Welcome to the 293rd edition of Coding Jag brought to you by TestMu AI!👐
This week made one thing clear: the agentic era isn't coming - it's already running the show. Meta cut around 8000 jobs while nearly doubling its AI spend to $135 billion - and it wasn't alone. OpenAI released GPT-5.5, its most agentic model to date. Microsoft took Copilot Agent Mode live across Word, Excel, and PowerPoint. Google Cloud put $750 million behind agentic AI partners at Cloud Next '26. The message from every corner of the industry was the same: agents are here, and they're moving fast.
In this week's edition, discover why LangChain flags that up to 20% of agent-written code references packages that don't exist. Dive into a practitioner's MCP migration guide, Microsoft's agentic DevOps playbook, JetBrains' CI/CD trust gap research, post-GPT-5.5 model rankings, TestMu AI's Kane CLI launch, and an AI state of the union with Simon Willison on dark factories.
Grab your coffee. This one's packed.
News
12 min
testmuai.com
TestMu AI announced Kane CLI on April 28, the first browser automation tool built natively for AI coding agents and developers alike. Installable via npm and skill-ready for Claude Code, Codex CLI, and Gemini CLI, Kane CLI validates rendered UI in a local Chrome browser end-to-end, closing the verification gap that AI-generated code leaves behind.
07 minopenai.com
OpenAI CEO Sam Altman announced GPT-5.5 on April 23, positioning it as the company's most capable model for agentic coding, computer use, and knowledge work. Rolling out to Plus, Pro, Business, and Enterprise users, GPT-5.5 scores 82.7% on Terminal-Bench 2.0, a significant leap over GPT-5.4.
07 minmicrosoft.com
Microsoft moved Copilot's Agent Mode from preview to general availability on April 22. The agent now takes multi-step, app-native actions inside Office files - drafting full documents in Word, building analyses in Excel, and rebuilding decks in PowerPoint - without switching between apps.
07 mingooglecloudpresscorner.com
Announced by Thomas Kurian at Cloud Next '26 on April 22, Google Cloud is funding AI value identification, agentic AI prototyping, agent building, and upskilling across its 120,000-member partner ecosystem. The investment also covers forward-deployed engineers and aims to make the partner network production-ready for agent-first workflows.
09 mincnbc.com
Meta CEO Mark Zuckerberg told staff on April 23 that layoffs, cutting ~10% of the workforce, would begin May 20. At the same time, capital expenditure guidance jumped to $115–135 billion (up from $72B in 2025), signaling one of the most aggressive AI infrastructure bets in the industry's history.
AI
07 minessamamdani.com
Developer Essa Mamdani shares a practitioner's deep dive on migrating to MCP-native architecture after moving AutoBlogging.Pro from custom OpenAI function-call wrappers to MCP, deployment time for new tool integrations dropped from three days to eleven minutes. Covers architecture (hosts, clients, servers), the three core primitives (Resources, Tools, Prompts), TypeScript implementation, and the 2026 roadmap, including stateless HTTP transport and async Tasks.
07 minbuildmvpfast.com
Updated after the GPT-5.5 launch - OpenAI's latest takes the #1 spot for agentic coding (82.7% Terminal-Bench 2.0), Anthropic's Claude Opus 4.7 holds strong for deep reasoning, Kimi K2.6 emerges as the top open-weight challenger, and MiniMax M2.7 is the budget standout. Includes real pricing comparisons and workflow fit analysis.
10 minmindstudio.ai
DeepSeek V4 matches GPT-5.5 and Anthropic's Opus 4.7 on agentic benchmarks at a fraction of the cost. Alibaba's Qwen 3.6 and Moonshot's Kimi K2.6 beat closed models on key agentic coding benchmarks. This MindStudio guide covers which open-weight models actually work in production coding pipelines, where they still fall short, and how to pick the right one for your stack.
Automation
10 mindevblogs.microsoft.com
David Sanchez from Microsoft's Developer Go-To-Market team lays out the strategic playbook: agents don't fix broken practices - they scale them. Covers a 6-dimension audit for DevOps foundations before scaling agent adoption, pipeline transformations for human-agent delivery, and GitHub Copilot Coding Agent patterns. Essential reading for engineering leaders preparing for agent-native CI/CD.
11 minindium.tech
Indium Tech walks through how Microsoft's Playwright MCP + GitHub Copilot Agent Mode creates a generate → run → heal loop inside VS Code. Covers the 3 built-in Playwright agents (Planner, Generator, Healer), how Copilot now auto-configures Playwright MCP with zero setup, and real examples of AI-assisted E2E test creation without writing selectors manually.
07 minblog.jetbrains.com
JetBrains' research (January 2026) reveals AI is heavily used in IDEs but barely trusted in CI/CD pipelines. The article maps a maturity model from "AI not in pipeline" to "agent-driven workflows" and explains why CI/CD is shifting from pure automation to control and validation as AI-generated code volume surges.
Tools
09 minfazm.ai
The most comprehensive April roundup - Anthropic's Claude Code shipped 30+ releases in five weeks, Cursor launched v3.0 with parallel agents, GitHub Copilot CLI hit 1.0.23 with new agent modes, and OpenAI Codex gained Windows sandbox networking and MCP improvements. Plus notable updates from Vercel and Supabase. One single reference for what shipped, what broke, and what to try next.
09 mintechgenyz.com
Cursor 3.0's agent-first interface, GitHub Copilot Workspace for full backend development, and Replit Agent for one-prompt deployments are redefining what "AI coding tool" means. LangChain's 2026 data flags that 15–20% of agent-generated code references fake packages (slop squatting). Also breaks down pricing across tools ($10–50/month) and the growing skill decay concern among junior devs.
Video & Podcast
12 minsimonwillison.net
Simon Willison (creator of Datasette, Django co-creator) joins Lenny Rachitsky to explain how GPT-5.1 and Claude Opus 4.5 crossed the "inflection point" in November 2025, why vibe coding is fine for personal projects but dangerous when shipped to others, and how "dark factories" with zero humans are coming to software development.
12 minyoutube.com
Google Cloud CEO Thomas Kurian's full keynote from April 22 - unveiling the Gemini Enterprise Agent Platform, 8th-gen TPUs (80% better inference per dollar), A2A protocol v1.0 now in production at 150 organizations, managed MCP servers across all Google Cloud services, and Project Mariner (web-browsing agent).
Events
12 minseleniumconf.com
Three days of workshops, keynotes, and track sessions on Selenium 5, WebDriver BiDi, Appium at scale, AI in testing, and accessibility. Srinivasan Sekar, Sai Krishna, and Sri Harsha from TestMu AI are among the speakers, alongside experts from Google, Amazon, Sony PlayStation, and The New York Times.