Claude

Claude is an AI assistant developed by Anthropic, built for reasoning, analysis, coding, writing, and agentic task execution. Unlike a basic chatbot that responds to one prompt at a time, Claude can carry out multi-step workflows autonomously, use tools, read and write files, browse the web, and execute code. It is trained using Constitutional AI, a technique Anthropic developed to align the model’s behaviour with a set of principles rather than relying solely on human feedback at every step. Claude is available as a consumer product at claude.ai and through an API for developers building production systems.

How Claude works

Claude is a large language model, which means it generates responses by predicting the most useful next token given everything in its context window. Its context window is very large, allowing it to process long documents, codebases, and extended conversations without losing track. Here is how the key parts of the system fit together:

• Model tiers: Claude is offered in three sizes for different price and performance trade-offs. Haiku is the fastest and most cost-efficient, suited for high-volume, latency-sensitive tasks. Sonnet sits in the middle, combining speed with frontier-level intelligence for coding and agent workflows. Opus is the most capable, used for complex reasoning, multi-step agentic work, and tasks requiring the highest accuracy.
• Extended thinking: Claude can be toggled into a reasoning mode where it generates a step-by-step chain of thought before producing its final answer. This improves accuracy on hard problems like maths, logic, and multi-part analysis.
• Tool use: Claude can call external tools defined by the developer, including web search, code execution, file reading, and MCP servers. It decides when to call a tool, what arguments to pass, and how to use the result to continue the task.
• Agentic loop: In agentic mode, Claude executes a sequence of actions, evaluates the result, and continues until the task is complete or it hits a checkpoint requiring human approval. This is the foundation of products like Claude Code and Claude Cowork.
• Constitutional AI: Anthropic trains Claude against a written constitution of principles, producing a model that reasons about its own outputs and declines requests that conflict with its values, rather than relying only on human labellers to flag every harmful case.
• Memory: Claude can store and retrieve information across sessions using the memory tool, letting it recall user preferences, project context, and prior decisions without the user repeating themselves.

What you can build with Claude

• Agentic coding assistant: Use the API or Claude Code to build a system that reads a codebase, plans changes across multiple files, runs tests, and commits fixes automatically, with the developer reviewing the final result rather than guiding each step.
• Document intelligence pipeline: Feed Claude long contracts, research papers, or internal reports through the Files API and have it extract structured data, summarise findings, and flag key clauses with citations pointing to the exact source text.
• Customer support classifier: Pass support tickets to Claude with a category list and have it route each one to the right team, including a rationale, at scale through the Batch API without a human reviewing each ticket.
• Multi-agent research system: Orchestrate several Claude instances where one searches the web, one synthesises sources, and one writes a structured report, all coordinated through the API’s tool use and agentic loop.
• Internal knowledge assistant: Connect Claude to your company’s documentation via MCP or RAG and give employees a chat interface that answers questions with citations from actual internal sources rather than guessing.
• Computer use agent: Deploy Claude with the computer use tool to automate browser-based workflows, fill forms, extract data from web interfaces, and complete tasks that previously required manual human interaction.