AI on X Boss - Edition #1

Monday, March 31, 2026 | ~5 min read

TL;DR

• Google launched Gemini 3.1 Flash Live and Anthropic shipped Claude Code "auto mode" - both aimed at making AI faster and more hands-off for real work. If you use voice assistants or AI coding tools, these are upgrades worth trying today.
• A major security vulnerability hit npm's most popular library (300M weekly downloads), and AI coding tools are making it worse by running installs without you checking what goes in. Pin your dependencies or risk getting burned.
• Anthropic's own data shows that experienced AI users get better results by iterating WITH the AI, not handing it full control - a big lesson for anyone trying to squeeze more value out of ChatGPT, Claude, or any LLM.

Model Updates & New Tools

Google Gemini 3.1 Flash Live - Voice AI gets real

Google DeepMind CEO Demis Hassabis announced Gemini 3.1 Flash Live, their best voice model yet. Lower latency, better precision, more natural conversations. Available now in the Gemini app and Google AI Studio.

Why you care: If you run a business with any phone-based customer interaction - support, sales, scheduling - voice AI just got significantly better. Test it in the Gemini app. The gap between "robot voice" and "human assistant" is closing fast, and the businesses that adopt voice agents early will handle 2-3x the customer volume without hiring.

Claude Code Auto Mode - AI coding without the babysitting

Anthropic launched "auto mode" for Claude Code, their AI coding tool. Instead of asking your permission for every single action (annoying) or running wild with no guardrails (dangerous), auto mode uses classifiers to make smart approval decisions on its own.

Why you care: If you or your dev team use AI to write code, this is a meaningful productivity boost. Less clicking "approve" every 30 seconds, but with safety nets so it does not delete your database. Alex Albert from Anthropic summed it up: "Goodbye --dangerously-skip-permissions, hello auto mode."

Anthropic hints at a model "dramatically smarter" than Opus 4.6

Matt Shumer flagged that Anthropic has trained a model that is "dramatically" smarter than their current best (Claude Opus 4.6). No release date yet, but the signal is clear: the next jump in AI capability is not far off.

Projects, Repos & AI Products

Anthropic's multi-agent system for software engineering

Anthropic published how they use multiple AI agents working together to handle frontend design and long-running coding tasks. Think of it like a small AI dev team where agents hand off work to each other.

Business angle: This is the direction all AI coding is heading. If you pay developers, the cost of building software is about to drop again. One developer plus multi-agent AI tools will soon do what a small team did a year ago.

Claude for Excel and PowerPoint

Anthropic hosted a webinar on using Claude directly inside Excel and PowerPoint. If you missed it, look for the replay - this is practical, "use it Monday morning" stuff for anyone who lives in spreadsheets or decks.

OpenAI's Stargate data center - steel is going up

Sam Altman shared that the first steel beams are up at OpenAI's Michigan Stargate site, built with Oracle and Related Digital. This is a massive AI infrastructure play.

What it means for you: More compute capacity means AI tools get cheaper and faster over the next 12-18 months. If cost has been holding you back from going all-in on AI, the trend is firmly in your favor.

Case Studies & Actionable Insights

The "argue the opposite" trick (30K+ likes)

Andrej Karpathy (former Tesla AI lead, OpenAI co-founder) shared this gem: He spent 4 hours using an LLM to polish a blog post argument until it was "so convincing." Then he asked the AI to argue the opposite - and it demolished his entire case.

Your takeaway: Before you publish that sales page, send that proposal, or commit to that strategy - ask your AI to argue against it. It is the fastest, cheapest devil's advocate you will ever find. This one trick alone is worth the cost of any AI subscription.

Experienced AI users iterate more, automate less

Anthropic released data from their Economic Index showing that people who have used Claude longer actually give it LESS full autonomy, not more. They iterate carefully, attempt harder tasks, and get better results.

Your takeaway: Stop trying to "set it and forget it" with AI. The power users are treating AI like a sharp junior employee - give clear instructions, review the work, refine, repeat. That collaborative loop is where the real ROI lives.

A physicist got 10x faster with AI

Theoretical physicist Matthew Schwartz found Claude Opus 4.5 to be roughly equivalent to a second-year grad student and used it to accelerate his research 10x.

Your takeaway: You do not need to be a tech expert to get massive value from AI. If a physicist can 10x his workflow, you can probably 3-5x yours in areas like research, writing, analysis, and planning. The key is giving AI enough context about what you actually need.

Conversations & Content

The supply chain attack every business owner should know about

Karpathy flagged a new supply chain attack targeting npm's axios library - used by 300 million projects weekly. The scary part: AI coding tools are running package installs constantly, and most users never check what gets installed. Karpathy himself said he "can't feel like I'm playing russian roulette with each pip install or npm install."

Action item: If you have developers using AI coding tools (Cursor, Claude Code, Copilot), make sure they are pinning dependency versions and reviewing what gets installed. One bad package can compromise your entire system.

Sam Altman on AI as a "research institute in your pocket"

Altman shared the story of Paul, who used ChatGPT and other LLMs to create an mRNA vaccine protocol to save his dog. Altman quoted Paul: "The chatbots empowered me as an individual to act with the power of a research institute."

The bigger point: AI is not just a writing tool or a chatbot. It is a thinking partner that can help you tackle problems you would never have attempted alone. The business owners who internalize this will find opportunities everyone else walks past.

That is a wrap for Edition #1 of AI on X Boss.

Was this useful? Reply and tell me - I read every response.

>

Share AI on X Boss with a friend who wants to stay sharp on AI.

>

-- The AI on X Boss Team