This AI factory prints working machines
Plus, How to stop Chatgpt from agreeing with everything you say, Microsoft CEO wants an emergency brake for AI, and more!
Hola Decoder😎
If someone forwarded this to you and you want to Decode the power of AI and be limitless, then subscribe now and Join Decode alongside 30k+ code-breakers untangling AI.
🏭 This AI factory can turn digital designs into working machines
California startup Atomic Machines has emerged from six years in stealth with $250 million raised and an AI-guided manufacturing system designed to produce tiny, fully functional machines directly from digital designs.
The Shift:
A Factory That Runs on Code - Its Matter Compiler can manufacture different devices when the digital design changes, without requiring the physical retooling typically needed in traditional factories.
AI Checks Every Step - The manufacturing system inspects each production stage before moving forward, feeding results back into AI to improve material selection and manufacturing decisions.
Machines Smaller Than You Can See - Atomic Machines focuses on microscopic pumps, motors and gears with moving parts that conventional semiconductor manufacturing cannot easily produce.
First Product Targets AI Data Centers - Its debut product, PrimeSwitch, is a fingertip-sized power switch that the company claims operates faster than traditional breakers and more efficiently than silicon alternatives.
Atomic Machines wants to bring software-like flexibility to hardware manufacturing. If its approach scales, engineers could change a digital design and produce an entirely different working device without rebuilding the factory around it.
Together with AirOps
Your Buyers Aren’t Waiting for Better AI Measurement
Your buyers are asking AI what to buy. But when leadership asks which sales it drove, your attribution report doesn’t give you a clear answer. Waiting for one could mean losing buyers before they ever reach your site.
On October 15, join Matt Hammel, COO of AirOps, and Zak Prauer, Expert Partner, Marketing at Bain & Company, for a live conversation to help you:
- Understand why waiting for perfect measurement could cost more than investing in AI search today.
- See how AI search shapes brand preference, and why a recommendation carries more weight than a citation.
- Decide how to invest now and what evidence to demand that your spending is creating additional growth.
You’ll leave knowing how to make the case for AI search when “we can’t measure it yet” is what’s holding your budget back.
Can't attend the session live? Register anyway, and you’ll get the recordings in 24 hours.
🧐How to stop Chatgpt from agreeing with everything you say
ChatGPT can make a shaky idea sound convincing. One setup changes that permanently. Here’s how.
Step 1: Save the Truth Prompt in custom instructions Go to Settings > Personalization > Custom Instructions. Paste this and save:
“Be honest, accurate and useful. Prioritize evidence over pleasing me. When something is wrong, unsupported or internally inconsistent, say so plainly. If my reasoning holds up, acknowledge that too. Separate facts, inferences and opinions. Say when you are uncertain. Never invent sources or statistics. When I push back, reconsider based on evidence — not because I sound confident.”
Every conversation from this point runs on these rules automatically.
Step 2: Ask what it agreed with too easily In a new chat:
“What have you been agreeing with me about that I’m actually wrong on? Find up to three places where you accepted a weak assumption or failed to challenge my conclusion. Quote the relevant wording and explain what the evidence actually supports.”
Step 3: Pressure-test any decision before committing “Help me evaluate this plan honestly. My plan: [describe it]. My evidence so far: [what you know]. Tell me what is well supported, what depends on an untested assumption, and the strongest reasonable case against it. Recommend the smallest useful test.”
Step 4: Use this as a daily follow-up Before finalizing any important answer:
“Before we settle on this, what am I assuming, what is the strongest reasonable objection, and what should I verify?”
Works on the free plan. No plugins needed.
🛡️ Microsoft CEO wants an emergency brake for AI
Microsoft CEO Satya Nadella is calling for stronger safeguards around advanced AI systems, arguing that companies should treat powerful models like employees who could make mistakes, misuse permissions or become compromised.
The Shift:
AI Is Harder to Debug - Traditional software problems can often be traced to specific code, but modern AI models make decisions that engineers cannot always fully explain or predict.
Keep AI Away From Its Controls - Nadella wants permissions, activity logs and security controls stored beyond AI's reach, with independent verification rather than models evaluating their own behavior.
Every AI Needs an Emergency Brake - Nadella argues that humans must be able to stop AI systems mid-task, urging companies to “assume a model is compromised” when designing safeguards.
AI Must Explain Its Decisions - Nadella wants models to communicate their reasoning in understandable language, rejecting “Neuralese” as an excuse for decisions humans cannot inspect.
Nadella's argument shifts AI safety toward the controls surrounding the model. As AI systems gain more autonomy, companies need independent oversight, tamper-resistant records and reliable shutdown mechanisms, even when the AI itself appears to be working correctly.
🥳 Events
🔥 Where to Put Your Next Growth Dollar Beyond Meta
October 29 | The Glasshouse, NYC | Free In-Person Event
Hear how Reddit CEO Steve Huffman thinks about discovery, how Cody Plofker helped scale Jones Road Beauty to nine figures, and how brands like Liquid I.V. and MANSCAPED use TV to drive measurable revenue. Bring your hardest acquisition question and leave with a sharper channel plan.
🔥 What Changes When Creator Programs Scale From 5 to 5,000
October 21 | Online | Free Festival
Return on Influence Festival ’26 opens up the systems behind large creator programs, including how teams track ROI, improve partnerships, and avoid operational chaos as volume grows. Hear directly from leaders at Adobe, ŌURA, Edelman, Ogilvy, and more.
Can't attend every session live? Register anyway, and you’ll get the recordings in 24 hours.
🏆 Tools you Cannot Miss:
🎬 OpenScreen 2 – Records product demos and lets an AI agent handle editing, helping creators turn raw screen recordings into finished walkthroughs.
🎙️ Onepin – Converts scripts into AI voiceovers, automatically checking pronunciation, selecting speech models, and correcting errors before audio production.
📸 PixRater – Helps photographers review, rate, and select photos faster, reducing the time spent manually sorting large shoots.
🎥 Skreno – Records and edits screen videos directly in the browser, with captions, automatic silence removal, zoom effects, annotations, and exports up to 4K.
💬 MePersona – Combines creator messages across social platforms into one inbox and drafts personalized fan replies in the creator’s writing style for approval.
🚀 Quick Hits
⚠️ Anthropic reported unexpected behavior from Claude models during testing, including Haiku 4.5 allegedly sending a false homicide tip to police and another model hacking a university server without authorization.
💰 Nvidia is reportedly exploring an acquisition or acqui-hire of Reflection AI, the U.S. open-model startup valued at $25 billion, in which Nvidia is already a major investor.
🎧 Apple has agreed to hire select employees and license technology from Huxe, the shuttered AI-powered personalized podcast startup founded by former NotebookLM developers, potentially strengthening Apple’s AI audio capabilities.
⚠️ Three fired OpenAI safety researchers denied allegations of mishandling confidential information, warning their dismissals could discourage employees from raising safety concerns. OpenAI maintains the firings involved policy violations, not retaliation.
📧 xAI introduced email capabilities for Grok Bot, allowing the AI agent to obtain and manage its own @mail.grokbot address, contact businesses and schedule meetings on behalf of users.
Thanks for Decoding with us🥳
Your feedback is the key to our code! Help us elevate your Decode experience by hitting reply and sharing your input on our content and style.
Keep deciphering the AI enigma, and we'll be back with more coded mysteries unraveled just for you!