Perplexity Monitor tracks web changes for $1 per 1,000 runs
Today's top 20 insights for PM Builders, ranked by relevance from X, GitHub, and LinkedIn.
Perplexity Monitor tracks web changes for $1 per 1,000 runs
#1 𝕏
Aravind Srinivas announced Perplexity Monitor, an API for tracking events, topics, or webpage changes and receiving updates via webhooks. The stated cost is $1 for 1,000 runs.
#2 𝕏
Philipp Schmid recapped seven Google DeepMind developer launches shipped during the prior week: Gemini 4 Carbon, Gemini 4 Hydrogen, Gemini Decisions API with Gemini Decision 1, self-hosted sandboxes for Managed Agents, a Multimodal Knowledge Agent benchmark, Gemini Robotics 2.5, and AlphaGenome 2.
Also covered by: @Philipp Schmid
#3 𝕏
OpenAI shared that ChatGPT for Healthcare is advancing with more capable computer-use agents, stronger clinical answers, and 20+ new healthcare plugins. AbridgeHQ, AmbienceAI, Cigna, CVS Health, and other healthcare organizations are using OpenAI’s models for prior authorizations, referrals, discharge planning, and patient outreach.
#4 𝕏
NVIDIA Nemotron for Science was announced at SC26 by Ian Buck, NVIDIA’s VP of Hyperscale and High-Performance Computing. The collection includes open models, datasets, tools, and post-training recipes for building more accurate and efficient scientific AI agents that can plan tasks, run simulations, check results, and adjust as they proceed.
#5 𝕏
Hugging Face announced a direct OpenRouter integration to accelerate open-source AI, offering the same token and endpoint alongside live model statistics on the Hub and an inference widget redesigned by both organizations.
#6 𝕏
Google DeepMind announced that the Google Antigravity program for 10,000 scientists is expanding to include free access to Gemini 4 Carbon.
#7 𝕏
Aravind Srinivas shared an open-world browser driving game built by a Teammate session on Fable 5.2, featuring a code-generated neon 1986 Miami with no game engine, Three.js, or external assets. The 9 MB game uses a custom WebGL2 renderer and about 63,000 lines of JavaScript; Teammate rendered a 40-second film on a rented cloud GPU at 1080p60.
#8 🐙 elementalsouls/Claude-BugHunter
elementalsouls/Claude-BugHunter - elementalsouls/Claude-BugHunter, created by Sachin Sharma and hosted by elementalsouls, is an MIT-licensed bundle of reusable agent skills for external bug hunting and red-team work; its README lists 83 skills (the repository summary says 82), 15 slash commands, and 681 disclosed-report patterns across 24 vulnerability classes. It has 8,355 stars, gaining 1,007 daily and 1,747 weekly. Worth trying if you need structured, external-only security workflows in a supported agent harness.
#9 🐙 google-research/paper-orchestra
google-research/paper-orchestra - google-research/paper-orchestra is an Apache-2.0 multi-agent framework shared by google-research that turns ideas and experimental logs into LaTeX papers, with specialized agents, templates, a CLI, Bash script, and Streamlit demo. It has 2,344 GitHub stars and 1,175 daily trending gains. Worth trying if you need an automated paper-writing workflow and can provide raw materials and a LaTeX template.
#10 𝕏
DeepLearning.AI recapped Anthropic’s S-1: revenue rose from $1.2B in H1 2025 to $16.3B in H1 2026, with $974M in Q2 operating profit as compute costs fell from 251% of revenue to 48%. Claude wrote more than 90% of codebase contributions in September.
#11 𝕏
OpenAI recapped its study of more than 1,800 ChatGPT for Academic Researchers participants, in which scientists reported nearly doubling their productivity with AI. Using privacy-preserving methods, the study examined ChatGPT use across developing ideas, analyzing data, and writing code.
#12 in
Dharmesh Shah announced HubSpot agentic users, which admins can provision, assign permissions to, and audit like regular users. The feature is intended to make AI agents productive contributors while maintaining control and governance.
#13 𝕏
Metric Lab in ParseBench lets teams weight tables, charts, forms, and text, set costs for critical, major, and minor errors, and re-rank all 19 parsers live in the browser. Users can start with a default profile or build a custom one based on their document needs.
#14 𝕏
Mercedes-Benz uses AI agents powered by NVIDIA Cosmos to detect assembly defects, flag missed steps in real time, and improve production quality. Mercedes-Benz Board Member Michael Schiebe described how physical AI is helping build self-learning factories where machines adapt and work alongside people.
#15 in
Peter Yang recapped five takeaways from Nate at Anthropic for using Claude in design: provide context, research familiar UX patterns, document design biases, start with wireframes, and request 5 options before remixing preferred elements into 5 more.
Also covered by: @Peter Yang
#16 𝕏
Udi Menkes 🚢 recaps that OpenAI announced its ChatGPT Wallet “last Friday,” enabling ChatGPT’s personal agent to complete purchases after user approval in a limited US Pro rollout through “dots,” without accessing card details and with a required security check. In a demo—not a real order—an unbranded gift request for about $100 produced a $59.97 herb-box suggestion, highlighting the need for accurate, current, agent-accessible product data and risk systems that accept legitimate agent purchases.
#17 𝕏
Guillermo Rauch shared four intelligence roadmap recommendations: use open SDKs and harnesses, retain model choice through gateways, own traces and data while enforcing ZDR, and own models through private inference.
#18 𝕏
Guillermo Rauch shared a Vercel read on how the company conducts evals, contrasting traditional human-user interviews and mocked-up journeys with an agent-user workflow where evals inform the software engineering loop.
#19 in
Greg Isenberg shared a six-part summary of his 49-step carousel for making companies more AI-driven by organizing company context, turning workflows into skills, assigning agents, building internal tools, updating playbooks, and applying agents to growth. He recommends starting with one folder documenting company decisions and recorded calls, and says his team plans to select 10 companies to work with in 2027 through latecheckout.agency.
#20 𝕏
Garry Tan shared a full report on evaluations for gbrain, linking to the gbrain-evals repository under GitHub owner garrytan.