.NET 11 Preview 3 is now available - Features across the .NET runtime, SDK, libraries, ASP.NET Core, .NET MAUI, C#, EF Core, container images, and more.

Microsoft Agent Framework 1.0 - The merger of Semantic Kernel and AutoGen has now hit version 1.0! Production-ready agent framework for .NET (and Python), with state management, multi-agent orchestration, and first-party MCP support. Shortly after, A2A v1 also shipped for cross-runtime agent communication.

Windows App SDK 2.0 released - The first major version of Windows App SDK since 1.0 back in November 2021.

ASP.NET Core 2.3 end of support announcement - Microsoft announced that ASP.NET Core 2.3 - the only ASP.NET Core version still supported on .NET Framework - reaches end of support on April 13, 2027.

Writing Node.js addons with .NET Native AOT - Drew Noakes walks through building Node.js native addons in C#.

2000th episode of .NET Rocks! - I don't think this podcast needs any introduction. But 2000 episodes! Just wow! A massive congratulations to Carl and Richard for reaching this impressive milestone 🎉

Anthropic launches Claude Mythos Preview and Project Glasswing - Yikes, this was HUGE! This new model was leaked in March, but only officially announced in April. It's been deemed so powerful that Anthropic is only allowing access to certain big companies to give them a chance to fix all the security vulnerabilities it's discovering. They've deemed this initiative: 'Project Glasswing'. There's been a lot of talk online about whether it's genuinely as powerful as they say, or if it's just really clever marketing (especially given they're about to IPO). The evidence is pointing to the former! I'd strongly recommend a listen to this SecurityNow podcast episode where Steve talks about it. Another related link: Claude Mythos finds thousands of zero-days.

Claude Opus 4.7 released - Note that this is not the above-mentioned Mythos, and Anthropic explicitly states that it still trails the unreleased Mythos (link).

OpenAI GPT-5.5 ("Spud") released - The long-rumoured "Spud" finally landed - and as GPT-5.5, not GPT-6. SWE-bench Pro came in at 58.6%, well short of the leaked "high 70s", which OpenAI cite as the reason for keeping the GPT-5 branding. Strong on Terminal-Bench 2.0 (82.7%) and FrontierMath. GPT-5.5 Thinking and GPT-5.5 Pro launched the same day in ChatGPT (paid plans only); API access opened the day after with stricter safeguards.

DeepSeek V4 released - V4-Pro and V4-Flash - Exactly one year after V3's "Sputnik moment", DeepSeek is back. V4-Pro is 1.6T total / 49B active params; V4-Flash is 284B / 13B active. Both have 1M context and dual Thinking / Non-Thinking modes. V4-Pro reportedly trades blows with GPT-5.4 and Opus 4.7 on coding benchmarks at a fraction of the price. Open weights on Hugging Face. MIT Tech Review on why it matters.

Claude Design by Anthropic Labs - Another startup-killer - Anthropic have announced Claude Design, which allows you to design your website/slides/prototypes/etc. with text prompts and by attaching various types of assets/screenshots/links/etc. You can then live-edit its design and then hand off to Claude Code to build it. It also ships with a Canva integration and can export to Canva/PDF/PPTX/HTML.

Anthropic launches Claude for Creative Work - Anthropic released a set of nine connectors letting Claude work directly inside Adobe Creative Cloud, Blender, Autodesk Fusion, Ableton, Splice, Affinity by Canva, Resolume Arena/Wire, and SketchUp.

Claude Code desktop app rebuilt around parallel sessions (and Routines) - Anthropic rebuilt the Claude Code desktop experience around parallel sessions - a sidebar managing multiple concurrent sessions, drag-and-drop workspace layout, integrated terminal, in-app file editor, rebuilt diff viewer, expanded preview pane (HTML/PDFs/local app servers), and a side-chat (Cmd+;) for branching off a running task without polluting the main thread. Same announcement introduced 'Routines' - scheduled automations that run on Anthropic's cloud infrastructure independently of your machine.

Anthropic explains the Claude Code quality drop - I was seeing a lot more negative comments about Claude Code's quality recently on the socials. Anthropic publicly traced this to three engineering missteps: (1) March 4 default reasoning effort dropped from high to medium (reverted April 7), (2) March 26 caching bug that made the model continuously discard its own reasoning history mid-session (fixed April 10 in v2.1.116), (3) April 16 system-prompt tweak capping responses at 25 words between tool calls (reverted four days later). All three were resolved by April 20; usage limits were reset for all subscribers on April 23.

Cursor 3 launches with Agents Window - Cursor's biggest UI rewrite since launch - built around an "Agents Window" for orchestrating many agents in parallel across local, cloud, worktree, and remote SSH environments. Adds Design Mode, Agent Tabs, and seamless handoff between local and cloud agents.

GitHub Copilot CLI had a busy April with two notable releases...

Claude Managed Agents are now in public beta - A hosted agent harness on the Claude Platform - sandboxed code execution, checkpointing, credential management, scoped permissions, and end-to-end tracing all handled by Anthropic. Define tasks, tools, and guardrails; Anthropic runs the infra. A persistent memory feature was added on April 23 so agents can learn across sessions. Worth noting: this is billed at API rates - your Pro or Max subscription credits don't apply here, you'll need a separate API account.

