Agentic systems
Agentic
LLM-driven loops that plan, take actions in the world (call tools, edit files, hit APIs), observe results, and iterate — rather than just answering a single prompt. The dominant 2026 paradigm for AI engineering.
An "agentic" system is one where the LLM is in a loop: it picks an action, executes it (via tools, code, or another agent), reads the result, and decides what to do next. Compare to a chat completion, where the model produces one response and the loop is closed.
The agentic shift happened because tool-use got reliable enough that you can string 20+ steps without the agent collapsing. Coding agents (Claude Code, Cursor Agent, Aider), research agents (Deep Research, OpenAI's o-series with tools), and operator-style agents (browse-the-web) are all expressions of the same pattern.
Production agentic systems care about: tool reliability, checkpoint/resume, observability (because debugging a 50-step run is hell without traces), cost-per-task budgets, and human-in-the-loop checkpoints for irreversible actions.
See also
-
MCP
Model Context ProtocolOpen standard from Anthropic (Nov 2024) for connecting AI assistants to external tools, data sources, and prompts. Think "USB for LLM tools" — one protocol, many servers, any client.
-
Tool use
Tool use / function callingThe LLM emits a structured request to call an external function (search, calculator, API), the host runs it, the result goes back in the next turn. Foundation of every agent worth shipping.
-
Eval
EvaluationSystematic measurement of LLM/agent quality — accuracy, hallucination rate, latency, cost. The discipline you wish you'd started 6 months earlier. Without it, you're shipping vibes.