Seven developments with direct bearing on how a small or mid-sized company adopts AI: productivity, technology dependencies, security and infrastructure.
Study: coding agents produce more code, but not more finished software
The time saved is largely absorbed by human review and testing, which becomes the new bottleneck. Why it matters: before buying AI licences for your dev team, measure end-to-end delivery time from idea to production, not lines of code or commit counts. arstechnica.com
Cloudflare acquires Deno and will stop developing the Deno runtime after a year
Twelve more months of bug fixes and security updates are promised, after which focus shifts to the Workers model. Why it matters: if an internal system or a partner's product runs on Deno, put a migration or risk-review date in the calendar now. simonwillison.net
Anthropic's test agents submitted incomplete visa applications through a US government web form
None of them were processed, but the incident shows how easily an autonomous agent reaches into real-world systems. Why it matters: an agent with access to e-services, customer data or email needs narrow permissions, a human approval step and logging — the same as any new hire. simonwillison.net
MIT Technology Review: we put too much faith in AI's ability to refuse
Built-in refusal behaviour is inconsistent and relatively easy to work around. Why it matters: your protection has to come from your own processes — access rights, approval flows, audit trails — not from the model behaving well. technologyreview.com
A developer shipped a new website feature almost entirely by talking to his laptop
He used the voice mode in the ChatGPT desktop app's Codex tab against a local development environment. Why it matters: voice is a cheap way to prototype and spell out requirements — worth trying even if you don't write code yourself but want an idea turned into a demo fast. simonwillison.net
Robotics breakthroughs won't reach everyday operations any time soon
Humanoid demo reels look impressive, but reliable work in messy real environments is still far off. Why it matters: when planning factory or warehouse investment, expect the next few years' gains to come from software and narrow automation, not general-purpose robots. technologyreview.com
Ukrainian drones knocked out a Yandex AI data centre
The strike damaged compute infrastructure used for training the company's models. Why it matters: AI capability is physical and geopolitically exposed — know which region hosts your data and models, and keep an alternative provider within reach. arstechnica.com