Today's Author

Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.

Claude Opus 4.8 just dropped, and it's a meaningful step up from 4.7. Same price, but noticeably smarter about knowing what it doesn't know.

The big story here is autonomous work. Opus 4.8 is more likely to flag uncertainties and less likely to make unsupported claims. That matters a lot when you hand off a long task and walk away.

Here's what's new and useful:

• Dynamic Workflows (research preview): you can ask Claude to plan the work and run hundreds of parallel subagents in a single session, then verify outputs before returning them. Think: migrating a massive codebase hands-free.
• Fast Mode is now 2.5x the speed and three times cheaper than before. Turn it on with /fast in Claude Code.
• Effort control: you can now choose the specific effort level for a response, with higher effort translating to deeper thinking.
• You can update Claude's instructions mid-task without breaking the prompt cache.

Opus 4.8 supports a 1M token context window and 128k max output tokens. It's available now on the API, Claude Code, and all major cloud platforms.

You have the ICP. You have the messaging. You just don't have the hours to build the list, find contacts, write sequences, and do it again next week.

Lightfield's outbound agents run on your CRM. They score accounts against your real won deals, draft sequences from your actual customer language, and surface warm intro paths from your team's network.

You set the strategy. The agents build the list, run the sequences, and escalate the replies that need you.

Anthropic just closed a $65 billion Series H round at a $965 billion valuation, making it the most valuable AI startup in the world right now, ahead of OpenAI.

So what's the money actually for? Three things:

• More compute, because Anthropic has struggled to meet demand, forcing usage limits during peak hours
• Safety and interpretability research, meaning figuring out why AI models behave the way they do
• Scaling Claude Code, Cowork, and other products you can build on top of

The growth is real. Run-rate revenue crossed $47 billion this month, driven by enterprise customers deploying Claude deeper into their workflows.

For you as a builder, this matters because Claude is now available across AWS, Google Cloud, and Microsoft Azure, so you can access it wherever you already work. More compute also means fewer rate limit headaches when you ship something that actually gets used.

Tired of tuning XGBoost?

TabPFN-3 beats tuned classical ML, 90% win rate, out of the box.

• #1 across industry standard benchmarks
• from world-class research to production
• drop your dataset, get predictions in seconds

The model handles the complexity so you don't have to.

You know the pain: Claude Code generates a nice HTML report, and then you're stuck copy-pasting feedback back into the chat to tweak it. Back and forth, forever. Someone just fixed that.

A new open-source Claude Code skill turns any HTML page into a live commenting surface, like Google Docs but for local web pages. You highlight text, click an element, drop a note, and Claude edits the page on the spot.

Here's what actually happens under the hood:

• A small script gets injected into your HTML pages that captures your clicks and highlights
• Your comments land in a local inbox file that Claude watches in real time
• The page auto-reloads and walks you through every change Claude made
• The server auto-shuts down after 10 minutes of idle so nothing leaks

To get started: git clone the repo into ~/.claude/skills/, build any HTML page, then just say "make this page interactive." Claude handles the rest. Works for reports, mockups, docs, anything HTML.

How was today's email?

Awesome Decent Not Great

{"AlphaSignal": "214 Barton Springs Rd, Austin, USA"}