UI/UX in 2024: The Good, The Bad, And The AI-Driven (2026 Recap)

Why Fora Soft wrote this UI/UX retrospective

Fora Soft has been shipping software products for almost two decades, and design has been a core part of every project we deliver. We do not run a separate “design studio” team that hands wireframes to engineering. Designers, product managers, and engineers sit on the same squad and own the same outcome — that is the only way to land an AI-driven UI on time.

This article is the playbook we use internally to brief new product owners on what changed in UI/UX during 2024 and how that has shaped 2025–26. It is written for CTOs, product directors, and founders who are deciding what to build, what to retire, and which design/engineering roles to hire next. We will walk through the wins, the loud failures, the platform shifts, and the practical commercial implications for anyone scoping a custom software product.

Our team uses agent-assisted engineering — in-house Claude- and Cursor-based pipelines — to compress the commodity work in any product build, including the design system, the component library, the empty states, and the boilerplate flows. That keeps the budget on the parts of the UI that matter: the AI affordances, the workflow that the customer pays for, and the polish that takes the product from “works” to “loved.” Our 2026 quotes therefore run noticeably tighter than typical 2024–2025 industry ranges.

The end of the InVision era and the platform consolidation

InVision shut down on December 31, 2024 after losing its place as the prototyping standard. By that point Figma had already taken over — and the consolidation went much further during 2025.

Figma is the new design substrate. Figma Make and Figma Sites turned the canvas into a deployable product platform. Pair that with Figma AI for component scaffolding and Figma Slides for design system documentation, and most B2B SaaS teams now ship from Figma all the way to production-grade prototypes. Adobe’s blocked acquisition in 2023 turned out to be the moment Figma stopped being a tool and started being a platform.

Adobe XD is functionally retired. Adobe stopped active development; XD users have largely migrated to Figma or to Adobe Express for marketing pages. Sketch has retreated to indie Mac shops. Penpot is the credible open-source alternative for self-hosted teams.

The handoff stack changed. Zeplin and Avocode lost ground because Figma Dev Mode plus the design-token export ecosystem covers most of what they did. Storybook is still the bridge to engineering. The new layer underneath is V0 by Vercel, Cursor, Lovable, and similar tools that turn a Figma frame into shippable React directly.

AI design tools that actually shipped in 2024–26

2024 was the first year AI design tooling stopped being demoware and started being a daily-use part of the design pipeline. Two trends matter for product teams.

Generation moved from screen mockups to working components. V0 (Vercel), Lovable, Bolt, and Cursor turn a prompt or a Figma frame into shippable Tailwind/React code. Useful for scaffolding, design-system stress-testing, and rapid prototyping. Not yet a substitute for senior design judgment — but it is a substitute for week three of a junior designer iterating on a sign-up form.

Personalisation moved to runtime. Apple Intelligence, Galaxy AI, and Samsung One UI 7 made on-device personalisation a platform feature: writing tone adjustment, photo cleanup, content summarisation, and contextual replies happen in the OS, not in your app. The implication for product teams: stop building features that the OS now does for free, and start building affordances that benefit from those OS features.

Generative UI is the next frontier. Anthropic’s Artifacts, OpenAI’s Canvas, and Google’s in-Gemini canvases prove the pattern: the AI does not return a paragraph, it returns a small interactive UI fragment that the user can manipulate. By 2026 mature B2B products are starting to ship dynamic UI surfaces — tables, charts, forms — assembled at runtime from an LLM plus a strict component library.

The good: redesigns that actually worked

Five product redesigns from 2024 are worth studying because they shipped genuine improvements without alienating users.

1. Samsung One UI 7. A complete refresh of Samsung’s Android skin around AI-first interactions: Now Bar, Now Brief, AI Select, Cross-app Actions. The redesign delivered on a clear job: turn the lock screen and the home screen into surfaces where the OS does work for the user. Reception was overwhelmingly positive on review sites and forums.

2. Linear’s 2024 platform refresh. Cycle planning, sub-issues, customer requests, and AI-suggested triage shipped in coordinated waves rather than one big-bang release. Each shipped feature respected the muscle memory of existing power users. Linear retained > 90% of its high-NPS user base while expanding into PM-track work.

3. Notion AI integrated as an inline citizen. Rather than a sidebar bot, Notion AI lives inside any block. Ask it to summarize, rewrite, or translate without leaving the page. The pattern (“AI is a verb on the existing surface”) is now a default for productivity products.

4. Stripe’s onboarding revamp. Stripe rebuilt their merchant onboarding around progressive verification, with the AI-driven KYC/KYB layers explicit in the UI. Time-to-first-payment dropped measurably for new accounts. The lesson: AI explanations belong in the foreground when the user is being asked to trust an automated decision.

