AI Agent Engineering

A chatbot replies to messages. An agent pursues a goal: it plans, calls tools, observes results and loops until the task is done or it needs a human, so it can take actions, not just produce text.

You describe tools with names and JSON schemas. The model returns a structured request to call a tool with arguments; your code runs it, sends the result back, and the model continues with that new information.

Retrieval-augmented generation fetches relevant documents at query time and adds them to the prompt. Use it when knowledge changes often or must be cited; fine-tune when you need a consistent style or behaviour, not fresh facts.

An open protocol that standardises how AI applications connect to tools and data. A server exposes tools, resources and prompts; any MCP-compatible client can use them without custom integration code.

Build a dataset of realistic tasks with expected outcomes, score final results and intermediate steps (right tool, right arguments), use code checks where possible and LLM-as-judge where needed, and run it on every change to catch regressions.

Malicious instructions hidden in user input or retrieved content that try to hijack the agent. Defences include treating external content as data, least-privilege tools, allow-lists, output validation and human approval for sensitive actions.

Route easy steps to smaller models, cache prompts and results, trim context, cap loop iterations, batch requests where possible and monitor cost per task with tracing.

Only when tasks are truly separable or need different tools/permissions. Multi-agent systems add latency, cost and failure points, so start with one well-designed agent and split only when evals show a benefit.

Short-term memory is the conversation/state passed each step; long-term memory is stored externally (database or vector store) and retrieved when relevant. Summarise old context to stay within the context window.

Every model call (input, output, tokens, latency, model version), every tool call with arguments and results, errors, user feedback and final outcome — with sensitive data redacted.