Hey there,

Welcome to issue #2 of Vibe Coding & AI Jobs — the weekly newsletter where I break down what's actually happening in the AI job market, the tools reshaping how we build, and the career moves that matter right now.

I'm Guy. I run LLMHire.com (real-time AI job intelligence) and EndOfCoding.com (vibe coding resources). Signal — no noise.

Let's get into it.

📊 This Week's AI Job Market Pulse

Data from LLMHire.com — updated April 13, 2026

Top paying companies: Salesforce AI ($550K), Stripe ($475K), Citadel ($375K), Perplexity ($314K), OpenAI ($301K), Google DeepMind ($300K), Anthropic ($300K), Cursor ($295K).

Notable shift from last week: remote roles dropped from 28% to 9%. Companies are pulling people back to offices — hard. If you want remote AI work, the window is narrowing fast.

👉 Browse all 5,769 AI jobs on LLMHire

🔥 The Story of the Week: OpenClaw and the Agent Arms Race

If you missed it — OpenClaw, an open-source autonomous AI agent framework, hit 302,000 GitHub stars and was all over CNBC. It reached 100K stars in just 2 days.

Why does this matter? Because it triggered a full-blown industry response. Anthropic, Nvidia, Perplexity, and Snowflake are all racing to launch competing or complementary tools for enterprise-grade autonomous agents.

Meanwhile, Anthropic launched Project Glasswing — a collaboration with Amazon, Microsoft, Apple, Google, and Nvidia to test Claude Mythos for defensive cybersecurity. The model has already found thousands of vulnerabilities across major operating systems and browsers.

We're watching the shift from "AI assistants" to "AI agents that actually do things in the real world" happen in real-time. This is the biggest inflection point since ChatGPT launched.

🧠 The Agentic Web Is Getting Its Infrastructure

Cloudflare and GoDaddy just partnered to build infrastructure for controlling how AI agents access websites. Think of it as a permissions layer for bots — site owners can now allow, block, or charge AI crawlers.

New standards being introduced: - Agent Name Service — verified identities for AI agents - Web Bot Auth — permission-based access protocols

This is foundational. The agentic web needs rules, and the companies building the plumbing are going to be incredibly well-positioned.

🛠️ New Tools & Models Dropping

Meta launched Muse Spark — a new multimodal model powering Facebook, Instagram, WhatsApp, and smart glasses. It coordinates multiple sub-agents and offers both fast-response and deep reasoning modes.

Z.ai released GLM-5.1 (MIT License) — an open-source model built for long-horizon autonomous engineering. It can stay aligned on a single task for up to 8 hours and sustain thousands of tool calls. This is a serious contender for agentic workflows.

Microsoft launched 3 new MAI models on Foundry — MAI-Transcribe-1 (starting at $0.36/hr), MAI-Voice-1, and MAI-Image-2. All outperform competitors in speed and efficiency.

The AI Scientist-v2 — an autonomous research system that proposes hypotheses, runs experiments, analyzes data, and writes papers. A fully AI-generated paper was just accepted by a major conference. Let that sink in.

🎯 Hottest Skills by Salary (Updated)

From LLMHire salary data across 625 roles with compensation info:

1. Distributed Training — top of the charts
2. MLOps — operations is the new gold
3. JAX — still the research crowd's weapon
4. RLHF — alignment work = premium pay
5. PyTorch / CUDA — the fundamentals never go out of style
6. Agent Orchestration — the fastest-growing category
7. Transformers / Kubernetes / Python — baseline requirements for serious roles

Agent Orchestration climbing the ranks is the signal. Companies aren't just building models anymore — they need people who can make multiple agents work together reliably.

📚 Free Resource

The Complete Vibe Coding Guide — 22 chapters, 200+ prompts, and a full framework for building with AI.

👉 Free preview at vibecodingebook.com

🤔 One Thing to Think About

The compute wars are heating up. Anthropic more than doubled its capacity to train Opus 4.5, getting close to OpenAI's total. OpenAI plans to pull away in H2 2026, but 2027 will be close.

Here's what this means for you: the models are going to keep getting dramatically better, dramatically fast. The skill that matters most isn't mastering any single model — it's learning how to evaluate, switch between, and orchestrate multiple models for different tasks. The era of "I'm a GPT person" or "I'm a Claude person" is over. The winners are model-agnostic operators.

That's it for issue #2. If this was useful, forward it to someone who'd benefit:

👉 buttondown.com/endofcoding

See you next week, Guy @ EndOfCoding

P.S. — Reply to this email and tell me what you want to see more of. I read every response.