AI Act Digital Omnibus: which deadlines moved and what small teams still owe
The Digital Omnibus pushed the AI Act's high-risk rules to 2027 and 2028. Literacy, chatbot disclosure and two 2 December 2026 deadlines stay put.
5 min read
AI Observer: News, guides and analysis on AI agents, workflow automation and chatbots, for teams that want to save time on repeated work.
Runway’s open Praxis-1 robot model needs an 80 GB GPU at full precision
The 27B world action model scores 46.2% on NVIDIA's RoboLab-120, just ahead of FLUX 3 Action within the error bars. Full precision needs one 80 GB GPU.
3 min read
Box Architect will build agent workflows from a plain-language brief
Box's new tool will design, test and tune document workflows from a description. The beta reaches Enterprise Advanced this quarter, with no price yet.
3 min read
Larridin raises $30M to track what companies get for their AI spend
Pelion led the Series A for a startup that ties AI bills to shipped work. Its customer data puts the median engineer at $213 a week in billed AI coding spend.
3 min read
Supermicro plans 10,000 SambaNova SN50 chips to run decode beside NVIDIA GPUs
Supermicro will sell racks where NVIDIA GPUs read the prompt and SambaNova SN50 chips write the answer. Speed figures exist for SN50, not for these systems.
3 min read
Claude Meetings joins Google Meet calls to take notes and draft docs
Anthropic's beta puts Claude in Google Meet calls on Team and Enterprise plans. It is on by default for Team, and two settings deserve a look first.
3 min read
Clinicians rated AMBOSS’s grounded AI above ChatGPT in a study it helped shape
Doctors and students gave a grounded clinical AI better marks than ChatGPT, but the study measured impressions, not accuracy, and AMBOSS chose the questions.
3 min read
Latest stories
- AI Act Digital Omnibus: which deadlines moved and what small teams still owe
- Runway’s open Praxis-1 robot model needs an 80 GB GPU at full precision
- Box Architect will build agent workflows from a plain-language brief
- Larridin raises $30M to track what companies get for their AI spend
- Supermicro plans 10,000 SambaNova SN50 chips to run decode beside NVIDIA GPUs
- Claude Meetings joins Google Meet calls to take notes and draft docs
- Clinicians rated AMBOSS’s grounded AI above ChatGPT in a study it helped shape
Eight desks, one place to look
Every story on AI Observer lives on a desk. Pick one to see what is new there.
- Models and researchNous says its Arete-27B fine-tune cut Qwen propaganda answers from 22% to 0.9%2 this week
- Agents and automationClaude Meetings joins Google Meet calls to take notes and draft docs6 this week
- Chips and hardwareSupermicro plans 10,000 SambaNova SN50 chips to run decode beside NVIDIA GPUs2 this week
- Business and startupsLarridin raises $30M to track what companies get for their AI spend1 this week
- Policy and regulationAI Act Digital Omnibus: which deadlines moved and what small teams still owe5 this week
- Health and scienceClinicians rated AMBOSS’s grounded AI above ChatGPT in a study it helped shape1 this week
- RoboticsRunway’s open Praxis-1 robot model needs an 80 GB GPU at full precision3 this week
- Design and productChatGPT Remix lets Business and Enterprise teams fork a teammate’s site1 this week
Latest stories
All news
The AI Observer Awards will recognise AI work that was checked, shipped and still holds up a year later.
A planned programme. Nominations, jury and criteria will be published here before anyone is asked to enter.
Planned categories
- AI agency of the year
- Best AI tutorial
- Best interview
- Tool that held up