- The Shift AI
- Posts
- Anthropic Just Looked Inside Claude's Mind
Anthropic Just Looked Inside Claude's Mind
Plus, 🛒How to recover abandoned carts automatically with Omnisend, Meta says its next frontier model has reached GPT-5.5-level performance, and more!

Welcome back to The Shift. Let’s get straight to what matters in AI today…
Today we have:
🧠 Anthropic Peeks Inside Claude's Mind
🛒How to Recover Abandoned Carts Automatically With Omnisend
🍉 Meta Says Its Next AI Model Has Caught Up
🔨Tools and Shifts you cannot miss
🧠 Anthropic Peeks Inside Claude's Mind
Anthropic published new research showing Claude uses an internal reasoning workspace called "J-space," giving researchers a rare look at how concepts are formed and manipulated before an answer is generated.

The Decode:
Claude Has A Hidden Workspace - Researchers identified J-space, a small internal reasoning area that temporarily stores and manipulates concepts, acting like a shared workspace for Claude's more complex thinking.
Hidden Thoughts Shape Answers - Replacing the internal concept of "spider" with "ant" changed Claude's answer from eight legs to six, showing hidden representations directly influence final responses.
Reasoning Happens Here - When researchers suppressed J-space, Claude could still chat naturally and recall facts, but its ability to complete multi-step reasoning tasks dropped significantly.
A New Window Into AI - Anthropic believes J-space could eventually help researchers inspect hidden concepts, detect manipulation attempts, and better understand how advanced AI systems reason internally.
Understanding how models think is becoming as important as measuring what they can do. Research into internal reasoning systems like J-space could lead to more transparent, reliable, and safer AI models in the future.
Together with Wati
That lead in your CRM? Gone.
Over 3.5 billion people open WhatsApp, Instagram, or Facebook Messenger every day. Your customers are already there, asking questions and comparing options.
Wati puts your business across WhatsApp, Instagram DM, Facebook Messenger, SMS, RCS, and web chat in one AI-powered inbox.
Automations instantly respond, route conversations, and keep every interaction tracked in one place.
Meet customers where they already are, before your competitor does.
🛒How to Recover Abandoned Carts Automatically With Omnisend
70% of shoppers abandon their cart. Here's how to bring them back on autopilot.

Step 1: Connect your store - Link Shopify, WooCommerce, or BigCommerce. Omnisend starts tracking cart activity immediately.
Step 2: Launch the abandoned cart workflow - Select the pre-built Abandoned Cart template in the automation builder. Trigger, timing, and sequence already set. Customize or launch as-is.
Step 3: Set your three touchpoints - Email at 1 hour. Email at 24 hours. SMS at 48 hours. Each message pulls the exact products the shopper left behind automatically.
Step 4: Add a discount to the final message - Drop a time-sensitive offer into the last touchpoint. Omnisend applies it dynamically, no manual coupon management.
Step 5: Let it run - The sequence fires automatically every time someone abandons. You set it once, it works around the clock.
$79 back for every $1 spent across 150,000+ brands. You can set up your abandoned cart flow today.
🍉 Meta Says Its Next AI Model Has Caught Up
Meta's AI chief Alexandr Wang told employees the company's upcoming model, Watermelon, has reached GPT-5.5-level performance as Meta prepares its next major AI release.

The Shift:
A Bigger Model In Training - Watermelon is still being trained using roughly ten times more compute than Muse Spark, Meta's current flagship model released earlier this year.
Coding Upgrade Coming - Meta says the next Muse Spark update will deliver major improvements in coding and agentic capabilities, with the release planned for both Meta AI and its developer API.
Agents Still Moving Slowly - During the same town hall, Mark Zuckerberg said progress on AI agents across the industry has not accelerated as quickly as expected, a comment later clarified by Wang.
Meta Keeps Investing - The company continues expanding its AI efforts through larger infrastructure spending, aggressive hiring, and more powerful models aimed at closing the gap with OpenAI and Anthropic.
Meta is signaling that its AI investments are beginning to translate into more competitive models. If future releases match these internal claims, competition among frontier AI labs could become even tighter over the coming months.
Together with tldr.tech
Don't be the one behind at standup
Your team is already talking about the launch you missed. TLDR is the 5-minute daily brief that keeps you ahead, curated by ex-Google and Anthropic engineers. Free, and read by 7M+ subscribers.
🔨AI Tools for the Shift
🎬 Story Notes – Analyzes your screenplay and delivers detailed feedback on characters, structure, premise, pacing, and themes in minutes.
🛠️ MCP Server by Picsart – Connects Picsart's AI models to your CLI or AI agents, giving developers access to 120+ image, video, audio, and design models.
👕 AI Tech Packs – Creates editable apparel tech packs with specifications, measurements, and bills of materials ready to share with manufacturers.
✍️ Unbotify – Learns your writing style to generate authentic first drafts from notes or humanize AI-generated content.
🎨 Modeinspect – Lets you design and refine product interfaces directly inside your codebase using an AI-powered design canvas.
🚀Quick Shifts
💼 Microsoft is laying off 4,800 employees, including major Xbox and sales teams, as it restructures for the AI era while emphasizing the eliminated roles are not being directly replaced by AI.
🚀 xAI has rebranded its X account to SpaceXAI, reflecting its integration with SpaceX following the acquisition and introducing a new company logo.
🛡️ Reddit is using AI to fight AI-generated spam, with new LLM-powered moderation systems blocking around 25,000 spam posts daily and reducing users’ spam exposure by 20%.
🇨🇳 ByteDance and Alibaba are removing AI companion features from Doubao and Qwen, ahead of new Beijing regulations aimed at limiting emotionally dependent relationships with AI chatbots.
🧠 OpenAI’s GPT-5.6 Ultra is coming to Codex, with the company confirming its highest-end model will power the coding platform as the GPT-5.6 family is expected to launch this week.
🧩 Prompt of the Day
Balance Multiple Subjects Without Falling Behind
Most students study whatever feels urgent and end up neglecting other subjects. A structured rotation keeps every course moving forward while naturally shifting focus as exams get closer.
Study smarter instead of constantly catching up.
Paste the prompt: Drop this into ChatGPT, then fill in your study details.
Prompt to paste
Create a weekly study plan based on my [Insert subjects and exam dates], [Insert current performance in each subject], and [Insert total weekly study hours]. Allocate study time according to subject difficulty, current performance, and exam priority while ensuring no subject is neglected. Recommend a weekly rotation, daily study blocks, review sessions, and practice schedules. Explain how the plan should adapt as exam dates approach by increasing focus on higher-priority subjects without sacrificing long-term retention in the others. Focus on building a balanced, sustainable study system that maximizes performance across all courses.
That’s all for today’s edition. See you tomorrow as we track down and get you all that matters in the daily AI Shift!
If you loved this edition, let us know how much:
How useful was today's edition? |
Forward it to your pal to give them a daily dose of the shift so they can 👇



Reply