Transparent AI costs require showing the unit before the work runs
If you're building or advising on agent features, this gives you a concrete pattern: surface the cost unit at the decision moment, not buried in a settings panel.
Context between products becomes the interface design problem
It's a useful provocation for positioning work: if agents bridge context across products, the thing worth designing might be the handoff, not the screen.
Design systems are principles in need of enforcement
A concrete prompt to write down the reasoning behind decisions in the design system you maintain, since no tool will enforce that discipline for you yet.
LLMs erodetrust in software signals
Kingsbury's point that generated docs and code erode trust signals means you'll need new ways to prove rigor that can't be faked by looking polished.
AI UI generators still miss error handling and recovery
A reminder to run AI-generated screens through basic heuristics for error handling and recovery before anyone mistakes polish for completeness.
Arca's agent architecture for long-horizon financial work with human guardrails
A rare look at agents doing multi-week work with persistent memory and a human making the final call, useful if you're designing for regulated or high-stakes workflows.
New York Times anniversary mark coordinated across editorial, advertising, and events
Worth studying how a 175-year-old institution keeps one mark legible across editorial, ads and events without it fragmenting into noise.
AI color palettes pass tests but fail in practice
A specific check worth running: AI-generated color palettes can pass individual contrast tests while failing as a system, so audit holistically, not component by component.
Transparent AI costs require showing the unit before the work runs
If you're building or advising on agent features, this gives you a concrete pattern: surface the cost unit at the decision moment, not buried in a settings panel.
Pick one AI feature you've shipped or are prototyping and write down, in one line, what unit of cost (seconds, tokens, requests) the user should see before they commit to running it.