhere to enterprise processes - DevSecOps and MLOps, for example - that provide the rigor needed to move mission-critical applications, and soon agents, into production. They must adhere to the expectations of all enterprise applications to become discoverable, observable, operable, secure, and trustworthy." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"Context engineering is the practice of selecting, structuring, and delivering the right information to an agent’s LLM at the right time. Because LLMs do not “know” your current situation beyond the text you provide and the tokens they can access, performance depends heavily on what context you place in the prompt: instructions, constraints, facts, prior steps, and goals. Good context engineering turns a general model into a task-competent assistant by shaping what it sees and how it should reason." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"For agents, principles are especially important because agents can act autonomously and operate with minimal human oversight and often influence critical business outcomes. Well-defined agent principles steer agents toward outcomes that reflect organizational values, regulatory requirements, and ethical norms. By embedding these principles into agent design, we - society, organizations, developers - can understand and then manage risk, build stakeholder trust, and ensure that AI-driven processes remain aligned with strategic objectives." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"In an agentic mesh, agents are the core participants, designed with governance, interoperability, and trust so they can collaborate, interact, and even transact in a broader ecosystem of agents. The key distinction, however, lies between the needs of an individual agent and those of the larger ecosystem. Ecosystems exist to enable collaboration at scale, raising questions of how thousands of agents, each operating independently, can plan, execute, and deliver consistent outcomes. These are the challenges that agentic mesh is designed to address." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"LLMs enable an agent to interact with people and other agents using natural language. LLMs are designed to interpret and convert human inputs into data that can be used to plan and execute complex operations. By translating the words that humans type or speak into usable information, LLMs give agents the ability to interact with people, to reason, and to plan and execute all sorts of tasks." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
"The concept of explainability in autonomous agents is rooted in the need to understand how agents reach their decisions. Given that LLMs underpin many agent functions, their inherent nondeterminism makes it critical to have a system that clearly outlines each operational step. Briefly, nondeterminism for our purposes refers to the variability in the output of systems like LLMs, meaning that identical inputs can yield different outputs upon each execution. This variability arises from probabilistic decision-making processes embedded within these models, where multiple plausible responses exist for a given prompt." (Eric Broda & Davis Broda,"Agentic Mesh: The GenAI-Powered Autonomous Agent Ecosystem", 2026)
No comments:
Post a Comment