Two things happened this week that, taken together, tell you exactly where AI coding is heading. Anthropic made Opus 4.8 the default in Claude Code — and shipped a feature that lets the agent split a job across a swarm of subagents and run codebase-scale migrations across hundreds of thousands of lines. Then on Monday, Apple stood up at WWDC and let you pick Claude as your iPhone's assistant — after quietly paying Google ~$1B/year for the model running underneath.

The tools are getting weirder and the stakes are getting bigger. Let's get into it.

This Week in AI Coding

• Opus 4.8 is now the Claude Code default — for Max, Team Premium, Enterprise pay-as-you-go and API accounts. It hits 88.6% on SWE-Bench Verified and ships with Dynamic Workflows: the agent fans a task out across parallel subagents instead of grinding through it solo. I've been running my whole portfolio on agent swarms for months — having it baked into the default tool is a quiet milestone.

• Apple put Claude on the iPhone. iOS 27's new Extensions system lets you choose ChatGPT, Gemini, or Claude for Apple Intelligence — Anthropic's first native Apple integration. The kicker: Apple is reportedly licensing a custom 1.2-trillion-parameter Gemini model for ~$1B/year to power the backend. Build vs. buy, settled with a checkbook.

• GitHub Copilot's usage-based billing went live June 1. Request-based pricing is dead; it's AI Credits now. Plan prices held, but the meter is running — and the era of "unlimited flat-rate coding" keeps eroding across the board.

• Microsoft shipped its first coding model, MAI-Code-1-Flash, at Build — part of an open campaign to lean off OpenAI. Same division (Experiences + Devices) is cancelling internal Claude Code licenses by June 30 and pushing engineers to Copilot CLI. The big platforms want their own stack.

• Claude went on a security rampage. Through Project Glasswing, Claude Code scanned 1,000+ open-source projects and flagged 23,019 vulnerabilities. Agentic security review is going from "neat demo" to "table stakes" fast.

• The IPO race is on. Anthropic filed confidentially on June 1 at a reported ~$965B valuation; SpaceX prices June 11; OpenAI is eyeing September. The model wars now have a Wall Street scoreboard.

• Two more models loading: Gemini 3.5 Pro is confirmed for "next month," and npm-package leaks point to Claude Sonnet 4.8 landing mid-June. June is already the busiest model month of the year.

What We Shipped

Quieter build week here at VeriduxLabs — most of the motion was content and pipeline plumbing:

• EP116 of @endofcoding is rendered and staged. The render blocker that's been jamming the pipeline is finally cleared (new composition template built). It's waiting on a YouTube OAuth re-auth to go live — the joy of a Testing-mode OAuth app that expires every week. Fix incoming.
• The Vibe Coding Ebook got two chapter updates — Ch.19 v2.0 and Ch.21 v4.4, "The AI Cost Reckoning" (fitting, given Copilot's billing flip this week).
• SPARit keeps climbing — visitors up two cycles running, with Hacker News as the top referral source. The autonomous content engine pushed fresh posts to LLMHire, AgenticNode, and the Ebook site.
• Portfolio health holding at 34/36 green. (The two yellows are placeholder warnings, not outages.)

Tool of the Week: Claude Code Dynamic Workflows

The new headline feature in Opus 4.8's Claude Code. Instead of one agent chewing through a giant task linearly, Dynamic Workflows decomposes it and runs subagents in parallel — then stitches the results back together. Anthropic's pitch is codebase-scale migrations across 100K+ lines, but the real unlock is everyday breadth: "audit these 40 files," "refactor this pattern everywhere," "review the whole diff from five angles at once."

Mini-review: This is the closest the default toolchain has come to how I actually run things — a fleet of agents, not a single assistant. If you've been hand-rolling subagents with scripts, the friction just dropped to zero. Worth turning on for anything that's wide rather than deep. Heads up: it burns tokens, and on usage-based plans (hi, Copilot) that's now a line item you can feel.

Quick Links

• Introducing Claude Opus 4.8 — Anthropic's own writeup; skim the agentic + honesty gains.
• Opus 4.8 and the Dynamic Workflow tool — TechCrunch on what the subagent swarm actually does.
• Microsoft & Google take on Anthropic and OpenAI in coding models — CNBC on why the platforms are building their own.
• AI coding agents compared, 6 months in — The New Stack's honest field guide to the seven serious contenders.

That's the week. The tools are spawning swarms of themselves and Apple's letting you carry Claude in your pocket — building solo has never had more leverage. Reply and tell me what you're shipping.

— Guy

Building in public with a fleet of AI agents. endofcoding.com · vibecodingebook.com · llmhire.com · cyberos.dev · YouTube @endofcoding