News
Anthropic launches Claude Duet, an agent that messages you first
Anthropic’s Claude Duet is a proactive, approval-gated agent for US Pro and Max subscribers that watches your connected apps and starts work before you ask.
Anthropic’s draft AI safety bill: 50 deaths, $1bn, and a judge
Anthropic has put specific harm thresholds into a draft US bill for halting frontier model releases, and whoever writes the first numbers tends to set the terms of the debate.
Microsoft says frontier models can’t be trusted, ships Guardian Agents
Microsoft now sells both the AI agent and the watchdog for the agent, and its security chief says frontier models cannot be trusted, so the real question is who controls your oversight layer.
AI agents are moving money. Banks can’t yet verify who sent them
Agents are spending real money, but no standard lets them prove who they are, who authorised them and what they may do, so builders should assume a hostile perimeter.
DeepSeek V4.1-Flash-1017: MIT weights and a $0.75 search API
DeepSeek’s V4.1-Flash-1017 puts a 763B-parameter model under MIT and prices web search as a metered commodity, and the second move matters more than the benchmark gains.
Modal raises $750M at $15.75B as run rate tops $1B: what it signals
Modal’s $750M Series D and claimed $1B+ run rate suggest the unglamorous layer where AI workloads actually execute is now one of the biggest line items in production AI.
Microsoft finishes MAI-Base-2: a 2.8T in-house model bound for Foundry
Microsoft has finished pre-training a 2.8-trillion-parameter model built without OpenAI, so Azure builders should stop assuming one model supplier and make their stack swappable.
Apple’s Minions turns idle Macs into a GPU inference pool
Apple’s researchers have published Minions, a system that schedules inference jobs onto idle Apple Silicon Macs, and the interesting part is the orchestration, not the hardware.
Meta’s Muse Wonder Full scores 56 on AA but isn’t on the API
Meta’s first frontier models are close to the top of the independent rankings, but the stronger one is only available inside a Muse Code subscription, with no API access.
Meta’s Muse Spark helps solve five open maths problems via chat
Meta’s Muse Spark co-authored six maths papers through a standard chat window, which shifts the hard part of AI work from model access to problem selection and verification.
Gemini gets its own email address: Google bets on the identity layer
Google’s new Gemini agent arrives with its own Workspace account, email address and agent-attributed audit trail, and routes work to Anthropic’s Claude. The real competition is now over who owns AI’s workplace identity, not who has the best model.
Mistral Large 4 lands: 1T parameters, $1.36/M, weights due 31 Oct
Mistral’s trillion-parameter Large 4 is a strong specialist and a weak generalist, priced far below closed frontier models, and it was trained on compute Mistral owns in Europe.
GPT-6 reaches all ChatGPT tiers with Intelligent UI: what changes
GPT-6 is now rolling out to every ChatGPT tier, and its Intelligent UI turns answers into streamed, interactive interfaces, which is a pattern every team shipping AI output should study.
FTC will compel AI lab testimony; former chair Khan: self-regulation failed
US regulators are using existing consumer protection law rather than waiting for AI-specific legislation, so any business deploying agents on customer data needs a defensible paper trail now.
Amazon Ads Agent replaces Ad Console: what changes for advertisers
Amazon has merged Ad Console and its DSP into Amazon Ads Agent, a plain-language interface that sets up and optimises campaigns. The strategy and the measurement still need to stay with you.
What We Do
Product
Consumer and enterprise software, built with taste and shipped fast.
Infrastructure
AI agents, automation, and orchestration - production systems, not demos.
Advisory
Hands-on engineering and strategy for teams shipping AI into production.