Tool Use, Memory and Planning: The Anatomy of an Agent
Strip an AI agent down to its parts and you find a language model surrounded by three things that turn it from a text generator into something that can get work done. Tools let it act on the world. Memory lets it keep track. Planning lets it decide what to do next. Here is what each part does, how they fit together, and what happens when one is missing.
The model at the centre
The language model is the reasoning engine. It reads the situation, works out what is needed, and produces the next step. On its own it can only produce text, which is why the three surrounding parts matter. They are what the text gets turned into.
Tools: the hands
A tool is a function the agent can call: search the web, read a document, query a database, check availability, send an email, run a calculation. The model does not run tools. It writes a structured request naming the tool and its inputs, the surrounding system executes it, and the result returns to the model as text it can read.
Tools are what separate acting from describing. Without them the agent can explain how to book a table. With them it can book one.
The quality of the tools shapes the quality of the agent. Clear names, precise descriptions and compact structured results mean the model picks the right tool and reads its output correctly. Vague tools produce confused agents.
Memory: the notebook
Memory is everything the agent can refer back to. It comes in layers.
Working memory is the current context: the goal, the steps taken so far, the results of each tool call. It is what the model sees right now, and it is limited by the context window, so long tasks need the agent to summarise and drop what no longer matters.
Task memory persists across a task but not beyond it: a scratchpad of findings, a list of what remains to be done, intermediate results too large to keep in context but retrievable when needed.
Long term memory persists across tasks: user preferences, past outcomes, facts learned. It is usually stored outside the model, in a database or a knowledge graph, and retrieved into working memory when relevant.
Without memory, an agent repeats itself, forgets what it already found, and cannot learn that the first approach failed.
Planning: the map
Planning is how the agent decides what to do and in what order. Simple agents plan one step at a time: look at the situation, choose the next action, act, look again. More capable agents sketch a whole sequence up front, then revise it as results arrive. The best ones also know when to stop, when to ask, and when to back out of a path that is not working.
Planning is where the model's reasoning does its heaviest lifting, and it is the part most sensitive to the model's quality. A weak planner picks the wrong tool, loops on a failed step, or declares success too early.
How they fit together
A single turn of an agent runs like this. The model reads its working memory, including the goal and everything so far. It consults its plan and decides the next action. It calls a tool. The result lands in working memory. The model reflects: did that work, does the plan still hold, what is next. Repeat until done.
Each part covers the others' weaknesses. Tools give the plan something to act with. Memory lets the plan build on results. Planning gives the tools a purpose and the memory a shape.
What it means on the receiving end
When an agent interacts with a business, each part shows up. Its tools decide whether it can reach you at all. Its memory decides whether it remembers what it learned about you. Its plan decides whether it chooses you for the step it is on. Being reachable by tools, consistent enough to remember, and clear enough to choose is what it takes to be part of an agent's work rather than skipped by it.
The takeaway
An agent is a model with hands, a notebook and a map. Tools act, memory tracks, planning directs. Together they turn a system that writes sentences into one that completes tasks.