Claude Code Just Dropped Memory 2.0

In This Issue

• Claude Code just got permanent memory. Here's what AutoDream actually does (and why it changes everything about long-term sessions)
• $240/year saved by ditching Zapier for n8n -- and Claude built the whole thing in one session
• Google's new realtime voice model just beat GPT Realtime on benchmarks (and it's built specifically for voice and vision agents)

New Video:

Claude Code Just Dropped Memory 2.0

Anthropic quietly released AutoDream, a background feature that runs a sub-agent to consolidate, prune, and reorganize your Claude memory files automatically. Turn it on once and your sessions start feeling sharper instead of cluttered -- like Claude actually remembers you.

The mistake most people make: Most people let their Claude memory files grow into a wall of text and wonder why sessions feel inconsistent over time. Bloated context is the hidden reason your AI feels sharp on day one and fuzzy on day 100.

How this helps you: You'll know exactly how to enable AutoDream and what it does in the background to keep your sessions clean. Cleaner memory means less repetition, better recall, and an AI that builds on what it knows instead of starting from scratch.

Winning Is All We Do

$240/year back in her pocket (and it only took one session)

What happened: Debbie fully migrated off Zapier to a self-hosted n8n setup on a Hostinger VPS, connecting WishList Member webhooks, the Brevo REST API, Airtable, and a custom WordPress endpoint. She cancelled her Zapier subscription and saved $240 a year. Claude helped her build, debug, and document the entire workflow in a single session.

The opportunity for you: Membership site owners are stuck on expensive polling-based tools like Zapier when native webhooks and n8n can do the same work faster and for free. That's a real pain point you can solve and charge for.

This is one of many success stories from AI Automation Society Plus. If you're not a member, you're missing out.

In the News

Google Just Launched a Realtime Voice Model That Beats GPT Realtime

The news: Google launched Gemini 3.1 Flash Live, a new realtime model built specifically for voice and vision agents. After more than a year of improvements to the model, infrastructure, and developer experience, it scored 95.9% on the Big Bench Audio speech reasoning benchmark. That puts it well ahead of GPT Realtime (83.3%) and GPT-Realtime 1.5 (81.1%), and just behind Step-Audio R1.1 at 97.0%.

How it affects automation: If you're building voice agents or AI phone workflows, Gemini 3.1 Flash Live is now one of the strongest options available. The quality and latency improvements mean you can build more reliable voice-first automations without the performance tradeoffs that made realtime models frustrating before.