Model Improvements

Model Changes

• OpenAI: New powerful reasoning models
  • Updated gpt-4.1 model (great for everyday tasks)
  • o3 and o3-pro (great for hard coding problems)
• OpenAI: Price changes
  • Prompt caching can save you a lot of money (also on Azure)
  • Significant price drop for o3 (not (yet?) on Azure)
  • 50% savings for new "Flex Processing" option (not (yet?) on Azure)
• Anthropic: Claude 4
  • Sonnet and Opus
  • In general, more expensive than OpenAI
  • Both models are great for coding
• Gemini?
  • Sorry, I don't use Gemini a lot

Coding Tools and Models

• GitHub Copilot
  • Claude 4 is available in preview
  • New OpenAI o3 models are not (yet?) available
  • Pricing model has changed 🔗
• Cursor
  • All new models are available (and more)
  • Pricing model has changed 🔗 (new o3 pricing applied)
• Windsurf
  • Will be acquired by OpenAI 🔗 - consequences 🔗?
  • New o3 pricing applied

Coding Tool
Improvements

Autonomous Agents

Code
Completion

Edit
Mode

Agent
Mode

Agents in the GitHub World

• GitHub Copilot launched Agent Mode 🔗
  • Now also available in VS 🔗
  • Agentic behavior based on Function Calling
• GitHub launched Coding Agent 🔗
  • For GitHub Enterprise and Pro+ customers
  • Runs on GitHub Actions
• My key takeaways
  • Innovation happens in GitHub
  • No cloud, no AI
  • VSCode is GitHub's/Microsoft's Hero editor for AI

Beyond GitHub

• Cursor launched Background Agents 🔗
  • Ubuntu-based container
  • Get/push code from/to GitHub (other providers will follow)
  • Environment customization is possible
  • Cursor now has C# support (including debugging, note DotRush)
• Agentic CLI Tools from Anthropic and OpenAI
  • Claude Code and OpenAI Codex
  • Good options to agent mode in Copilot/Cursor
  • Great for non-VSCode IDEs
• Google?
  • Sorry, I don't have experience with Firebase Studio
• ⚠️ Agents can burn tokens pretty quickly
  • Anthropic: Max plans (starting at 100$/month)
  • Cursor: Ultra plan (starting at 200$/month)

What About UI and Databases?

• UI Option 1: All agents are multi-modal
  • Paste your design into your agent
  • Let it generate UI code
• UI Option 2: Use specialized UI agent
  • Lovable, v0, bolt, etc.
  • Have good integrations with Figma
  • They 💗 modern web dev frameworks (other than Angular and Blazor)
• New generation of Serverless Databases-as-a-Service offerings
  • Multi-Tenancy built-in from ground up
  • Generous free plans
  • E.g. supabase (Postgres), Neon (Postgres), Turso (SQLite)

API/SDK
Improvements

OpenAI API Improvements

• New Responses API 🔗
  • Chat Completions API will still be supported
  • Assistants API will never come out of Beta ➡️ stop using it
• Advantages of Responses API
  • Built with AI Agents in mind
  • Supports all features, including Function Calling, Background Calling, Web Search, File Search, Image Generation, Code Interpreter, Computer Use

OpenAI .NET SDK

• Current versions
  • Latest final: 2.1.0
  • Latest preview: 2.2.0-beta.4
  • Preview version supports Responses API (since beta.3 2025-03-11)
• My general impression
  • ​Solid for typical use cases
  • Not at the same level as SDKs in Python and TypeScript

XPlat AI Development?

• Similar to mobile development
  • XPlat vs. platform-specific implementations
• Consider reverse proxies for protocol translation
  • E.g. OpenRouter, LiteLLM

.NET 10 💗 SSE 

Model Context Protocol

🤘