Lukas Oppermann

LLMs fabricate tokens and forget between sessions, so prototypes drift. Hardik Pandya’s method makes design systems LLM‑readable: spec files, closed token layer, CI audit, drift detection. @Atlassian trial: 0 hardcoded values, 64 specs, consistent AI output. #DesignSystems #AI #UX → buff.ly/wKfYSwL

Sound can make web UI feel faster by confirming actions and signalling state changes before visuals catch up. The useful lesson for #UX is not “add audio everywhere,” but design subtle, optional cues that complement visual feedback. → buff.ly/CXTfYsp

Bill Gates warns the #AI transition could be history’s most turbulent decade: permanent job loss, amplified cyber/bio threats & weakened human relationships. He pushes for early #policy, “Human Reserved” domains & taxes on #automation to fund safety nets → buff.ly/u9np0mO

An interesting take on the #AI hype by @dancohen.org. His point is simple: the big question is not whether a machine is “intelligent,” but what role it plays in human work and life, just as minicomputers once reshaped offices. →

AI coding has made speed less scarce, but #alignment is still the hard part. @chriscoyier.net argues the bottleneck is getting humans, tools, and goals lined up, which matters for #AI work and for teams trying to ship together. → buff.ly/5otrIFk

Readymag rebuilt its web text editor after a 2013 legacy widget and a 2020 Draft.js stack hit limits. The #webdev takeaway is practical: browser APIs can handle more than many teams expect, and native-feeling text editing starts with fewer surprises for users. →

According to @microsoft.com #design is not dying, it is shifting left. The important decisions move closer to data & logic. #AI driven faster prototyping helps, but the bigger shift is bringing human signals into the system early →

Interesting @figma.com #plugin by @nathanacurtis.bsky.social: Spec-driven UI, using spec as a hub that normalizes & synchronizes design intent across Figma, prototypes, code libraries & docs. #designsystems #UX Maybe we can get a spec standard as they are super helpful with #AI → buff.ly/5vPqPUe

#ContextEngineering decides what #AI knows, how it remembers & where its boundaries sit, before users interact. @mcdeandres.bsky.social calls it “IA for machines” & asks #UX & #Design to work on system prompts, memory architecture, guardrails & persona. → buff.ly/6ISIB2Q

Colour drives branding, hierarchy, theming and accessibility, but workflows break because tools differ. Hex in Figma, RGB in one tool, HSL in another. #DTCG fixes that by treating colour as typed, machine-readable tokens, not loose string. Read @sturobson on #DesignSystems. → buff.ly/hqzehLG

Modernizing legacy UX without chaos using Strangler Fig pattern, proxy layer, shadow reads. @vitalyf explains how to start with business-critical slices, write tests before code & keep #UX continuity with progressive disclosure & optional old/new switch. #DesignSystems → buff.ly/rvN3NER

Most teams ship a component lib, not a #DesignSystem. Matt shows #AI agents with same tokens & shadcn/ui still producing different settings pages. The fix wasn’t more components, but a named vocabulary & composition rules. #UX → buff.ly/jMP2lpx

The Great #AI Transition: knowledge goes public, products → APIs, agents choose services & firms turn into AI‑run operation graphs. Work shifts corporate jobs → individual value creation via daemons & matching according to @danielmiessler.bsky.social → buff.ly/yIqM5LB 🫠

Matthew Butterick @mxb argues: The biggest AI risk isn't Skynet, but AI accelerating #Capitalism. Corporations replace labor, promise "goodies" like cheap goods in return. #AI makes citizens dependent on @OpenAI & co., not the state. This is #Technofeudalism in its purest form → buff.ly/d0m45CR

AI accelerates your process—good or bad #A11y. Out-of-the-box agents: 8–25% of automatable checks pass. With skills + deterministic tests: ~86%. Shift-left saves 10×–1,000× cost. #AI #UX #Accessibility → buff.ly/0ury8wO

When #AI can turn almost any idea into working software, production stops being the main constraint. This essay says #taste becomes a scarce skill: judging what deserves to exist, learning through failure, and asking @OpenAI tools for less, not more, when ideas can be weak. → buff.ly/prJvTFV

A variable font that turns text into chart: buff.ly/0wKUdeO Pretty cool idea allowing you to easily integrate data into text-only UIs. I would love to see this enabled in more places. Give it a spin!