5. Apple’s Vision Pro spatial UI patterns. Even with limited consumer adoption, the visionOS interaction grammar — pinch, gaze, tap-with-eyes — is now the reference for any spatial software project. Mature spatial UIs are still mostly enterprise (training, surgery prep, design review) but the patterns set in 2024 are sticking.

The bad: redesigns that lost users or revenue

2024 also produced the most public, most expensive UI controversies of recent years. The pattern is consistent: redesigns that prioritized aesthetics, growth metrics, or platform politics over the user’s primary task.

Apple Photos (iOS 18) ran into a wall of complaints. The single-page library reorganized core actions, removed the bottom navigation users had built muscle memory around, and surfaced AI-generated “Memories” ahead of recent photos. App Store reviews and Twitter feedback were unusually negative for an Apple stock app. Apple iterated through 18.1 and 18.2 to claw back the prior layout. The lesson: do not trade memorisation cost for AI promotion in a stock app that 1B users already know how to use.

Spotify’s home-screen and DJ pivot annoyed power users. The redesign pushed AI-generated playlists, podcasts, and short-form video higher in the discovery hierarchy at the cost of user-curated playlists. Power users complained loudly; subscriber growth slowed at exactly the time Spotify needed it to accelerate. The lesson: never demote a feature that drives retention in service of one that drives engagement minutes.

Sonos’s app rewrite became a textbook bad rollout. The new Sonos app launched with broken core features (alarm, sleep timer, queue management). Customer trust collapsed, the CEO was eventually replaced, and the company refunded long-time customers. The lesson: a rewrite that ships without parity on the user’s primary task fails, regardless of how clean the new architecture is.

Nike’s $25B revenue blunder. A redirect of digital strategy away from data-driven personalization toward brand-led top-of-funnel experiences cost Nike an estimated $25B in market cap by mid-2024. The cautionary tale here is not a UI choice; it is a UX strategy choice that downplayed the data layer behind the UI. We are still in the recovery phase as Nike reverses course.

Reddit’s API turmoil. The 2023 API pricing change rolled into 2024 with the third-party-app shutdowns, mod tooling churn, and a search-redesign that made the official mobile app harder to use for power moderators. Reddit went public anyway, but the UX debt became visible in user retention reports.

Generative UI: the new pattern to learn now

The most consequential UI shift starting in 2024 is generative UI — interfaces that are assembled at runtime by an LLM from a strict component library. Three things make this real now.

The component library is the contract. The LLM does not invent buttons. It picks from a typed registry of components your team owns: DataTable, SummaryCard, Chart, FormField, FilterBar. Each component declares its props in JSON Schema. The LLM emits a JSON blueprint; your renderer maps it to React.

The state model is server-side. The LLM is stateless and the UI fragment is ephemeral. The data behind it lives in a server-side store that gets revalidated as the user interacts. This pattern is what Anthropic Artifacts, ChatGPT Canvas, and Vercel’s AI SDK Generative UI all converged on by mid-2025.

Design becomes inventory management. Designers no longer hand off “the screen for the dashboard.” They hand off the catalogue of components and the prompt patterns that compose them. The role rebalances toward design systems and information architecture — a major reason senior designers are getting paid more in 2026 than in 2022.

The rise of the design engineer

The most important hiring shift of 2024–25 is the design engineer — a person who can write production React, owns the component library, and contributes to design tokens at the same level as a senior designer. Linear popularised the title; Vercel, Stripe, GitHub, and most well-funded SaaS companies now hire for it explicitly.

Why the role is needed. Generative UI, AI-assisted code, and design tokens make the boundary between “design” and “engineering” less useful. A design engineer ships the component, the variants, the dark/light themes, and the keyboard/a11y handling in one PR. Time from Figma frame to shipped component drops from weeks to hours.

What it means for hiring. If you have one design seat to fill on a small team, hire a design engineer over a pure visual designer. They will land more value because the visual designer’s output gets bottlenecked at the engineering handoff regardless of quality.

Five design roles that became real in 2024–26

Job titles in design teams shifted faster in the last 24 months than in the previous decade. Five roles you should know if you are hiring or being hired.

1. AI Interaction Designer. Owns the conversational and generative-UI surfaces, the prompt templates, the failure modes (refusals, hallucinations, low-confidence), and the “AI explanation” layer. Sits between PM, design, and applied science.

2. Prompt Designer. Treats prompts as a first-class artifact: versioned, tested, monitored, A/B-tested in production. Often a senior writer who learned engineering, not the other way around.

3. Design Engineer. Owns the component library and the bridge from Figma to React. Already covered above — the most consequential of the five new roles.

