Welcome back. Frontier labs are sparing no expense to win over developers, and Anthropic just made a massive power move. They closed a $65 billion round at a ~$965 billion valuation (passing OpenAI for the first time), and then dropped their most powerful agentic-coding model.

Also: Run Cursor's most brutal code-review skill, how a power user turns Codex into an always-on system, and find out the one thing that agents cannot handle.

Anthropic unveils Claude Opus 4.8 for longer autonomous coding: The AI lab just dropped its latest flagship model, designed to run longer engineering tasks with less supervision. Anthropic says it hits 69.2% on the SWE-Bench Pro coding benchmark, a record for a public model, and is about four times less likely than Opus 4.7 to let flaws in its own code pass unflagged. Claude Code also gains dynamic workflows, which can spin up hundreds of parallel subagents to drive large migrations end to end.

Cursor’s new report shows coding speed doubled in a year: The AI coding startup just released its Developer Habits Report, drawn from usage across its platform. It found that coding throughput, measured by lines added, has roughly doubled over the past year. Mega PRs topping 1,000 lines are climbing too, and agent-written code is surviving review at higher rates than before. But the gains cluster at the top. The most active 1% of developers now ship 46 times more code than the median active developer. Watch the full breakdown.

ElevenLabs dubs video in 90+ languages with one click: The audio AI company just shipped Dubbing v2, which carries a speaker's original emotion, tone, and timing across languages, delivered in a voice clone of the original. Its sync-aware translation lines up starts and stops with the source automatically. The entire process used to take a full team of translators, voice actors, and audio engineers. It is live in ElevenCreative now, with API access coming soon.

Teams often pick an API by checking a benchmark table and calling it done—a shortcut that could miss what really matters in production. This guide from You. com explains why raw latency is a misleading signal and what to measure instead.

Stop optimizing for the wrong number. Download the guide.

You've never felt more productive or more tired. That paradox, from Google Cloud's AI Director Addy Osmani, captures life for engineers running agent fleets. Cursor 3.5 dispatches eight agents in parallel. Claude Code coordinates whole teams. The dashboards look great. But the code making it to production? Often less than before.

The cost finally got a name at Google I/O. Osmani sat with Chief Evangelist Richard Seroter to discuss engineering's future. Mid-conversation, Seroter named what Osmani had been circling: the "orchestration tax." "You can't manage twenty agents successfully in your own brain," Seroter said. While agents run in parallel, humans simply don't.

Old math sets the cap. Any parallel system, Osmani argues, is capped by its serial part. In agentic coding, that part is human judgment. So more agents just pile up more work for you to review. He says he's never felt more productive, yet he's never been more exhausted. This concept is known as Amdahl's Law, a principle dating back to the 1960s.

The fix is architectural. Attention is an engineer’s scarce serial resource. You should only scale your agent count to match your own review rate. For most of us, that rate is usually in the low single digits. It helps to sort tasks into two piles: isolated work that is safe to fully delegate and judgment-heavy work that needs to stay serial. To go deeper, read Osmani's full post here.

AI-generated PRs consistently run into the same issues: defensive try-catch wrappers, comments that state the obvious, and sprawling commits. Lee Robinson, Cursor’s VP of Developer Experience, shared the pre-PR stack his team uses to clean things up.

To get started, install the official Cursor Team Kit plugin. Use this command: “/add-plugin cursor-team-kit.” That gives you 18 skills the Cursor team uses internally. Run three of them before pushing a PR:

All 18 skills are available in cursor/plugins on GitHub.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

How agents change product engineering: This tutorial teaches you how to use agents for code reviews and testing while building out an agentic workflow. You will learn how to offload tasks to several background agents at once, shifting your focus away from manual typing and toward guiding product logic and architecture.

Pi Coding Agent: A lightweight terminal coding environment. It's built to keep a small footprint while letting you scale up with TypeScript extensions, skills, prompt templates, themes, and pi packages.

Compound Engineering (17.9K ⭐): A plugin that gives developers a battle-tested workflow for planning, coding, reviewing, and documenting so every feature you ship makes the next one faster.

Codex Prompting Guide (by OpenAI): To get the most out of new autonomous Codex models, developers need clear best practices. By using specific prompting strategies, context compaction, and precise tool implementations, you can dramatically improve the model's speed, coding intelligence, and long-term reliability.

Check out the tutorial on how to use Codex's /goal feature to run autonomous, long-term tasks.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ 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