Federated #DesignSystems are the least favoured model, used by only 13% of teams per @zeroheight's report. As @amyhupe.bsky.social notes, systems reinforce power gaps. Layoffs and #AI hype push orgs toward federation, but diffused ownership means no one really maintains it. →

Great if you assume your #designSystem is clear because humans can find the rules. Rachel Cantor argues agents may read a Storybook manifest, not the full docs, and that manifest can drop intent. #storybook’s ecosystem makes that gap easier to inspect. →

Most #AI skills fail because they rely on prompts alone. @pbakaus breaks down 9 techniques like sub-agent debate, divergence, hooks, and memory that enforce behavior and beat model defaults. The goal is turning prompts into reliable harness extensions across tools and models → buff.ly/XG5i88d

#AI did not remove effort, it moved the work. Less time “doing”, more time: framing problems, setting constraints, judging edge cases, stitching AI output into messy systems. Dashboards are green, headcount is down, but people doing orchestration & review feel burned out → buff.ly/Gc9QpVe

#AI makes implementation cheap, but understanding systems remains slow. Huge AI-generated PRs can hide weak architecture, unclear ownership, and technical debt. The useful skill is judgment: use AI to build faster, not to avoid understanding the #software. @fhh → buff.ly/GQdq9Jn

#AI prototyping is not just about speed but also about control. UI-first tools are great for fast visual exploration. Code-gen tools are great for functional prototypes, but when you need serious UX validation, direct control over interaction logic becomes the real differentiator. → buff.ly/qh6EumV

Does #AI make work harder? Berkeley study: AI increased confidence → people take on more tasks & managed parallel threads Result: fewer pauses, broader job scope & a workday that feels more continuous, not less demanding: faster task completion → higher expectations & more work buff.ly/PCxK6u1

#AI mostly solved time & attention. The bottleneck is judgment. #Agents handle tedious migration work, but humans must decide what is maintainable, secure, reliable & robust when key people leave. Next: expert context → reusable skills Once judgment is encoded, creativity is next buff.ly/PZlw7h6

#DesignSystems are not replaced by #AI, they become more important. v0 now learns your real system from code, docs, and apps, then validates every build. That means no fake tokens or drift. Teams get consistent UI grounded in truth, not approximation via @vercel tools. →

AI won’t fix a broken workflow, it will speed it up. Aaron Gustafson’s talk says accessibility has to be built in from planning to release, or #AI will scale the same barriers teams already ship. Good reminder for product and #design teams @Microsoft #accessibility →

Polar’s Orbit is a design system built for #LLM-generated UI. It shifts styling from open-ended strings to typed tokens, removes raw layout escape hatches, and relies on CI checks so consistency is enforced by code, not by memory. →

#UI quality is not just start and end states. This post says every frame should be understandable, because users read polish as a signal of care. That means no flashes, no partial loads, no inconsistent copy, and no weird mid-animation states. @nikitonsky →

#AI is killing the #design model. Research → Design → Build where humans manually made the thing. Now AI gen changes the real skill to taste, judgment & intent. The new role is defining goals, constraints, architecture & codified taste to avoid generic AI slop. → buff.ly/UxC32gM

Overwhelmed by AI hype, do this: 1. Don't read predictions, read field reports only 2. For each, ask 3 questions: - new outcomes - actions you can take in response - value of those outcomes to you 3. Stay outcome oriented: what result are you chasing? → buff.ly/CI86skm

#UX #research: statistically significant ≠ worth caring about It means: "is this effect probably real?" Practical significance asks: "is it big enough to matter for users or the business?" Always pair stats with: would users notice, and does this move metrics at our scale. → buff.ly/dx8Lo6o

#AI #UX feels off b/c they ship raw capabilities, not rethought interactions. Chat is overused, moving friction from builders to users. Fix: right modality per task (not always chat), less prompting more smart defaults, design for tiny trustworthy wins, UX is about how it feels. buff.ly/hDkqS34

Remember when Mac & iOS apps looked alive? @lmnt.me argues that flat minimalism made everything fell samey. Harder to see what's clickable, harder to spot quality craft, harder for users to understand the UI. It's not just style, we lost a shared visual language for software wdyt? buff.ly/SD2pC5K

#Slots available in @Figma.com now, a great time to revisit your #designSystem buff.ly/sEiHnq0 Slots help avoid users breaking components, and make maintaining some components a lot easier. Instead of adding lists of 20 items, with 15 of them hidden, just use a slot, which lets you add items.

