AI Agents: How They Are Built
The shift underway is from writing prompts to designing the systems that prompt. Start with what an agent is and is not, then build one — without code or in code — and go deeper into the patterns and frameworks.
13 articles · start with the first and read in order, or jump to what you need
Understanding agents
What an agent is, and how it differs from generative AI, chatbots, copilots and RPA.
- 1 · BeginnerAI Agents vs Agentic AI vs Generative AIAI agents, agentic AI and generative AI, compared side by side. One variable separates them: how much the system decides and does for itself.5 min read
- 2 · BeginnerChatbot, Copilot or Agent: What’s the Difference?Chatbots answer, copilots suggest, agents act. The difference is who takes the action — and that decides the risk, the cost and how you test it.6 min read
- 3 · BeginnerAI Agents vs RPA and Workflow AutomationRPA and workflow tools follow steps written in advance; AI agents choose their own. When each one fits, and why the best systems combine them.6 min read
Building agents
Build an agent without code or in code, and give it tools, MCP connections and memory.
- 4 · IntermediateA Learning Path for Building AI AgentsFive stages, in the order the work actually demands them. Opinionated, deliberately short, and honest about which stages most teams can skip.4 min read
- 5 · BeginnerHow to Build an AI Agent Without CodeThe six building blocks every no-code agent builder shares, where to find them in ChatGPT, Claude, Gemini, Microsoft 365 and Grok, and how to start safely.6 min read
- 6 · IntermediateBuild Your First AI Agent in CodeA working AI agent in about sixty lines of Python, no framework: tools, the loop, and limits enforced in code. Then what to add before production.8 min read
- 7 · IntermediateTool Use, Function Calling and MCP, ExplainedHow models call tools, how to design tools they use well, how the Model Context Protocol standardises the connections, and how to keep tools safe.6 min read
- 8 · IntermediateAI Agent Memory: Short-Term and Long-TermModels remember nothing between requests. How agents get short-term and long-term memory, the three kinds worth storing, and how memory goes wrong.5 min read
Patterns and frameworks
Agent loops, multi-agent graphs, and the frameworks people build them with.
- 9 · IntermediateLoop Engineering: Design the Loop, Not Every PromptA single prompt cannot fix a bug it has never seen run. Loop engineering builds the harness that lets an agent act, observe, reason and retry.5 min read
- 10 · IntermediateAnatomy of a Good Agent LoopFive things every well-built agent loop needs — and the specific failure each one prevents. If it cannot tell done from not-done, it is not a loop.4 min read
- 11 · AdvancedGraph Engineering: Coordinating Specialized AgentsWhen one agent is not enough: who does what, in what order, and what happens when a step fails. Graph engineering is how that gets specified.4 min read
- 12 · IntermediateLoops vs Graphs: How the Two Patterns NestOne agent iterating and many agents passing work are different things at different levels. Holding the distinction shortens architecture debates.3 min read
- 13 · AdvancedAI Agent Frameworks Compared: What Each One Is ForLangGraph, CrewAI, OpenAI Agents SDK, Claude Agent SDK, Google ADK, Microsoft Agent Framework and more — grouped by the problem each solves.9 min read
Common questions: AI Agents
What is the difference between AI agents and agentic AI?
An AI agent is a kind of system: one given a goal, which chooses and takes its own actions toward it. Agentic AI describes a property — how much a system decides and does for itself. It is a spectrum, so a system can be more or less agentic.
What is the difference between a chatbot, a copilot and an AI agent?
The difference is who takes the action. A chatbot answers questions and the person acts on the answer. A copilot suggests inside the tool the person is using, and the person accepts or rejects each suggestion. An agent takes the actions itself, within limits a person sets.
Is an AI agent the same as RPA?
No. RPA and workflow automation follow steps written in advance and are predictable and cheap to run, but cannot handle input nobody wrote a rule for. An AI agent chooses its own steps, which copes with variety but costs more and is harder to test. Strong designs use a fixed workflow with a model only where judgement is needed.
How do I build an AI agent without code?
Use a no-code builder in an assistant or workflow tool and assemble six parts: instructions, knowledge files, tools and connections, skills, triggers, and limits and approvals. Start with a read-only assistant over your own documents, test it on real questions, then add one action at a time behind a confirmation step.
What is MCP in AI?
MCP, the Model Context Protocol, is an open standard for connecting AI applications to tools and data. A capability is wrapped once as an MCP server, and any compatible application can use it. Introduced by Anthropic in 2024, it is now governed by the Agentic AI Foundation under the Linux Foundation.
Do AI agents have memory?
Not on their own — language models keep nothing between requests. Agents get short-term memory from what the application keeps in the context window during a task, and long-term memory from information stored outside the model and loaded back when relevant.
What is an AI agent?
An AI agent is given a goal rather than a single instruction, and chooses and takes its own actions toward it — calling tools, observing the results and adjusting — until a stop condition is met.
What is the difference between generative AI and agentic AI?
Generative AI produces content when asked; every step is chosen by a person. Agentic describes how much a system decides and acts for itself. It is a spectrum, not a separate product category.
Which AI agent framework should I use?
Start from your language and the hard part of your problem. Graph frameworks such as LangGraph, Microsoft Agent Framework and Google ADK suit long-running, stateful work; LlamaIndex and Haystack suit retrieval-heavy work; a model vendor’s own SDK is the thinnest path on that vendor’s models. Many production systems use no framework at all.