Welcome back. It’s been a big week for everyone working in IT. Today, we’ll break down the shifting tech landscape for IT leaders, along with key moves from Google, Anthropic, and other trending developments you should know this week.
Today: How to talk to a real human instead of a support bot, the silly AI mistake that unmasked an alleged bank hacker, and a vibe-coded bug that cost $10,000.
Today’s Brief
State of the tech industry in 2026 and its IT impact
Anthropic’s newest model is 75% cheaper to run
AI interviewer melts down over gibberish answers
Build an AI-native software factory
THIS WEEK IN ENTERPRISE IT
Click here to watch Google’s Gemini agent in action.
Google drops a one-stop Gemini agent for enterprise work tasks: The search giant just unveiled a “universal” Gemini AI agent designed to run in the background across apps and devices. It works inside Workspace apps and is reachable from Slack and Microsoft 365. Each task routes automatically to whichever Gemini or Claude model fits best, with built-in cost controls. See how it works.
US suspends Microsoft, Adobe, and major IT firms from the green card program for skilled foreign workers: At a White House press conference on Thursday, Vice President JD Vance accused Microsoft of fraud, saying it laid off 6,000 American workers last year while securing 6,300 H-1B visas. Microsoft countered that 80% of its roughly 6,000 H-1B filings last fiscal year went to existing staff. The suspension now blocks those employees from seeking green cards through Microsoft.
Big Tech firms reportedly want their employees to stop using Claude: The Information reports that Microsoft and Meta aim to reduce their reliance on Anthropic’s models. Microsoft has asked its staff to use Copilot more, and The Information estimates that the company’s internal Claude spending will reach at least $1 billion this year. Meanwhile, Meta has cut the number of Claude Code users in half, bringing the total to 30,000, and it has imposed token limits and spending controls. Both companies are shifting AI spending toward tools they own.
YOUR OPINION MATTERS!
This is the second issue of The IT Code. It's a quick 5 min read designed to keep you up to date on the latest IT developments and emerging technologies that could impact your organization. Let us know what you think of this email in the poll at the bottom of this email (or by replying to this email).
We're building this just for you, so your feedback will shape what we cover next!
INSIGHT
The state of the tech industry in 2026 (and what it means for IT)
Click here to watch the entire keynote.
A snapshot of now. Veteran engineer Gergely Orosz, who writes the popular newsletter The Pragmatic Engineer recently gave a keynote to over 2,000 tech leaders in New York. His goal was to capture where the industry stands right now. To prepare, he spent weeks with teams at OpenAI, Anthropic, Uber, and other fast-moving companies. A lot of what he uncovered directly impacts IT:
AI writes the code now: Top engineers at these companies direct AI agents instead of typing code. Migrations that took years now take weeks, so stalled legacy projects deserve a second look.
Reviews can’t keep up: There is more new code than anyone can read, so many engineers approve changes without checking them. Testing and change approvals now carry more of the safety load.
Cloud capacity is running short: CPUs, not just AI chips, are hard to get, and reservations need months of notice. Orosz suggests heavy users lock in fixed capacity now, even at a higher price.
AI bills need a plan: Orosz says two moves deliver most of the savings: cheaper open models and sending each task to the model that fits it.
Infrastructure teams are changing: Most larger tech companies now run their own platforms for AI agents, and infrastructure teams increasingly own them. Orosz expects agents in deployments, monitoring, and incident response next.
The bottom line. Software is being built faster than teams can review it, but the core principles remain the same: teams still own their code, and engineers still decide what ships to production. If you’re planning an AI SDLC (software development lifecycle) transformation, here is Atlassian's playbook.
IN THE KNOW
What’s trending on socials and headlines
Meme of the day.
🧩 Support Quest: Getting customer support used to mean waiting on hold. Now it means battling an AI chatbot first, and one user’s workaround to reach a human is wild (76K likes).
🕵️ Amateur Hour: One person allegedly hacked South Korea’s biggest banks with a stack of five AI tools. Then he exposed his identity by asking Claude to write his résumé (1.3M views).
🧾 Runaway Bill: A vibe-coded Cloudflare feature got stuck in an infinite loop, racking up 6 trillion database operations. With no spending cap, it left the developer with a $10,811 invoice.
💸 Price Drop: Anthropic says its newest model costs about 75% less to run than its predecessor. It’s also a big step up in coding, computer use, and knowledge work (6.1M views).
🌀 Fine Print: Not all $200 AI plans provide the same amount of compute. Research shows that Anthropic's top plan delivers more than 5x the value of OpenAI's plan, while Codex users report that GPT-6's reasoning power drops by up to 12x.
🤖 Stress Test: AI interviewers are screening more job candidates. One candidate answered entirely in gibberish, and the bot had a full meltdown (1.3M views).
PRODUCTIVITY
How to take AI projects from pilot to production (by CIO): Many enterprise AI projects stall after the pilot because leaders aren't involved, production infrastructure isn't planned and models aren't managed once deployed. This guide gives IT leaders 5 steps for getting pilots into production.
How to get your AI token bill under control (by AWS): AI costs are rising fast and are hard to trace when people, apps and agents all call models. This guide shows how to track, attribute and optimize AI spend, much as FinOps does for cloud infrastructure.
How to decide whether a workflow needs an AI agent, automation, or a recurring task: Building agents for predictable work adds cost, testing, permission complexity and upkeep without better results. This guide offers a practical way to pick the right option in Microsoft 365.
What did you think of today's newsletter?
You can also reply directly to this email if you have suggestions, feedback, or questions.
Until next time — The IT Code team