Cuts to the real problem: AI-generated UI that can't survive past a demo.
The piece argues that most AI-generated UI fails because it is produced in a way that mismatches how real products are actually built, making it disposable after a proof-of-concept stage. It explores what design qualities and structural decisions make generated UI genuinely reusable and iterable beyond the initial demo.
A practitioner runs a real UX review through AI and reports back, honestly.
The author shares a first-person account of delegating a UX heuristic review to AI, covering which models worked best for different tasks and how to structure prompts with real data. Key takeaways include using AI as a first-pass partner, working in steps, and codifying your own severity heuristics into reusable prompts.
Concrete guide on pairing 3D modeling precision with AI's speed for visual exploration.
Designers can start with a base 3D model and use AI to rapidly explore different visual directions without losing control over the output. The article frames this as a collaborative workflow where traditional modeling precision and AI adaptability complement each other, helping designers iterate on ideas before committing to a direction.
Designers have always known things they can't fully explain. AI is forcing a reckoning with that gap.
Drawing on Donald Schön's concept of 'knowing-in-action,' the piece argues that expert design judgment is tacit and context-dependent in ways that resist codification. As AI tools accelerate execution, the author contends that the harder, underdiscussed crisis is articulating *what* to make, not *how* to make it faster.
A designer takes end-to-end ownership of craft-level changes without waiting on engineers. Here's the actual workflow.
This practitioner walkthrough shows how designers can use Figma Make with production code to own small, high-craft UI changes themselves, while engineers handle heavier lifting. It focuses on closing the gap between design and what actually ships, with a concrete split-responsibility model between designer and engineer roles.
Using reference images to steer AI output is a real skill. This breaks down how to actually do it.
Visual prompt engineering means attaching images to AI prompts to guide output style, layout, or context without writing code. The piece covers which tools support it, how to structure image-plus-text prompts effectively, and where it fits into a design workflow including Figma Make and Claude-based tooling.
A candid first-person account of what an all-AI design-to-code workflow actually feels like to live in.
The author describes a fully AI-mediated workflow — Figma AI for designs, AI for coding, AI for client emails — and grapples honestly with the fatigue and identity questions that come with it. It's a rare practitioner reflection on the psychological cost of letting AI handle the entire stack, not just productivity gains.
The real AI win for design leads isn't mockups — it's the 80% of the week that isn't design.
Drawing on the State of AI in Design Report 2025, the piece argues that AI's biggest impact for design leads is automating the administrative and process overhead that consumes most of their week. It notes that 84% of designers use AI in research and ideation but only 39% use it during delivery, and explores what that gap means for how design leadership evolves.
Cuts through the noise to show what AI actually does in a product designer's daily work.
Rather than listing tools or making broad predictions, this piece focuses on concrete integration of AI into real product design workflows. It examines where AI genuinely accelerates meaningful outcomes versus where it falls short, offering a grounded counterpoint to the hype cycle around AI-generated design.
The creative technologist archetype is bleeding into every design role, and here's the receipts.
The piece traces how the creative technologist skill set — building prototypes, writing first drafts in code, staying close to AI models — is now showing up in standard designer job descriptions. It connects several real hiring and portfolio trends to argue this is a structural role shift, not a niche specialization. Designers shipping vibe-coded work as self-expression are the early signal.
A sharp look at why Figma components and shipped code keep drifting apart, and how to fix it.
Design systems exist simultaneously in Figma and in code, but the two inevitably diverge. The author proposes using structured contracts (JSON/YAML) rather than markdown to define components, so both humans and AI agents can execute intent precisely without ambiguity. The argument centers on why labeled, machine-readable specs outlast prose documentation in AI-assisted workflows.
How to turn fuzzy creative intent into specs that AI agents can actually execute without hand-holding.
As token efficiency pressures push teams toward tighter, spec-driven workflows, the author introduces the SAID framework for translating creative ideas into precise, actionable intent. The argument is that designers need to get comfortable in Git environments and learn to write specs legible to both humans and AI agents. It's positioned as reclaiming craft, not abandoning it.
Plain-language walkthrough of prompt-driven app building for non-coders inside monday.com.
Monday Vibe lets users build functional apps by describing what they want in plain text, with AI generating the underlying code and accepting follow-up adjustments through chat — no code editor involved. The support doc walks through the step-by-step process of creating and iterating on an app entirely through natural language prompts on monday.com/vibe.
A real studio shows exactly how AI handles documentation and FRDs so designers skip the busywork.
Design studio DUOTONE describes a concrete workflow where AI drafts functional requirement documents directly from Figma specs and component data, turning visual canvas decisions into structured engineering handoff documents automatically. The approach is reported to save over 80 hours per project by offloading maintenance and documentation to AI while keeping design consistency intact.
Honest stress-test of AI prototyping tools against real design scenarios. The results are more sobering than the hype.
A hands-on evaluation of AI-powered prototyping tools using real design scenarios found that while they can follow high-level instructions, they routinely fail to weigh design tradeoffs or produce quality output without heavy human guidance. The piece argues that AI tools currently lack the sophistication to replace nuanced design judgment, even as marketing claims otherwise.
