Search, Calculator, Calendar

If you could give an AI agent only three tools, a strong case exists for search, calculator, and calendar. Together they cover the three most common ways an agent falls short on its own: it does not know what is happening right now, it is unreliable at precise math, and it has no sense of time. Understanding each tool clearly — what it does, what it does not do, and when an agent should reach for it — is a foundational skill for working with agents.

The Search Tool: A Window on the Present

A search tool connects the agent to live information — web pages, databases, or internal documents. The agent passes in a query string (the thing it wants to find), and the tool returns a set of results: page titles, snippets of text, URLs, or full document contents, depending on how the tool is built. The key word is live. A search tool can return information published five minutes ago. It bridges the gap between the model's frozen training knowledge and the world as it exists right now. When a user asks an agent to find the latest news about a topic, compare prices across three stores, or pull up a company's current job listings, those are all search tasks. But search is not perfect. Results can be wrong, outdated, or irrelevant. The agent must read the returned content carefully and decide whether to trust it, ask for clarification, or search again with a better query.

The Calculator Tool: Exact Arithmetic Every Time

Language models generate the next likely token based on statistical patterns. When the input is a math problem, the model outputs what a correct answer looks like — and it is often right for simple problems. But for multi-step arithmetic, large numbers, percentages, or currency conversions, models make errors that look confident. A student who relies on an AI to compute 1,847 times 293 without a calculator tool has a real chance of getting a wrong answer stated with certainty. A calculator tool (or a code execution tool that runs a small snippet of math) returns the exact result. There is no ambiguity, no estimation, no statistical noise. The agent sends the expression; the tool evaluates it; the verified number comes back. This is also why agents that handle financial, scientific, or engineering tasks almost always have access to a code execution environment. Precision matters, and language model guessing is not precision.

The Calendar Tool: Anchoring the Agent in Time

A language model has no built-in clock. It does not know what day it is, what your schedule looks like, or whether a date in the future falls on a weekend. A calendar tool solves all three of these at once. At its simplest, a calendar tool provides the current date and time. This alone unlocks a huge range of tasks: calculating how many days until a deadline, figuring out what day of the week a birthday falls on, or deciding whether a store is open right now. More powerful calendar tools read from and write to an actual calendar — checking availability, scheduling events, or sending meeting invitations. These write-capable tools belong to a special category we will examine more closely in a later lesson, because writing to a calendar changes something real in the world.