OpenAI Workspace Agents launched - Always-on Codex-powered agents that plug into Slack, Salesforce, Google Drive, Microsoft apps, Notion, Atlassian Rovo, and 60+ other enterprise apps. Available on ChatGPT Business / Enterprise / Edu / Teachers. Explicitly positioned as the successor to custom GPTs - OpenAI is deprecating the custom GPT standard for organisations.

OpenAI launches $100/month ChatGPT Pro tier - New mid-tier built for longer, high-intensity Codex sessions, sitting between the $20 Plus and $200 Pro plans. Includes unlimited GPT-5.4 access and access to GPT-5.4 Pro, with 5x more Codex usage than Plus. A clear move to compete with Claude Max-style "all-you-can-eat" tiers.

OpenAI launches GPT-5.4-Cyber - A new GPT-5.4 variant built for defensive cybersecurity tasks. OpenAI is rolling it out with tiered access controls. Follows directly in the wake of the above-mentioned Mythos/Glasswing news.

ChatGPT Images 2.0 - New image generation model in ChatGPT. Lots of hype around this on the socials, and I must say - the resulting images look pretty impressive!

OpenAI brings its models to AWS Bedrock - Days after OpenAI ended its Microsoft cloud exclusivity, OpenAI's models showed up on AWS - and AWS launched ‘Amazon Bedrock Managed Agents’ as the hosting surface.

Google launches "Skills" in Chrome - reusable Gemini prompts - Save any Gemini prompt as a one-click "Skill" you can invoke from any tab via / or the plus-button. Ships with a library of 50+ pre-built Skills; custom prompts sync across devices via your Google account. Free for all Chrome desktop users on Mac/Windows/ChromeOS.

GitHub launches gh skill for agent-skill management - New gh skill command in GitHub CLI v2.90.0 lets you discover, preview, install, update, and publish agent skills. Supports Copilot, Claude Code, Cursor, Codex, Gemini CLI, and Antigravity. Public preview. GitHub is upfront that installed skills are not verified and may contain prompt injection - always gh skill preview first!

Google releases Gemma 4 open-weight family - Apache 2.0 licensed, runs on everything from Raspberry Pi to Android phones to enterprise on-prem, free for commercial use and modification. Most capable Gemma yet.

Claude Cowork now GA on macOS and Windows - Following the limited preview referenced in edition 034, the standalone Claude Cowork desktop app is now generally available on both macOS and Windows, with expanded analytics, OpenTelemetry support, and role-based access controls for Enterprise. (I'd flagged in edition 032 that I couldn't try Cowork because it was Mac-only - this finally fixes that).

Computer Use lands in Claude Cowork and Claude Code - Pro and Max users can now grant Claude direct screen-control access in both Cowork and Claude Code, on top of the original research preview Computer Use feature.

Anthropic shifts Claude Enterprise to usage-based billing - Enterprise no longer bundles any tokens in the seat fee - flat $200/user/month replaced by $20/user + usage at standard API rates, with pre-committed spend. Reportedly could triple costs for some heavy users. Pairs with the GitHub Copilot usage-based shift below as part of a broader "AI billing is getting real" trend.

GitHub Copilot moving to usage-based billing from June 1 - Big one for Copilot subscribers! GitHub is dropping the request-based subscription model entirely. From June 1, you get a monthly allotment of "AI Credits" (1 credit = $0.01) consumed by token usage across input, output, and cached operations. Pro: 1,000 credits/month; Pro+: 3,900; Business: 1,900/user; Enterprise: 3,900/user. Promotional credits run through August; a billing-preview tool ships in May. Pro/Pro+/student sign-ups paused from April 20 ahead of the change.

"Tokenmaxxing" - A new term that seems to be floating around, where, because some staff’s performance is judged on how many tokens they’re using - they’re then doing everything they can to max out their token use, even if what they’re doing doesn’t necessarily equate to increased value/productivity.

GitHub Stacked PRs and the gh stack CLI - Private preview (waitlist at gh.io/stacksbeta). This is to help avoid having massive PRs that are hard to review and work with. With stacked PRs, you have multiple PRs for different areas of your change, and the tooling manages separate “stacked” PRs that can be all merged in together.

GitHub sunsetting SHA-1 in HTTPS - GitHub is discontinuing SHA-1 in HTTPS for the platform and its CDNs. Affects browser access.

Microsoft Build 2026 registration is open - June 2-3, San Francisco (Fort Mason) plus livestream. ~2,500 in-person seats at $1,099, livestream free.

PowerToys v0.99.0 released - The headline new feature for me is ‘Grab And Move’ - hold Alt and drag a window with the left mouse button to move it, or right mouse button to resize it (Linux KDE users will find this very familiar). I used to use the AutoHotKey KDE_Mover-Sizer script for exactly this, so it's great to see it baked into PowerToys. Note that v0.99.1 was also released at the end of the month with fixes.

Supply chain attacks - April was another rough month for open-source supply chain security, with the late-April week being particularly bad...

Developers are System Thinkers, not Code Monkeys - Do you worry that AI is coming for our jobs? My argument is that our real value has never been typing code, it's been systems thinking - breaking problems down, choosing the right components, and orchestrating them together. AI is just one more component to reason about, and the developers who lean into that framing will thrive.

The Many Use Cases of AI Coding Agents - It's easy to think of AI coding agents as just "code generators", but that really sells them short. In this post I walk through a bunch of ways I've been using them beyond writing app code - running CLI commands, debugging by looking at logs and source simultaneously, knocking out throwaway scripts, hooking into collaboration tools, and more.