-
Models
Google cuts free Gemini to Flash-Lite from 9 October
Engadget reports that from 9 October free Gemini users lose Flash and Pro and keep only Flash-Lite. The $5 AI Plus plan keeps Flash and Flash-Lite but loses Pro, while the $20 AI Pro plan keeps the advanced models. Google is also adding low, medium and high effort levels across tiers. The report does not cover API changes.
Why it matters: If you prototype or demo on the free Gemini app, expect weaker answers from tomorrow. Check which model you are really testing against before you compare results.
Source: Engadget ↗ -
Models
Claude Haiku 5.5 is on Amazon Bedrock
AWS says Claude Haiku 5.5 is now available on Amazon Bedrock and Claude Platform on AWS, including GovCloud. Anthropic calls it the fastest and most efficient model in the Claude 5.5 family, about 75% cheaper than Haiku 4.5 for most tasks, and the first Haiku with effort controls. It is aimed at subagents, code review, classification and other high-volume work, paired with Opus 5.5 for planning.
Why it matters: A cheap, fast model for the execution steps is the pattern that makes agent workflows affordable. Treat the 75% figure as Anthropic's claim and test it on your own prompts.
Source: AWS Machine Learning Blog ↗ -
Agents
Shopping and booking agents keep getting blocked by websites
TechCrunch reports that personal agents such as Meta's Muse and ChatGPT's Dots are being stopped by sites like Amazon, Walmart, Delta and eBay, sometimes on purpose and sometimes by anti-bot tools. Meta, Walmart, Stripe, Sierra and several others are working on an open standard for agent-to-agent communication in online commerce, meant to tell legitimate consumer agents apart from malicious bots.
Why it matters: If you build browser or checkout automations, plan for blocks and prefer official APIs. A shared standard would help, but it is not here yet.
Source: TechCrunch ↗ -
Agents
How a 3-person team cut database diagnosis from 45 to 10 minutes
In an AWS case study, Cornerstone OnDemand describes Orion AI, a hub-and-spoke system of 13 specialised agents built in six months on Amazon Bedrock and the Strands Agents framework. Diagnosis fell from 45 to 10 minutes (78%), and repeated alerts dropped by 65%. Their design choices: agents scoped by domain, keyword routing with a semantic fallback, and session memory that bypasses stale data for live metrics.
Why it matters: A rare, specific multi-agent write-up with numbers and trade-offs. Note it is published by AWS about its own platform, so read it as a design reference, not a benchmark.
Source: AWS Machine Learning Blog ↗ -
Startups
Anthropic offers startups a free year of Claude Team and $1,000 in credits
TechCrunch reports the programme includes a year of Claude Team with up to five seats, $1,000 in API credits, access to the Claude Marketplace and virtual office hours with Anthropic's applied AI team. Companies founded in the last five years, or funded in the last two, can apply through claude.com/programs/startups.
Why it matters: Worth a look if you are shipping an AI product on a tight budget. Read the eligibility terms before you rely on it.
Source: TechCrunch ↗
Each brief is compiled with automation and edited by a person. Summaries stick to what the linked source reports - read the source for the full story.