{{first_name | Leader}}, welcome back.
Every AI company eventually runs into the same three questions: Can people trust it? Can you afford to run it? Will customers pay for it? Today's stories answer one of those. These are today's updates.
ElevenLabs just watermarked its audio output
A Databricks veteran says he can cut AI power use 1,000x
Claude's paid user base is growing fast
Tools, resources, and a prompt to train ChatGPT to handle your routine tasks. ⬇️
. Top News.
The deepfake audio problem has been building quietly for years. Anyone with a browser can now clone a voice. Most people can't tell the difference.
ElevenLabs just made detection easier. The company has adopted Google's SynthID, embedding invisible watermarks directly into AI-generated audio. It's already live for free users on text-to-speech output, with full coverage across all ElevenLabs generations rolling out over the coming weeks. Users can verify content through the ElevenLabs Audio Detector.
This matters beyond ElevenLabs. SynthID is becoming an industry standard across Google's own tools and is now a major third-party audio platform. The infrastructure for verifying AI-generated content is being built piece by piece.
AI's power consumption problem isn't going away. Every major lab is burning through electricity at a rate that's becoming a genuine infrastructure constraint.
Naveen Rao, formerly head of AI at Databricks, just launched Unconventional AI with a direct shot at that problem. The company's oscillator-based architecture claims to make inference up to 1,000 times more power efficient than conventional systems. They've already released Un-0, an image-generation model built to demonstrate that the architecture can match standard diffusion models in output quality.
ChatGPT still has more paying users. That's not the interesting part.
Claude's consumer revenue has grown roughly 75% since January 2026, according to credit card transaction data from Indagari. Its paying consumer base has grown month over month consistently. And "Claude" has become one of the most searched AI terms, with demand for Claude courses among independent learners running well above ChatGPT.
For anyone tracking where enterprise AI adoption is actually heading, the paid consumer market is usually the leading indicator. It's where professionals try tools before bringing them into work. Claude closing that gap with ChatGPT, without the brand recognition advantage or the consumer marketing spend, says something real about product quality and the momentum.
. Signals.
Tools
250K leaders read this every morning before making AI decisions. Mindstream is the daily AI newsletter for people who need clarity, not just coverage. No hype, no jargon — just the signal: what's new, why it matters, and how smart teams are actually applying it. If you're serious about staying ahead, it belongs in your inbox. Subscribe for free.
Links
. Poll.
Where is AI improving enterprise sales fastest?
. Market.
Funding
Coval raised $28M (Series A) to advance its voice AI evaluation platform.
Runlayer raised $30M (Series A) to help enterprises become AI-native.
Socials
. Prompt of the Day.
Train ChatGPT to Handle My Routine Tasks
When to use this?
When you want ChatGPT to reliably handle routine work the same way every time, without micromanagement.
You are learning how I handle routine tasks.
I’ll share examples of how I usually do this kind of work.
Your job:
Identify my preferred format and level of detail
Note what I care about vs what I skip
Capture rules I follow (length, tone, structure, speed)
Summarize a short execution checklist you should follow
Confirm by completing one small task using these rules
Apply this checklist to similar tasks going forward unless I say otherwise.
Examples:
[paste 2–5 examples of routine work: emails, summaries, reports, task lists]P.S. Get more such prompts in the Prompting Playbook (free for you)
Stay curious, {{first_name | leaders}}
PS. If you missed yesterday’s issue, you can find it here.