Thursday’s signal: AI is moving from a tab into the stack. Replit’s Free Mode, powered by OpenAI’s GPT-5.6 Luna, promises Core subscribers up to 30 times more creation, 30 hours of chat each month, and usage resets every 5 hours for $20 per month on annual billing. The model is becoming background infrastructure just as Etched raises $700 million at a $21 billion valuation to build the racks underneath it. The strategic question is which entire workflow becomes cheap enough to stay inside the agent.
Cheap interfaces still require expensive machinery. Etched says its latest inference round raised $700 million at a $21 billion valuation after Jane Street tested the hardware and took delivery of its first rack; the startup also says fewer than 1% of people can access frontier models. Replit is pushing that expensive capacity outward through a $20-per-month annual plan promising 30 hours of chat and up to 30 times more creation. The margin battle is migrating from model access to the full cost of completing useful work.
Once assistants enter work, the real product becomes permissioning. Meta’s Mac app can now operate across 3 surfaces—web, mobile, and desktop—and connect to 4 business systems: Facebook, Instagram, ad campaigns, and Google Workspace. Replit’s Free Mode resets every 5 hours, while Meta’s assistant can create 3 artifact types—decks, documents, and spreadsheets—and run recurring tasks. The workflow race is no longer about a single chat box. It is about how much context companies will centralize, which controls remain visible, and who earns the next default surface.
The Pulse
Alexa+ is now included on compatible Fire TV devices at no additional cost
Free Alexa+ turns the television into Amazon’s default ambient AI surface.
Google brings a student hub and interactive study tools to Gemini and Search
Interactive simulations make search a classroom, but citations still carry the grade.
Calendly launches Notetaker and its Callie AI assistant
Calendly is betting meeting context matters more than another perfect transcript.
Wuying-Browser-Agent sets an open-source state of the art on browser-use benchmarks
Open browser agents are becoming credible enough to force serious permission design.
Latest from PulseMark
 |
How to Configure OpenRouter Model Fallbacks That Actually Work
Build fallbacks that preserve schemas, tool behavior, and privacy instead of treating every HTTP 200 response as a successful recovery. |
 |
How to Run Qwen3.6 27B Locally With llama.cpp
Choose a Qwen3.6 27B quant with enough room for context, buffers, vision, and GPU offload before launching it in llama.cpp. |
 |
Grok Bot Tutorial: Build an AI Teammate That Works 24/7
Set up Grok Bot as an always-on cloud-computer teammate that can keep working after your laptop closes. |
 |
Unsloth Dynamic 3.0 GGUF Guide: Choose a Qwen3.8 Quant
Compare Dynamic quants by file size and memory headroom so Qwen3.8 fits the machine you actually have. |
That’s the signal.
— The PulseMark Team
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.