engblogs

summaries of the latest blog articles from your favorite tech companies.
OpenAI

Perplexity and Intelligence

How perplexity measures progress in AI models, and why it could help developers establish a shared approach to evaluating emerging capabilities.

Publisher summary

11/22/2026
AWS ML

Meta ads is now available in AWS Quick

AWS Quick’s Meta ads MCP connector lets marketers analyze and manage campaigns in natural language, combine ad data with other business sources, and create reports or workflows, while requiring approval before write actions.

11/22/2026
Together AI

Security event: Prompt leakage for requests in June and July of 2026

An exposed repository credential let an unauthorized party download prompt samples from selected Together AI serverless model requests in June and July 2026; the company is investigating exposure and has strengthened credential controls, code scanning, AI-agent workflows, and default zero data retention for new accounts.

11/21/2026
Together AI

Qwen4 Max is live on Together AI

Together AI’s Qwen4 Max handled a 652,519-token Tornado repository and bug report in one streamed request, producing a patch that passed the maintainers’ regression tests and full 1,673-test suite; automatic prompt caching made a repeat request substantially faster and cheaper.

11/21/2026
AWS ML

New in AWS Quick: More capable flows, an updated web experience, and agents that work together

AWS Quick adds chat-built workflows with deterministic steps and longer-running, high-volume processing, a web app for persistent agent work, coordinated multi-agent responses, and connectors to Meta Ads and Amplitude, alongside expanded skills and administrative controls.

11/21/2026
AWS ML

Cut self-hosted LLM cost and latency with the distributed KV caching agent skill

The AWS Agent Toolkit’s distributed KV caching skill uses workload traces and fit checks to size and configure an ElastiCache for Valkey shared cache for LMCache-enabled vLLM or SGLang fleets, helping reuse attention state across servers while guarding against mismatches and transfer failures.

11/21/2026
AWS ML

Introducing managed cluster extensions for observability and user management in Amazon SageMaker HyperPod Slurm

Amazon SageMaker HyperPod cluster extensions declaratively install, update, and monitor managed observability and user-management software across Slurm nodes, keeping configurations consistent as clusters scale while complementing custom AMIs and lifecycle scripts.

11/20/2026
Google Research

How Google Duplex evolved to also call businesses for you in Gemini

Google evolved Duplex from task-specific recurrent neural networks to a Gemini-powered calling system that combines speech understanding, planning, and dialogue generation to handle flexible real-world business calls with low latency and policy safeguards.

11/20/2026
Salesforce

How an AI Agent Cut Production Change Reviews From 24 Hours to 12 Seconds

Salesforce’s ROAM agent cut median production-change review time from about 24 hours to 12 seconds by normalizing evidence across Slack, change records, and release calendars, using deterministic approval gates, and escalating unverifiable requests to humans.

11/20/2026
AWS ML

Introducing Amazon SageMaker HyperPod Global Inference Gateway: One endpoint across Regions and clusters

Amazon SageMaker HyperPod Global Inference Gateway provides one SigV4-authenticated endpoint to route requests across registered clusters and Regions using KV-cache headroom, queue depth, proximity, and error rate, with PrivateLink connectivity and automatic failover.

11/20/2026
ElevenLabs

How Fundación Televisa interviewed more than 1,000 entrepreneurs with voice AI

Fundación Televisa integrated an always-available voice AI agent into POSiBLE’s web platform to conduct 1,036 five-minute applicant interviews in four weeks, and used a similar setup to interview 394 scholarship applicants in two days.

11/20/2026
ElevenLabs

How to fix text to speech accuracy problems

Improve TTS pronunciation with reusable PLS dictionaries or inline phonetic tags, make scripts easier to speak through clear wording and punctuation, and troubleshoot API failures by status code while testing difficult terms before production.

11/20/2026