Design shifts: applied craft → liberal art of technological culture, fewer folks designing, more becoming design literate: can read, analyze & critique UIs. It's not just vibes, buff.ly/5kK2VVe says we are mid “a design turn.” It reframes the job: being thoughtfully critical of what gets built.

Love this idea of using #AI: buff.ly/5u6PIrS Don't let AI jump straight into implementation, it should first do structured planning with you: 1. Capabilities 2. Components 3. Interactions 4. Contracts This way AI thinks about those steps & you can influence core concepts before implementation.

Why does claude pick certain products? → bit.ly/3ONQUIH I guess it's good it does not recommend paid products, I am sure this will come, the internet is all about making money. 😑 Are high pick rate tools is better? Not necessarily. Pick rate reflects how often it appears in training data.

What impact does #AI have on the well-being of employees? This post says, it may lead to burn out and more: buff.ly/wkEfrrY The pressure to deliver more and more is mounting. And using AI does not increase the speed of every task equally. Especially when it matters to get it right.

Great notes by @addyosmani.bsky.social on writing #Agent #specs: buff.ly/lvRn1jp - Start high-level - Use PRD structure - Cover essentials: Commands, testing, do's and don'ts - plan/refine spec before execution - sub-agents for specialized sections (e.g. DB) - Add self-checks

Will the #AI investment bubble burst? @hughhowey.com thinks so: buff.ly/Vym79ux Why? - 80% of 6,000 firms show zero productivity gains - First mover weak: OpenAI unlikely to dominate as Google/Amazon build own models - Chinese free open models like DeepSeek match paid ones Wdyt?

#Agents ship frontends without humans in the loop, a11y tooling needs to catch up. Agents need: structured violations + rule IDs, selectors, fix ops, fixability, context, guidance & diff based verification. more about #a11y tooling for agents by @ckundo.bsky.social: buff.ly/8YFIjpe

Some interesting thoughts on #Agent design patterns buff.ly/Bj6uP1G - combat finite tokens by allowing agent to write files - avoid tool dumps instead provide tools via mcp or file system when needed - use usb-agents to isolate context and share via summary files

Writing code is mostly obsolete according to @antirez.bsky.social. You can focus on higher problems and let #AI fix your tests or migrate to that new software, saving you hours if not weeks. This also allows small teams now punch like big ones, just faster → antirez.com/news/158

We talk about “inevitable” AI disruption, but this is a story told by people who profit from chaos. There are possible 3 futures: A) rapid automation, mass unemployment, social shock B) slower shift, painful but manageable adaptation C) Amish style discernment 🔗 buff.ly/r3OXPVn

A #designSystem functions like an immune system, allowing broad participation in shipping UI while keeping the overall experience coherent. DS ROI: directly improves AI output quality, no more messy UI but on-brand, accessible, production-ready interfaces 🔗 buff.ly/MnylkxK

Why do we even pay for software in a world of #AI. It's not because we couldn't build it, but because we don't want to own it. Owning a software solution comes at a constant cost, which distracts you from focussing on what you actually sell. 🔗 buff.ly/DIatG12 by @quinnkeast.com

#DesignSystems don't fail on components, they fail on enforcement according to @nngroupux.bsky.social A DS enforcer – Makes teams use the system – Folds good exceptions back in – Mediates conflicting product asks Goal: slightly messy, widely adopted system solving real problems 🔗 buff.ly/9vKJWRV

Should AI-era UX skips wireframes for data + context + evals? Analytics reveal friction; user stories the "why"; automated tests validate: buff.ly/TYj7PFZ Does it work? Can AI understand the friction without human intuition? Will it create the best or most common flow? What is your experience?

Saying No in the age of AI can't be argued with time constraints, so is it still valuable? @jim-nielsen.com says yes: buff.ly/EbvVXlY Saying no may matter more than ever! To build good products we need to keep in mind the users attention constraints. More is often not better, focused products win.