Mozilla has quietly run an MDN MCP server since June, giving coding agents live access to current web platform documentation instead of a training cutoff. It is a small experiment with a lesson worth applying to every AI tool we use.
A performance budget agreed before a line of code is written, not measured for the first time once a site is live. Rebuilding Gun Trader taught us why that document has to carry hard numbers, not a vague promise to make it fast.
The V&A's Global Design Forum returned for London Design Festival under the theme Worlds in Motion. Stage designer Es Devlin opened the three day programme with a keynote on new technologies and the discipline of designing futures.
Photoshop 27.10 adds a non destructive Light Adjustment Layer, a small change that quietly admits Lightroom and Camera Raw had the better interface for basic light edits for the best part of two decades.
A council planning officer, a landlord, and a family buying their first self build home want completely different things from the same container conversion. We build three separate render briefs from one underlying 3D model.
GreyNoise found an attacker running hundreds of AI agents in parallel to exploit PaperCut NG and MF, hitting 440 instances across 48 countries in hours. The operator's own exclusion list failed, hitting countries it meant to avoid.
The W3C's updated WCAG 3 draft moves accessibility from a page by page pass or fail checklist towards an organisational practice graded across a whole product. A recommendation is not expected before 2029, but the thinking is worth building towards now.
Turning on a scanner is the easy part. What a scan is for, when the light and access are right, and what will not still be there when the render ships decide whether a day of capture becomes a usable asset.
The V&A's Awakenings season, part of London Design Festival, pairs generative AI with centuries old textile craft rather than setting the two against each other. Two commissions in particular are worth a look.
DeepSeek's V4.1 Flash edged past Moonshot's Kimi K3 on Terminal-Bench 2.1 this week, 90.6 to 88.3. Open weight benchmark leadership is now turning over faster than most teams can update their internal recommendations.