Non-engineers building AI agent systems are repeating decades-old software architecture mistakes. Here's why that matters.
NN/g researchers identify a 'big ball of mud' pattern emerging among non-technical 'vibe architects' building complex AI agent systems. Without software engineering fundamentals, these systems become brittle and hard to maintain. The piece draws on a recent user study and connects current AI builder behavior to well-documented failure modes in software architecture history.
Designing for AI agents means arguing about API contracts, not just screens. A practical framing for UX practitioners.
The article argues that as AI agents operate across product surfaces without a fixed UI, designers need to develop API literacy — understanding rate limits, status codes, and service contracts — not just visual craft. It draws on service design methodology to help practitioners map invisible agent interactions and advocates for designers having a seat at the systems-architecture table.
A pointed critique of AI regulation theater, with an eye-opening detour through vibe-coded apps leaking user data.
The piece critiques the gap between regulatory intent and real-world outcomes, using GDPR's largely symbolic effect on US users as a case study. It references a 'vibe-coding graveyard' of AI-generated apps — including Lovable's 170 exposed apps — as evidence that fast AI-assisted shipping without engineering discipline creates serious privacy and security failures.
Real practitioners tested two frontier models on product and design work and got genuinely split results.
Every's team ran head-to-head comparisons of Claude Opus 5.5 and OpenAI's Sol (GPT-6) on coding and visual/design tasks. Opus 5.5 matched or beat competitors on product and design work and finished a webcam heart rate monitor Sol couldn't complete; Sol was faster and cheaper for everyday use. The piece includes concrete task examples and honest assessments of where each model fell short.
A designer walks through exactly when to hand tasks to AI and when to keep judgment for themselves.
A designer at a fictional outdoor app (Offtrail) works through real decisions about delegating to Figma's AI agent: cleaning up stale components, drafting design specs. The piece centers on a practical question — what needs human judgment vs. what just needs doing — and traces the workflow step by step rather than pitching the tool.
Why conversational AI interfaces strip away the affordances users need to actually trust a system.
Traditional GUIs give users visible cues about what's possible; conversational interfaces (LUIs) replace that with a blank text box, leaving users to guess at system capabilities, retained context, and failure modes. The piece argues these trust deficits are structural, not fixable by better copy, and draws on long-standing principles of human-centered automation to make the case.
Concrete look at five prompt-driven workflows that connect brand design systems to AI-generated campaign assets.
Figma's team documents five real creative workflows built on Figma Weave, showing how brand designers can use prompts anchored to an existing design system to generate scalable campaign visuals. The focus is on the logic behind prompt construction and how AI fits into, rather than replaces, established design language. Practical rather than promotional.
A day-by-day account of one designer actually integrating AI without losing their design judgment.
A practitioner walks through a full week of AI-augmented design work, covering how AI tools fit into a real UX process. The piece is grounded in daily practice rather than theory, showing where AI helped and where human judgment remained essential. It also touches on the Core Model methodology as a contrast to AI-first approaches.
A practitioner rethinks how framing, context, and delegation shape what AI actually produces.
The author examines how designers and thinkers like Amanda Askell and Andrej Karpathy use AI differently — not just as a chat tool but as a collaborator within broader workflows. The key insight: the relationship around the model (framing, context preservation, which actions you delegate) matters more than the model itself.
Senior-only hiring is quietly hollowing out design teams, and AI tools won't fix that.
Patrick Neeman argues that the UX field is cannibalizing its own future by refusing to hire and develop junior designers. He contends that AI tools actually favor juniors who can learn workflows from scratch, and that leaders who only hire senior talent are building brittle teams that can't sustain institutional knowledge.
AI makes designers faster, but the Jevons paradox means that speed may not mean less work.
The piece applies the Jevons paradox — where efficiency gains increase total consumption — to design in the AI era. It argues that as AI lowers the cost of producing design output, demand for design scales up rather than freeing designers from labor, raising questions about strategic positioning and what skills remain irreplaceable.
Practitioners compare AI coding agents head-to-head with refreshingly blunt, firsthand opinions.
The piece collects real quotes from builders actively using Claude Code, Codex, and Opus for production work, comparing their personalities and reliability. One GM's analogy — Codex as the grumpy-but-effective senior engineer, Opus as the fun-but-slow one — captures genuine tradeoffs. It also covers Claude Code onboarding for absolute beginners.
Sixty years of UX history reframed around a single shift: designers now define rules, not screens.
The article traces five distinct eras of UX practice and distills the durable lesson each left behind. Its core argument is that AI shifts designer leverage upstream — to information architecture, constraints, and system logic — rather than eliminating the role. Concrete historical comparisons ground what could otherwise be vague futurism.
A practicing designer's honest guide to rescuing AI-generated UI from looking generic and lifeless.
The piece breaks AI-assisted design output into three quality tiers and gives concrete techniques for moving beyond default AI aesthetics. It argues that vibe-coders especially need design literacy to avoid 'sloppy' results, and outlines what 'real design' intervention looks like at each level of effort.
Five working writers, including designer Maggie Appleton, get specific about where AI earns its place.
Journalist Laura Entis interviews five professional writers — including Every CEO Dan Shipper, designer Maggie Appleton, and NYT alum Kevin Roose — about their honest AI workflows. The piece surfaces concrete stances on where AI helps, where it doesn't, and what gets lost when writers outsource too much of the process.
