The third part of the series moves a local C# agent from free-form tool calling to deterministic Agent Workflows in Microsoft Agent Framework. A workflow collects deployment status, errors and Qdrant documentation in a fixed order, then hands the evidence to a tool-less agent for analysis, with a human approval gate before sensitive actions.
A tutorial that wires .NET 10 into Pinecone, the managed vector database, using HttpClient and the REST data-plane endpoints instead of the archived C# SDK. It covers options binding, DI registration, upsert and similarity query, plus validation that keeps embeddings consistent.
Microsoft Agent Framework lets .NET developers turn a console app into an AI agent that can call tools with only a few lines of code. The article walks through creating the agent, wiring C# functions as tools, testing it locally, and moving the same code to production on Microsoft Foundry Hosted Agents.
The model's failure on the "strawberry" question isn't a counting problem: it never sees letters, only numeric pieces called tokens. Using Microsoft.ML.Tokenizers 2.0.1 and the gpt-5 tokenizer, the author measures this from C# and shows that the word alone is a single token with ID 101830, so the r's inside it stay out of reach. The piece also covers how token cost varies by language, by numbers, and by random text such as GUIDs, plus how to count tokens in .NET.
FAIRouter, an open-source C# router that picks a language model per request and validates the reply with a judge model, is now published on NuGet as FAI.Router (0.1.0, .NET 10). A few lines of code connect OpenRouter or any OpenAI-compatible endpoint, score answers, collect feedback and train the router's weights.
Nick Cosentino explains how to package a recurring .NET review job as a project-scoped Claude Code skill. The example reviews endpoint changes across validation, authorization, response mapping and persistence, and the article stresses clear limits on what the agent may run. It is a practical starting point for teams that want repeatable AI-assisted reviews of ASP.NET Core code.
Claude Code is an agent that reads a repository, plans changes, edits files and runs dotnet build and dotnet test until the code passes. Mehdi Mohseni's guide shows how a .NET team can set it up with a CLAUDE.md, a PostToolUse build hook, plan mode and permission guardrails. It is worth reading if you want the agent to check its own work.
A developer built a Telegram assistant in n8n that can send email and book meetings, with its decisions guarded by a small ASP.NET Core API on .NET 10. The post lists six checks that run before the model sees a request: chat allowlist, duplicate updates, stale messages, voice length, contact ambiguity and time zones. It is a useful checklist for any bot that triggers side effects.
Martin Pisacreta built a small .NET 8 MCP server where the order cancellation tool requires a separately recorded, single-use, expiring approval instead of trusting the model's confirmation. The article walks through the claim-before-use approval store, the tool contract and the limits of the design, backed by 12 passing tests.
How should you configure Claude Code permissions so a coding agent can build and test a .NET project without gaining database or publish authority? This guide explains the documented rule order, settings scopes, wildcard risks, and boundaries for migrations and deployments. It is worth reading before you run agents unattended on .NET repositories.
This tutorial shows how to grow a single LLM call into a production-ready AI feature on ASP.NET Core. It covers authentication, retrieval-augmented generation, security and failure handling. It is a practical roadmap for .NET developers who want the model to stay a replaceable dependency inside their application.
Auto mode in Claude Code still lets an AI classifier decide which actions are safe. The author shows how to add ask and deny rules to a .NET project's settings file to keep control over EF Core migrations, running the app and database updates. He also explains that a deny rule wins over allow and ask rules, whichever settings file holds it.