⏱ 4-min read
Anthropic just found a hidden "workspace" inside Claude — a layer where the model silently reasons, notices it's being tested, and plans things it never says aloud. Researchers can now read and even edit it.
Also this issue: a robot doing colonoscopies solo at 86%, ChatGPT inside PowerPoint, Tencent's 295B open model, the origin story of Claude Code, and a $5/month Telegram booking bot you can build yourself. Let's get into it.

Anthropic says it found a small set of internal neural patterns inside Claude that act like a silent “workspace” for thoughts the model can reason with, report on, and use across tasks. The team calls it J-space — a hidden layer where concepts can appear even when Claude never says them out loud.

Image source: Anthropic
Highlights
Claude can report what appears in its J-space
It can silently reason through intermediate steps
Researchers can edit J-space and change Claude’s answers
The tool can reveal hidden goals, fake data, or test awareness
Anthropic says this does not prove Claude is conscious
Why it matters: AI safety is moving from watching what models say to inspecting what they may be thinking before they answer. The open question is huge: if labs can read and steer hidden reasoning, they may get a new way to catch deception, improve honesty, and understand whether advanced models are becoming more mind-like than anyone expected.

How AI-Era Pricing Is Reshaping Finance Operations
Usage-based and hybrid pricing models are changing how B2B companies generate revenue — and creating new headaches for the finance teams behind them.
Tabs co-founder Rebecca Schwartz and PwC Partner Amit Dhir sat down to unpack exactly what that means in practice: how pricing model decisions ripple into revenue recognition, forecasting, and financial ops — and what it takes to scale without piling on manual work.
Watch the on-demand recording to get practical frameworks, real-world examples, and a clear path to operationalizing usage-based revenue — including a forward-looking take on how AI will reshape financial workflows. If your team is navigating pricing complexity heading into the back half of the year, this is worth an hour.

A Robot Just Did a Colonoscopy by Itself
Japanese researchers built an AI-powered colonoscopy robot that completed 62 of 72 full insertions on a training model without human intervention — an 86.1% success rate. It is not ready for real patients yet, but it shows AI may one day handle the most operator-dependent part of colon cancer screening.
This issue also covers CRISPR expanding to 2-year-olds, an AI heart patch that detects deadly rhythms without the cloud, a hormone-tracking wearable backed by $11.6M, and the first generic Wegovy approval all pointing to one bigger shift: healthcare is moving from better tools to more automated care.
Want it weekly?

🔥Big news in AI
🔹OpenAI DevDay 2026 Applications Are Open
OpenAI opened applications for DevDay 2026, happening September 29 in San Francisco. Builders can apply to test new tools, compare notes with other developers, join technical sessions, and bring their hardest questions. AI developer events are becoming less about announcements and more about hands-on access to what builders will ship next.
🔹ChatGPT Is Now Inside PowerPoint
OpenAI made ChatGPT for PowerPoint generally available worldwide across all plans, letting users create slides, update existing decks, rewrite content, polish structure, turn notes into presentations, and edit directly inside PowerPoint.
🔹Tencent Just Dropped a 295B Open Model
Tencent launched Hunyuan Hy3, a 295B mixture-of-experts model built for agentic use cases, with Apache 2.0 licensing for commercial use and a free API window for two weeks. The signal is clear: open models are moving from “good enough experiments” to serious agent infrastructure where cost, licensing, and deployability may matter as much as raw model size.

37 Free Claude Prompts With The AI Report
Subscribe to The AI Report, the free 5-minute daily AI brief for 400,000+ business leaders, and you’ll get 37 Claude prompts free in your welcome email. They’re organised by the 8 situations every manager faces. You get both: the newsletter and the prompts.

Anthropic Revealed How Claude Code Was Built
Anthropic published the inside story of Claude Code, tracing it from an early VS Code assistant and internal CLI tool called clide to the agentic coding system developers use today. The team says the real breakthrough was giving Claude the tools to read, edit, run commands, test code, and keep improving through fast user feedback.

Image source: Anthropic
The bigger shift is software engineering itself. Claude Code did not start as a perfect product — it became useful as models crossed the capability threshold. Now developers are moving from writing every line by hand to managing agents, reviewing outputs, and shipping faster than traditional teams were built for.

Build a Telegram AI Chatbot That Books Appointments (Zero Code)
A clinic front desk wastes half its day on the same five questions. Now imagine a Telegram bot that handles all of it, understands natural language, books appointments, saves them to a Google Sheet, and pings the patient with a confirmation.
The full tutorial walks through the build with n8n, OpenAI, and SerpAPI. Total cost: under $5/month.
The kicker? A patient can say "book me for a broken bone, I'm free around 5ish tomorrow, name's Alex" and the agent maps it to Orthopedic at 17:00 and asks only for the missing phone number.
Same pattern works for salons, gyms, any appointment business. The full prompt, JSON trick, and exact node setup are in the build.

How 2M+ Professionals Stay Ahead on AI
What’s the secret to staying ahead of the curve in the world of AI? Information.
Luckily, you can join 2,000,000+ early adopters reading The Rundown AI — the free newsletter that makes you smarter on AI with just a 5-minute read per day.

AI Is Becoming America’s Investment Engine
McKinsey’s new Catalyzing Competitiveness report shows data center investment is up roughly 200% since late 2022, while non-AI investment is mostly flat. China is adding about $4.4T in net productive assets each year — roughly 4x the US — while America’s net-new capacity is increasingly tied to AI infrastructure.
This Sunday Special covers a $50 Microsoft AI hackathon, Fable 5 returning worldwide after export controls were lifted, the AI Marketing Playbook at $10, free AI courses, live events, and five AI career openings. The investment map is shifting fast — but the next move is still practical: follow the capital, build one project, and ship proof.
Want it every Sunday?

New here? Get the next issue Thursday.
70,000+ AI builders read The AI Entrepreneurs every week.








