The top things worth knowing about in AI today.
Microsoft launched Frontier Company, a $2.5 billion unit staffing about 6,000 engineers and specialists who embed inside customer organisations to build AI systems that show measurable results. Amazon announced a similar $1 billion venture two days earlier. The bet is that the hard part of enterprise AI is now deployment and integration, not model quality: one survey found 71% of large-company executives named organisational readiness, rather than the technology, as the main barrier.
Read moreMeta released Muse Image, the first image model from its Superintelligence Labs, now built into Meta AI across Instagram and WhatsApp, with Facebook and Messenger to follow. It plans a layout before drawing, renders legible text inside images, blends multiple photos, and lets users pull in public Instagram accounts as reference images. The effect is free image generation placed directly inside apps most people already use daily.
Read moreThe Government of Alberta ran about 50 Claude agents to scan 466 million lines of code across 1,280 applications in 20 hours, work it estimates would have taken 6.5 years by hand. Claude then generated and tested patches, and rebuilt a benefits portal in four to five days that had originally taken five months. It is a concrete case of AI compressing the security and legacy-modernisation work that governments and large firms usually defer for years.
Read moreStartups raised $510 billion worldwide in the first half of 2026, already more than the $440 billion invested across all of 2025, according to Crunchbase. More than 70% of second-quarter capital went to AI companies, up from roughly 50% a year earlier, and OpenAI and Anthropic together took $217 billion, or 43% of the half-year total. The money is real, but heavily concentrated in a small group of frontier labs.
Read moreOpenAI began a limited preview of three new models: Sol, its most capable, plus the cheaper Terra and Luna, available only to about 20 trusted organisations through the API and Codex and not yet in ChatGPT. At the US government request, OpenAI shared the models and its release plan with officials before the narrow rollout, citing stronger cyber and biology safeguards. Pricing runs from $5/$30 per million tokens for Sol down to $1/$6 for Luna, with wider availability due in the coming weeks.
Read moreThe EU Council gave final approval on 29 June to a package that eases and reshuffles parts of the AI Act. It pushes the deadline for national regulatory sandboxes to August 2027, but cuts the grace period for labelling AI-generated content from six months to three, now due 2 December 2026. The Commission gains enforcement powers over general-purpose AI models in August 2026, so firms operating in Europe get more time on some duties and less on content transparency.
Read moreThe Australian government signed a five-year volume-sourcing agreement with Microsoft, commencing 1 July, giving public-service agencies access to Copilot, Microsoft 365, Azure and Dynamics 365 under a single arrangement. It is aimed at speeding AI and cloud adoption across the Australian Public Service. The move points to AI settling into routine government operations rather than isolated pilots.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.