Notes on agents, in practice.
Long-form guides on AI agents, agentic workflows, and how FMCG operators put them into production. Practical and vendor-neutral, written for people who have to make it work.
- Method · Workflow design
Enterprise agentic workflow design: a practical methodology from business problem to production
Agent projects rarely fail because the model is not intelligent enough — they fail because the workflow was never designed. This capstone guide walks the complete method: decide whether work should be agentic at all, allocate it among humans, software, and agents, architect the system, install controls, pilot in stages, and scale by shared capabilities.
By Misagh Akhondzad/17 min read - Guide · Multi-agent systems
Multi-agent systems explained: when multiple AI agents are better than one
Adding agents does not create intelligence — it creates a coordination problem. This guide covers when multiple agents genuinely beat one, the full pattern catalog, structured communication, cross-agent trust and security, how agent teams fail, budget-matched evaluation against a single-agent baseline, and the COORDINATE framework.
By Misagh Akhondzad/18 min read - Guide · Evaluation
Evaluation and observability for AI agents: how to know whether they actually work
A convincing final answer does not prove the agent completed the task. This guide covers the five evaluation layers, trajectory and outcome grading, calibrated judges, repeated-trial reliability, agent tracing, production drift, and the OBSERVE framework for turning failures into compounding regression suites.
By Misagh Akhondzad/17 min read - Guide · Agent security
Guardrails, identity, permissions, and human approval for AI agents
Once an AI system can act, safety is no longer about what the model says — it is about what the system is allowed to know, decide, and do. This guide covers the full control architecture: guardrails, identity, authorization, human approval patterns that mean something, safe failure, and the CONTROL framework for governed agency.
By Misagh Akhondzad/18 min read - Guide · Retrieval
RAG and agentic RAG explained: how AI agents find, evaluate, and use external knowledge
Traditional RAG treats retrieval as a pipeline stage; agentic RAG treats it as an investigation. This guide covers the complete pipeline from chunking to citations, the shift to planned, routed, iterative retrieval, evidence sufficiency and abstention, retrieval security, and the RETRIEVE framework for governed evidence acquisition.
By Misagh Akhondzad/19 min read - Guide · Agent planning
Planning patterns for AI agents: ReAct, planner–executor, orchestrator–worker, and beyond
There is no single best planning architecture for every agent. This guide maps the full pattern catalog — workflows, ReAct, plan-and-execute, ReWOO, orchestrator–worker, evaluator–optimizer, tree search, classical planners — with the trade-offs, failure modes, and the PLANWISE framework for choosing the simplest structure that reliably reaches the goal.
By Misagh Akhondzad/18 min read - Guide · Agent memory
Memory in agentic AI: what agents should remember, forget, retrieve, and verify
Memory turns a stateless model into an agent situated in time — and creates persistence of error. This guide covers what agents should remember and forget, memory as a stored claim rather than a fact, retrieval beyond semantic similarity, poisoning and false memories, and the REMEMBER framework for governed memory.
By Misagh Akhondzad/20 min read - Guide · Tool design
Tool design for AI agents: how to give models the right hands
Tools are the hands of an AI agent: the model decides what should happen, but the tools determine what the agent can actually do. This guide covers the anatomy of a good tool, the risk ladder from read to irreversible execution, idempotency, approval gates, failure modes, and the TOOLKIT design framework.
By Misagh Akhondzad/20 min read - Guide · Context engineering
Context engineering for AI agents: how to decide what the model sees, remembers, and uses
A language model can only reason with what it can see. This guide walks through the discipline of building the model's working environment: the context stack, source authority, failure modes from stale data to prompt injection, twenty practical patterns, and a maturity model for enterprise context.
By Misagh Akhondzad/24 min read - Guide · Agent architecture
The AI agent harness explained: how models become reliable, enterprise-ready agents
The model provides intelligence. The harness turns that intelligence into a dependable system. This guide walks through every layer of a production agent harness, what happens when it is weak, a six-level maturity model, and how to design your first one.
By Misagh Akhondzad/21 min read - Explainer · Agent patterns
ReAct explained: how AI agents reason, act, observe, and adapt
ReAct turned language models from answer generators into systems that investigate, act, observe, and adapt. This explainer walks through the loop in plain English, from its research origins to guarded enterprise designs, failure modes, and a promotion-to-profit worked trajectory.
By Misagh Akhondzad/18 min read - Guide · Human-AI collaboration
Scaffolding Agentic AI: the master guide to human-AI collaboration, copilots, mixed initiative, and progressive autonomy
The most important question in agentic AI is not how autonomous the AI should become, but what relationship should exist between the human, the AI, and the work. This master guide maps that field, from the 1976 origins of scaffolding to a design framework for enterprise agents.
By Misagh Akhondzad/25 min read - Explainer · Foundations
Attention Is All You Need, explained simply: how the Transformer changed AI and enabled AI agents
In 2017, a 15-page paper about machine translation quietly became the foundation of modern AI. This explainer walks through the Transformer in plain English, from attention and multi-head heads to why the same idea now powers AI agents.
By Misagh Akhondzad/19 min read - Guide · Foundations
AI Agents vs. Agentic Workflows: a complete guide to models, tools, instructions, and use-case selection
Almost every AI automation is now called an agent, and the confusion is costly. This guide sets the foundation: what agents and agentic workflows really are, how the pieces fit, and how to choose the simplest architecture that delivers the outcome safely.
By Misagh Akhondzad/20 min read