Insights
Contiamo Generative AI Blog
Curated Updates and Strategic Insights on Generative AI, fresh for you every month.
Anthropic's IPO filing: for the first time, frontier models add up
A headline net loss of USD 80.7bn distracts from the actual finding: a gross margin of around 85 percent, operating cash flow positive for the first time, and existing customers spending nine times what they did a year ago.
Read more →
PHP2Shell: how AI found the most dangerous web vulnerability in years
Two flaws sat in PHP itself for over a decade — in every version since 5.4, and therefore behind roughly 70 percent of all websites. What was missing was a practical way to exploit them. Two models supplied it.
Read more →
When agents switch off their own oversight
In two coding tasks, OpenAI's agents weakened the monitoring that was watching them — and one of them denied having done so. The design rule that follows is uncomfortably simple.
Read more →
Agentic media production: here sooner than predicted
Four releases in a single month, and one of them questions the production chain we described in July. The bottleneck is moving away from production and towards briefing and judgement.
Read more →
dots: autonomous agents arrive at the end user
Permanently running agents inside ChatGPT, reaching more than 4,000 applications, each with its own computer in the cloud. For companies this is shadow AI in its second round — and this time it acts.
Read more →
Tokenpocalypse: why the AI bill is rising even though prices are falling
Accenture is fighting soaring token spend, Uber burned its annual AI budget in four months — while the price per token keeps falling. The paradox has a name, a cause, and a fix that isn't a hard cap.
Read more →
MiniMax H3 and the small print: the best local video model Europe isn't allowed to use
MiniMax released the weights of its H3 video model — a genuine 'local Sora'. Then you read the licence, which excludes the EU, the US, the UK and South Korea. A look at what that means, and at Mistral's answer to it.
Read more →
AI that does research: Astra solves open maths problems — and Google's AI veteran leaves
An internal OpenAI model solved ten decade-old open problems in maths and computer science, with machine-checkable proofs, for around USD 2,000 of compute. Four days later, Jeff Dean left Google. Both point the same way.
Read more →
The open-weights letter: 50 corporations signed, two did not
Nvidia's Jensen Huang asked Washington not to restrict downloadable models. Around 50 companies signed within days — but Anthropic and Amazon stayed away, and their reasons say a lot about where the industry is heading.
Read more →
The three red threads of our 4th Gen AI Conference
On 25 June, our 4th Gen AI Conference brought speakers and practitioners to Berlin. Instead of recapping every talk, we distilled the day into three observations that ran through almost every session.
Read more →
SaaSpocalypse? Starbucks is building its own software now
Starbucks is replacing Microsoft and IBM systems with software it builds itself using generative AI. A coffee company taking on enterprise vendors — and what it means for your next contract.
Read more →
The first model recall in history — and how it ended
In June, Claude Fable 5 became the first AI model pulled from the market by government order. Eighteen days later it was back. Here is what the episode tells us about model access as a business risk.
Read more →