A DotNet developer’s note on Technology Radar V33

As a new Technology Radar release with many interesting techs, a few drew my attention. I picked a few looked interesting for dotnet developer.

Surprised by so many tools, frameworks around LLM application. It is definitely exploding.

Techniques

Continuous Compliance: Automate compliance check to be process incorporated in CI/CD
Curated shared instructions for software teams: Share a proven instructions and prompts for AI coding assistance among teams.
Pre-commit hooks: Use Git hooks to check secret, etc before commit in minimal and risk focused way.
Using GenAI to understand legacy codebases: By giving code and context enough, it’s possible to improve the efficiency of understanding legacy code. Tools like Unblocked is specifically made for this purpose.

AGENTS.md: looks like similar to claude.md in Claude Code.
AI for code migrations: when used in smaller steps of all migration projects, it was more useful.
Self-service UI prototyping with GenAI: Product manager can generate interactive prototypes using Claud Code, Figma Make, Miro AI and v0. This technique can speed up feedback loops.
Structured output from LLMs: which is common nowadays, but try tools like Outlines or, Instructor
TCR (Test && Commit) || Revert

AI-powered UI testing: With the help of playwright mcp, it became more useful.
Anchoring coding agents to a reference application: Provide a reference application as a context when creating a new service.
Context engineering
- Context setup: minimal system prompt, canonical few-shot examples, token efficient tools
- Context management: context summarization, structured note-taking, sub-agent architecture
- Dynamic information retrieval: just-in-time context retrieval.
GenAI for forward engineering: modernizing legacy system through AI-generated descriptions of legacy codebase.
GraphGL as data access pattern for LLMs
LLM as a judge: using LLMs to evaluate the output of another system needs to be used carefully. In particular, human should be monitor the result.
For on-device information retrieval, it’s promising to use sqlite-vec and EmbeddingGemma.
SAIF (Secure AI Framework) can help secure LLMs system.
Small language models will be good for simple and repetitive agent workflow.
Spec-driven development: A good example is GitHub’s spec-ket
Team of coding agents: e.g. Claude Code’s sub-agents.
Toxic flow analysis for AI will be useful to get it secure AI agents or MCP.

Arm in the cloud: Arm based cloud instances are proved as cost and energy efficient. It was relatively simple to change x86 based application to deploy Arm based instance.

Model Context Protocol (MCP): Very useful, but be aware of security risk. Use MCP-Scan to scan MCP.
n8n: Workflow tool with rich features including AI agents. Developer friendly. I personally and our team heavily uses it and love it. Very flexible and easy to use.

Agent2Agent (A2A): a protocol for agent to communicate with other agents.
OpenFeature: OpenFeature is an open specification that provides a vendor-agnostic, community-driven API for feature flagging that works with your favorite feature flag management tool or in-house solution.
Restate: a lightweight runtime to turn AI agents, workflows, and backend services into durable processes.

ClickHouse: an open-source, distributed columnar online analytical processing (OLAP) database for real-time analytics.
NeMO Guardrails: open-source LLM guardrail toolkit from NVIDIA
pNpM: Fast, disk space efficient node.js package manager

Barman: open-source PostgrSQL backup and recovery manager
Claude Code: a favorite among other agent coding tool including GitHub Copilot.
Context7: Helps reduce hallucinations by bring framework docs and examples from official repository.
V0: an AI tool for generating front-end code from a screenshot, Figma design or simple prompt.

Augment Code: Similar to Claude Code but argues that good at large codebase
Docling: Docling simplifies document processing, parsing diverse formats — including advanced PDF understanding — and providing seamless integrations with the gen AI ecosystem.
MCP-Scan: is a security scanner for MCP
Serena: the agent no longer needs to read entire files, perform grep-like searches

Fastify: a fast web framework for Node.js that
LangGraph: an orchestration framework designed to build stateful multi-agent applications using LLMs
vLLM: a fast and easy-to-use library for LLM inference and serving.

FastMCP: Python framework that simplifies MCP implementation process.
MLForecast: Python framework for time series forecasting on high-volume data
Nuxt: “Next.js for Vue.js”
Presidio: It provides fast identification and anonymization modules for private entities in text such as credit card numbers, names, locations, social security numbers, bitcoin wallets, US phone numbers, financial data and more. It’s open-source by Microsoft.

Browser Use: an open-source python library that enables LLM-based AI agents to use web browsers and access web applications. It leverages Playwright.
Drizzle: a lightweight TypeScript ORM