Day 23: Tool Use and Function Calling — Connecting LLMs to APIs, Databases and Shells
By default a language model is an isolated reasoning engine. It cannot read the current time,...
Tag archive
By default a language model is an isolated reasoning engine. It cannot read the current time,...

MCP in agentic AI beats plain function calling for teams: one server works across every client. Function calling still wins for solo builds.
After two months porting a customer-facing agent to Microsoft Semantic Kernel and comparing it against LangGraph, I share where SK earns its keep with
Deep dive into how function calling actually works under the hood — JSON Schema to CFG compilation, pushdown automata, logit masking, and why the model literally cannot produce invalid structure.
Building AI Agents with Function Calling: Complete Implementation Guide Build...
AI Agents with Function Calling: Build Tools LLMs Can Use Create AI agents that can...
Panoramica pratica su Muse Spark (managed), Muse Glimmer (open weights), pricing a token e...
LLM tool-use reliability in 2026: how to evaluate models for AI agents Summary. The...
How tool use and function calling actually work in AI agents, and what it takes to build reliable systems that do things in the real world.
Streaming a text response is easy: print tokens as they arrive, order doesn't matter to the reader. Streaming a tool call is not, because the payload is JSON, and partial JSON is not valid JSON. The three ways people handle that mismatch, why two of them break in production, and the pattern that lets you show progress without ever executing on a half-formed argument.
Six patterns for reliable AI agent tool calling in 2026, with a side-by-side Anthropic vs OpenAI function-calling comparison and strict-mode rules.

Structured outputs measured on 12 LLM APIs: 100% schema-valid JSON, wrong values on 4 models with thinking on, same call billed 30 to 4,959 prompt tokens.