4. Service Designer. Maps the cross-channel customer journey (app, email, support, in-store, callbacks) and owns the experience as a whole rather than any single screen. Especially valuable in regulated verticals (healthcare, finance, government).

5. Spatial Designer. 3D affordances, gaze and pinch interactions, anchored content for AR, room-scale environments for VR. Still niche in 2026, but real budgets are now attached on enterprise training and design-review software.

Ethics, dark patterns, and growth design after 2024

2024 was also the year regulators and the design community pushed back on growth-led design choices. Two pieces of news set the tone.

The FTC’s Click-to-Cancel rule made it explicit: cancellation must be at least as easy as sign-up. Subscription products with three-step sign-up and seven-step cancel are now legally exposed in the US. EU’s Digital Services Act and Digital Markets Act add similar requirements. Design teams that built dark patterns into the cancellation flow have been quietly redesigning them all year.

Apple App Tracking Transparency completed its long arc. The opt-out default reshaped the mobile-ads market, and by 2025 Meta and other platforms had fully adapted with on-device modeling and privacy-preserving signals. The implication for product teams is that growth-design playbooks built on cross-app tracking are mostly dead. Build the loop on first-party data.

The community pushed back too. “Growth-design ethics” was the most contested topic at IxDA, Config 2024, and the Nielsen Norman Group’s 2024–25 reports. The pendulum is swinging back from growth-loop optimisation toward retention and trust. Pricing transparency, deliberately non-clickable confirmation steps, and explicit AI disclosures became the new normal in well-run products.

Tools comparison: 8 platforms shaping product UI in 2026

If you are setting up a new design+engineering pipeline in 2026, these eight tools are the ones we see most often in well-run product teams.

Pick the smallest pipeline that closes the loop from idea to shipped component. For most product teams in 2026 that is Figma → V0 (or Cursor) → Storybook → production React, with Maze and Dovetail running in parallel as the research cadence.

AI beyond chatbots: agents, copilots, and ambient UI

By 2026 the “chatbot” is the least interesting AI surface. Three patterns are taking over.

Inline AI. The AI is a verb on the existing surface. Highlight a paragraph and rewrite it. Right-click a chart and ask it to explain. Tap a row in a table and request a summary. No modal, no chatbot — the AI is part of the toolbar. Notion, Apple, Adobe, Microsoft 365, and Google Workspace all converged here.

Agents and copilots. Multi-step automations that can read your data, take actions, and surface a UI for human-in-the-loop confirmation. Cursor for code, Claude in Chrome for browsing, and the new wave of vertical agents (sales SDRs, finance, ops) all share this pattern. The UI challenge is showing what the agent has done, what it will do, and where the user can intervene.

Ambient AI. The OS does work in the background — smart replies, intent prediction, photo cleanup, voice intent recognition — without a clear surface. Apple Intelligence and Galaxy AI both did this. Implication for product teams: pick the part of your product that benefits most from ambient OS features and design around it instead of duplicating it.

Voice is back. LLMs make voice useful again, especially for hands-busy and accessibility-driven workflows. Real-time TTS/STT pipelines from OpenAI, ElevenLabs, and Deepgram are now production-grade. We use these in cloud video projects for transcription and live captioning, and in smart intercom apps for voice-driven door control.

Practical implications for product owners scoping a build

Five takeaways if you are about to brief a design/engineering team in 2026.

1. Budget for the design engineer first. If you have one design seat, hire someone who can also commit production code. Pure visual designers are still valuable, but they get bottlenecked at handoff in any team smaller than 20.

2. Decide on AI surfaces in v1. Not all of them — one or two. Inline AI on the primary surface, an explainability panel for any AI-driven decision, and a clear opt-out for users who do not want personalisation. Defer agents and ambient AI to v2 unless they are core to the value proposition.

3. Pick the design system before the screens. Tailwind + shadcn/ui is the de-facto B2B SaaS default in 2026. Material 3 for native Android, SwiftUI for native iOS. Custom design systems pay back only at > 50 engineers. Below that, a forked shadcn library with your tokens beats every alternative.

4. Plan the regulatory work. Click-to-Cancel, App Tracking Transparency, GDPR, BIPA, ADA. Add four weeks of design+legal review per release if you serve regulated buyers. Underestimating this kills product launch dates more often than any technical risk.

5. Re-test the obvious. Your sign-up flow, your settings page, your cancellation flow, and your onboarding have probably regressed since 2022. Run a usability test on the first three weeks of a new user’s journey at least once a year. The teams that win in 2026 take the obvious work seriously.

Mini case: shipping AI-driven product UX with a small team

Situation. A B2B media-research client needed a v1 product that would put AI-summarised interview insights in front of brand teams within ten weeks. The team was three engineers and one designer. The risk was that the design+engineering loop would consume the budget before the AI work landed.

