Welcome back. For years, the AI race has been one thing: bigger numbers, better benchmarks, and louder launches. OpenAI just blinked first. Their latest drop, GPT-5.3 Instant, skips the benchmark flexing entirely and focuses on something quieter.

Also: How to become a world-class Agentic Engineer, 50 practical hacks for AI-assisted engineering and a portable device that detects and jams nearby microphones.

OpenAI tunes its busiest model for smoother conversations: The AI lab shipped GPT-5.3 Instant, an update focused on tone, relevance, and conversational flow rather than raw performance gains. The model is less likely to refuse reasonable questions or add unnecessary warnings before answering, gives better answers when pulling from the web, and makes fewer factual errors (down by up to 26.8%). Developers can access it via the API now as “gpt-5.3-chat-latest”.

Apple refreshes MacBook lineup with M5 chips: The tech giant just launched new MacBook Pro and MacBook Air models, both powered by M5 silicon. The Pro version delivers up to 4x faster LLM prompt processing than its predecessor — a notable gain for teams running AI models locally. The Air doubles base storage to 512GB at $1,099, while the Pro starts at $2,199. You can pre-order here.

Google launches its fastest, cheapest Gemini 3 model yet: The search giant just released Gemini 3.1 Flash-Lite, priced at $0.25/1M input tokens — a fraction of larger model costs. It runs 2.5X faster than 2.5 Flash and comes with adjustable thinking levels, letting teams dial reasoning up or down depending on workload complexity. It's available now via the Gemini API in Google AI Studio and Vertex AI.

Spinach AI is the most accurate AI notetaker for humans, and now also serves your agents!

Give your coding agents the missing context they need by connecting Spinach MCP directly to Claude Code, Cursor, Codex, Devin, Glean and other agents.

Spinach also lets you create tasks and tickets in Linear, Jira, Asana, Monday, ClickUp or Trello and write documentation in Notion, Confluence or Google Docs.

Spinach AI is backed by top investors like Y Combinator, Zoom & Atlassian and trusted by over 500,000 professionals.

Give it a try

Block cuts its workforce. Last week, Block cut roughly half its employees due to AI while posting its best quarter in company history. But this wasn’t another corporate restructuring vaguely attributed to AI. CEO Jack Dorsey documented his reasoning in a post that’s racked up 62M views — and it offers a glimpse into what the future might hold.

The receipts. Block’s business isn’t struggling. It’s thriving. The company just posted 24% quarterly gross profit growth, raised guidance for next year, and can afford generous severance packages. So why the layoffs? Dorsey attributed the decision to an internal AI agent called Goose, which has boosted code output per engineer by 40% and “enabled a new way of working.”

The stock market’s verdict. With these layoffs, Block nearly doubled its revenue per employee from roughly $2.2M to $4.2M — and the stock soared 15+%. The message to every leadership team with stock-based compensation packages is now clear: Replacing headcount with AI gets rewarded.

But it’s a Two-Quarter Toss-up. Block is betting that AI has already crossed the threshold from "promising technology" to "load-bearing infrastructure." Jack Dorsey himself even admitted this decision was a bit preemptive. The real test will come in two quarters, when the smaller organization has to outperform last year’s numbers.

If you use OpenClaw, you've probably seen this before. Your agent says "on it!" but never actually starts. Or it says "done!" when nothing was actually shipped.

By default, agents give confident status updates whether the action worked or not. This means you won't catch failures until you check them manually. Tech founder Cathryn Lavery shared a simple one-line fix. Just add this to your agent's instructions:

When you're using Claude Code, just paste the instructions into the "CLAUDE.md" file in your project root, or use "~/.claude/CLAUDE.md" for global rules. For Cursor, simply add them to your ".cursorrules" file.

Instead of just saying "building," the agent now shows a process ID, the active branch, and the current build step.

50 practical hacks to optimize AI-assisted engineering: In this tutorial, a Meta staff engineer shares 50 practical tips for mastering Claude Code. You’ll learn how to optimize context engineering, automate validation loops, and manage complex projects using slash commands and custom skills. It also explores advanced workflows like running multiple parallel instances for faster development.

Skills-4-SE: A curated list of 180+ useful Claude Skills for Software Engineering and resources for customizing AI for SE workflows.

Impact of AGENTS.md files on the efficiency of AI coding agents: This paper discusses the lack of empirical evidence regarding how repository-level instructions affect AI coding agent efficiency. It reveals that AGENTS.md files reduce median runtime by 28.64% and output tokens by 16.58%, primarily by preventing high-cost, inefficient execution paths.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 200K+ engineers and 100K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team