The designer who shapes workflows will outlast the one who just works faster inside them.
The author distinguishes between using AI as a productivity boost within existing workflows (faster pencil) versus redesigning the workflow itself (architect). Early signals include Figma dropping in job-posting priority while design thinking moves into broader roles. The leverage of the architect role is real but shows up in slow-moving metrics, not polished demos.
Swapping in AI tools without changing your process is just theater with a bigger budget.
The piece argues that many teams adopting tools like Claude Code or Cursor are performing innovation rather than achieving it — using a powerful reasoning engine as a glorified checklist. The core critique: the tool changes but the underlying process does not, so the gains never materialize.
Anthropic's engineers share concrete patterns that actually make agents work reliably in production.
Anthropic's engineering team details practical patterns for building reliable AI agents, including how output format choices (diffs vs. full rewrites, markdown vs. JSON) meaningfully affect LLM performance. The guide covers orchestration, tool use, and when to prefer simple pipelines over complex multi-agent setups — grounded in what works, not what's theoretically elegant.
A real team shows how AI tools reshaped discovery work, not just production speed.
Carrefour's design and product teams use Gemini and NotebookLM to process raw user research — interview transcripts, workshop notes, reports — into structured insights during the discovery phase. The case study details how this accelerates problem framing and workflow structuring without removing designers from strategic decisions.
Explains why feeding AI the right context matters more than which tool you pick.
The article explores how 'design context' — goals, constraints, audience, and system logic — is the critical input that determines AI output quality in design workflows. Drawing on Figma's 2025–2026 survey data, it argues that designers who can translate complexity into clear briefs will extract the most value from AI tools.
A practicing designer reflects honestly on where AI helps and where human judgment still wins.
The author argues that AI tools like Perplexity and ChatGPT are genuinely useful in early research stages but cannot replace the systems thinking and creative judgment at the core of design work. The piece emphasizes that surface-level design tasks are being automated, shifting where designer value actually lies.
Concrete advice on keeping quality standards intact when AI quietly erodes craft at review time.
The article warns that AI-assisted design workflows introduce subtle quality drift that only surfaces at end-of-sprint reviews. It recommends building design token structures early, treating AI output as a draft requiring active critique, and using structured evals (borrowing from LLM-as-judge research) to catch regressions before they ship.
A sharp argument that taste can't be automated — and why designers who think it can are in trouble.
The piece argues that aesthetic judgment requires genuine exposure to exceptional work, historical knowledge, and trained perception — none of which can be outsourced to AI through prompts or consensus data. It positions taste as an orchestration layer that designers must own deliberately, especially as AI handles more execution.
DesignOps is changing from policing process to coordinating humans and machines together.
The piece argues that AI is forcing DesignOps to evolve from a governance-and-standardization function into one that orchestrates collaboration between human designers and AI agents. It examines what that shift means concretely for tooling decisions, quality control, and team structure, positioning DesignOps as a strategic layer rather than a process enforcer.
Honest look at what designers actually risk when they skip the craft for a single prompt.
The article traces vibe coding back to Andrej Karpathy's original framing and examines why the full-stack-from-one-prompt approach carries real risks for designers: hidden technical debt, loss of intentional design decisions, and brittle outputs that are hard to maintain or audit. The author argues designers should be skeptical rather than swept up in the hype.
The job of a designer is now partly about writing context that AI can actually use.
NNG introduces 'UX-context design' — the practice of structuring research, standards, and design rationale so AI models can access and act on that knowledge. As AI generates more interface work, the output of design shifts from human-readable documents to curated context files. The piece gives concrete guidance on what belongs in that context and why.
A team walks through exactly how they used AI tools to ship a major product launch, tool by tool.
The Every team recounts the concrete AI-assisted workflow behind their largest product launch, including use of Cursor, Codex, and other tools across writing, coding, and coordination tasks. The conversation covers why they chose to build a new subscriber tier and how AI compressed timelines that would normally require more headcount. It's a candid, process-level account rather than a retrospective gloss.
Practitioners compare working with GPT-5.6 Sol against other models in their actual daily workflows.
Three Every writers share first-person accounts of switching to GPT-5.6 Sol for real work tasks, noting its speed and steerability as standout traits. They contrast it with Fable (their preferred model for full-handoff tasks) and describe how Sol rewired day-to-day habits around file retrieval and context management. The piece reads as a practical, comparative model review rather than a benchmark summary.
A sharp argument that AI speeds up execution but not clarity, and what that means for design tokens.
The piece introduces 'hypertokens' as a proposed layer between design tokens and components, intended to encode designer intent in a form AI agents can act on without guessing. The author critically examines whether hypertokens are meaningfully different from composite tokens and questions where a single source of truth would actually live. It's grounded in the broader claim that human-authored intent must remain legible for AI-assisted design to produce coherent output.
Workshop tackles the unglamorous problem: keeping AI-generated copy consistent and on-brand at speed.
Co-founders of a content tooling startup present a workshop on building content systems for teams that are shipping AI-assisted products faster than their style guides can keep up. The session addresses how to maintain brand consistency and compliance when every tool in the stack is generating copy simultaneously.