Agentic Design
MetalHatsCats builds workflow systems, structured knowledge assets, and AI-ready products for complex work.
Consulting and development repository for agentic delivery and AI architecture. Patterns, bytes, and blueprints for building agentic systems that fit real workflows, review loops, and operating constraints.
Who This Is For
Developers, architects, operators, and delivery teams building agentic systems that need guardrails, structure, and usable workflow fit.
What This Cluster Covers
Agentic patterns, workflow automation, retrieval quality, orchestration, and practical AI delivery methods.
The Agent Loop: Observe → Plan → Act → Verify
Understand the minimal mental model of an AI agent so you can explain it clearly and design it reliably.
REF: agentic_dev_001.jsonTools vs Chat: When an Agent Must Act, Not Just Talk
Learn to clearly distinguish between 'thinking in text' and 'acting on the world', and explain why serious agents must use tools.
REF: agentic_dev_002.jsonOutput Contracts: Why Agents Must Speak JSON
Understand why strict output formats (JSON schemas) are critical for building agents you can trust, debug, and automate.
REF: agentic_dev_003.jsonChunking: How Knowledge Must Be Cut for RAG
Learn how to structure knowledge so an agent can reliably retrieve and use it without confusion or hallucination.
REF: agentic_dev_004.jsonMetadata: Teaching Agents What a Chunk Is About
Understand how metadata turns raw text chunks into navigable, filterable, and trustworthy knowledge for agents.
REF: agentic_dev_005.jsonReranking: Choosing the Right Knowledge After Retrieval
Understand why initial retrieval is not enough and how reranking helps an agent select the most relevant and safe knowledge.
REF: agentic_dev_006.jsonGuardrails: What an Agent Is Never Allowed to Do
Learn how to define hard boundaries so an agent behaves safely, predictably, and does not overstep its authority.
REF: agentic_dev_007.jsonSelf-Check / Critic: Teaching Agents to Verify Themselves
Understand how to add an explicit self-check step so agents catch their own mistakes before users do.
REF: agentic_dev_008.jsonPlan → Execute: Separating Thinking from Doing
Learn why agents must separate planning from execution to stay controllable, debuggable, and safe.
REF: agentic_dev_009.jsonHuman-in-the-Loop: Where Agents Must Stop and Ask
Understand where and why an agent must defer to a human, and how to design clear handoff points.
REF: agentic_dev_010.jsonGolden Set & Evals: How to Know Your Agent Works
Learn how to evaluate agents systematically so improvements do not break existing behavior.
REF: agentic_dev_011.jsonTracing & Observability: Making Agent Behavior Explainable
Understand how to trace, inspect, and explain what an agent did, step by step, in production.
REF: agentic_dev_012.jsonMemory: What Agents Should Remember (and Forget)
Understand different types of agent memory and how to use them without creating confusion, drift, or privacy risks.
REF: agentic_dev_013.jsonPrompt Injection & RAG Defense: How Agents Protect Themselves
Learn how to prevent agents from being manipulated by user input or retrieved content, especially in RAG systems.
REF: agentic_dev_014.jsonCost & Latency Budgeting: Designing Agents That Are Economical
Understand how to design agents with predictable cost and latency, so they are usable at scale and acceptable for business.
REF: agentic_dev_015.jsonVersioning: How Agents and Knowledge Evolve Safely
Learn how to change agents, prompts, and knowledge without breaking existing behavior or trust.
REF: agentic_dev_016.jsonFailure Modes & Fallbacks: What Agents Do When Things Go Wrong
Understand the most common ways agents fail in production and how to design explicit fallback strategies instead of silent breakdowns.
REF: agentic_dev_017.jsonSingle-Agent vs Multi-Agent: When One Brain Is Enough
Understand when a single agent is sufficient and when splitting responsibilities across multiple agents makes systems more reliable and maintainable.
REF: agentic_dev_018.jsonAgent Interfaces & Contracts: How Agents Communicate Safely
Understand how agents should communicate with other agents and systems using strict contracts instead of free text.
REF: agentic_dev_019.jsonOwnership, SLAs & Accountability: Who Is Responsible for the Agent
Understand how to assign clear ownership and service expectations so agents can be operated like real systems, not experiments.
REF: agentic_dev_020.jsonBusiness Value: Where Agents Create Real Impact (and Where They Don’t)
Learn to identify use cases where agents generate measurable business value, and avoid areas where they add complexity without payoff.
REF: agentic_dev_021.jsonFrom Bytes to RAG: Assembling an Agent Knowledge Base
Learn how to turn individual bytes into a coherent RAG knowledge base that agents can reliably use in production.
REF: agentic_dev_022.jsonOwning the Knowledge: Turning Bytes into a Personal Moat
Understand how structured agentic bytes become a long-term personal asset you can explain, reuse, sell, and build products on.
REF: agentic_dev_023.json