A real team replaced annotation sprints with Figma MCP + AI code gen. Here's what actually changed.
Adyen's design team built a workflow where a designer works in Figma, an AI reads the frames and generates design-system-accurate code, and engineers review before pushing to production. The result eliminated annotation sprints and alignment meetings. The piece details how Figma's MCP integration and a local sandbox pre-loaded with design tokens made this possible.
A sharp reframe: treat reading and writing as UX problems, and AI starts making more sense.
The author argues that reading and writing are interfaces with friction points, and that AI tools are better understood as friction reducers than content generators. The piece pushes back on blanket AI-slop criticism by pointing out that shallow, derivative content predates LLMs, and challenges designers to think more carefully about what good human output actually means.
Practitioners ran Opus 5.5 for seven days across real coding, design, and writing tasks and reported exactly what broke.
A team of designers and builders tested Claude Opus 5.5 in Claude Code and the desktop app for a week, comparing it head-to-head against Codex across coding, design, writing, and consulting. They found genuine improvements in readability but flagged real failure modes — like the model running out of time preparing materials instead of completing the actual task. Honest, specific, and grounded in daily use.
A sharp look at why most AI-generated UI falls apart once you try to ship it.
The piece argues that AI tools generate UI in ways that fundamentally mismatch how real products are built and iterated — making output useful for demos but fragile past the proof-of-concept stage. It outlines what would need to change for generated UI to hold up through real product development cycles, offering a concrete framing for designers evaluating these tools.
NN/g frames the shift from static design systems to real-time AI-generated, user-specific interfaces clearly.
Kate Moran and Sarah Gibbons explain Generative UI as a move from designing templates to designing outcomes, where interfaces are assembled at runtime for individual users. The piece covers adaptive design patterns, what UX practitioners need to rethink, and how outcome-oriented design changes the role of the designer. Research-backed and methodical.
Concrete look at how generating a dozen design directions in one session actually changes what a designer does all day.
Outlines five Claude-powered workflows for product designers, focusing on how rapid ideation changes the nature of design work rather than just speeding it up. The core argument is that when options are cheap, selection and testing judgment become the primary skill, and promising directions can be clickable prototypes fast enough to test before a team commits.
Designers who once heard 'no' from engineers can now ship a rough feature prototype by Monday. Here's what that costs.
Vibe coding has shifted power dynamics on product teams: designers can now prototype features that engineers previously blocked as too complex or not worth building. The piece examines the implications of this shift, including the quality tradeoffs when AI-generated code is 'good enough to make a point' but not production-ready, and what it means for cross-functional trust.
A designer-writer reckons honestly with AI's pull on craft, creative shortcuts, and what writing is actually for.
A designer who also writes reflects on the seductive ease of AI-assisted writing versus the discipline of doing it yourself. The piece references UX Collective co-founder Fabricio Teixeira's thinking on design writing and pushes back on frictionless AI output, arguing that the struggle of writing is inseparable from the thinking it produces.
Naveen Selvadurai argues that AI-coded apps routinely skip information architecture, creating brittle structures that collapse when bugs appear. Drawing on his own experience cleaning up vibe-coded codebases, he concludes it's often faster to rebuild from scratch than to patch the underlying structural issues — a pointed critique of the "just ship it" AI-coding ethos.
A working designer's two-tool AI workflow, plus why AI hasn't killed professional design yet.
Every senior designer Daniel Rodrigues shares his concrete two-tool AI workflow for producing precise, visually strong output. The piece also features a conversation with Figma's Matt Colyer on why professional design services have proven more resilient to AI displacement than predicted, grounding the optimism with real-world friction.
A rigorous ethical framework for deciding when AI-generated design crosses from efficient to unjust.
Rather than asking whether AI-generated design is ethical in the abstract, the author splits the question into two concrete tests: Was automation justified given available alternatives and resources? And how were the consequences — efficiency gains and harms — distributed across stakeholders? It's a disciplined framework that moves well past the usual hand-wringing.
A real 25-person company shares exactly which AI agents they built and how they structured each one.
Every's team describes building custom AI agents tailored to their specific workflows rather than adopting off-the-shelf templates. The core advice: use Notion AI to interview yourself about your problem first, then build the agent around those answers. Includes a reusable starter prompt and a breakdown of the four agents powering their operations.
Practical starting guide for designers and PMs who want to build working prototypes without waiting on engineers.
Explains how designers and PMs can use AI coding tools by describing what they want in plain language to generate testable prototypes. Covers when the approach makes sense, which tools to use, and how to iterate without writing code. Aimed squarely at non-engineers who have ideas stuck in the backlog.
Rounds up concrete examples of AI-generated UI in production, with real patterns worth stealing or avoiding.
Surveys how leading companies are integrating AI into their product interfaces, including generated UI patterns and lessons from building ChatGPT-style apps. Frames AI as "software as clay" — something to be shaped through many iterations rather than a one-shot output. More case-study grounded than typical trend roundups.
A practical guide for designers who want to write and ship code without a traditional engineering background.
Covers how designers can use Claude Code as an agentic coding tool to prototype and build functional interfaces directly. Touches on real workflow changes, including how about 41% of product teams say AI has shifted how the whole team collaborates, and positions Claude Code as a bridge between design intent and working software.