What we shipped. A Figma design system that mirrored shadcn/ui exactly so the designer could prototype in Figma while engineering shipped real components. V0 generated the first pass on every new screen; the designer refined and committed back to Figma; the engineer wired the data layer and the LLM calls. Inline AI lived on the existing transcript surface (highlight, summarise, ask). No chatbot, no modal.

Outcome. The product shipped on day 71 with the AI surfaces working in production. The designer+engineer loop closed in a single PR per feature instead of three. The customer’s NPS at week 12 was 67. We documented the pattern internally as a template for any small-team AI-product build.

Cost model: realistic UI/UX budgets in 2026

Conservative ranges for a senior agency in 2026 using agent-assisted engineering. UI/UX cost is a portion of the total project budget, never a separate line.

These ranges assume an existing product team. UI/UX as part of a full product build is roughly 20–30% of the total project budget — less if you start from a strong design system, more if AI affordances are a major surface.

A decision framework — pick your UI/UX path in five questions

Five questions on a sheet of paper before any design call.

Q1. What is the user’s primary task on this surface? Write it down in one sentence. Every UI choice flows from this. If you cannot answer it, you do not have a redesign brief — you have a Pinterest board.

Q2. Where does the AI live in this product? Inline, agent, ambient, none. Pick deliberately. Defaulting to a chatbot in 2026 is design malpractice for most B2B products.

Q3. Native, web, or both? Most modern B2B SaaS ships web first and native later (or never). Native pays back when the workflow is genuinely mobile or platform-specific. Be honest.

Q4. Custom design system or shadcn+tokens? Below 50 engineers, customise shadcn. Above 50, consider your own. Custom-from-scratch in a small team is a way to spend the design budget on plumbing instead of users.

Q5. What dark patterns will the new design remove? Each redesign should pay down at least one ethical debt — a hidden cancellation, a default opt-in, a confusing pricing display. Make this explicit in the brief.

Pitfalls to avoid in a 2026 redesign

Five places redesign projects keep burning weeks. None are exotic; all are easy to skip in a SOW.

1. Redesigning aesthetics without a job-to-be-done. The Apple Photos and Spotify lessons. If the new design does not pay back a measurable user task, do not ship it. A/B-test the change against the existing version before global rollout.

2. Shipping a rewrite without parity. The Sonos lesson. Map every existing feature, every shortcut, every accessibility affordance into the new system. Ship the rewrite when the parity is > 95%, not when the architecture is clean.

3. Demoting power-user features. Power users drive your retention numbers more than your acquisition team realises. Never push a power feature one layer deeper to surface a growth experiment. Retention loss compounds faster than acquisition gain.

4. Adding AI without explainability. Any AI-driven decision the user can see needs a one-tap explanation. Without it, trust collapses on the first wrong answer. Build the explainability layer into the design system, not as a bolt-on.

5. Skipping accessibility. ADA lawsuits keep climbing in 2026. WCAG 2.2 AA is the floor. Build keyboard navigation, focus rings, screen-reader semantics, and color contrast into every component, not as a phase-2 polish item.

KPIs that matter: Quality, Business, Reliability

Three buckets, with thresholds we use as the minimum bar to ship a UI/UX release.

Quality KPIs. Time-to-first-meaningful-action < 5 s on a new user. Task-completion rate > 85% for the primary task in usability tests. Lighthouse Performance > 90 on the top three pages. WCAG 2.2 AA conformance on every shipped component.

Business KPIs. Activation rate (users completing the primary task in the first session) > 50%, ideally trending up post-redesign. Trial-to-paid conversion not lower than baseline at week 4. NPS at week 4 and week 12 trending up. Support-ticket volume per active user trending down.

Reliability KPIs. JS error rate < 0.5% of sessions. Frontend p95 LCP < 2.5 s. AI feature latency < 2 s for inline calls, < 5 s for agent calls. Crash-free sessions > 99.5% on native (Crashlytics).

When NOT to redesign your product in 2026

Three scenarios where a redesign is the wrong answer. We tell roughly one prospect in five to skip the redesign — they almost always thank us later.

Your retention is healthy and you are not adding new surfaces. A working product with strong retention is worth defending, not refactoring. Spend the budget on adjacent features instead.

You do not have user research. A redesign without research is gambling. Run six interviews and an unmoderated usability test before scoping the rebuild.

Your team cannot ship the rewrite to parity. The Sonos lesson. If your team is not big enough to maintain feature parity through the rewrite, do an iterative refactor instead.

Your timeline is under eight weeks. A meaningful redesign of a B2B SaaS surface is 10–14 weeks at minimum. Anything shorter is a coat of paint.

FAQ

What to Read Next

Where UI/UX is heading from here

2024 was the inflection point: AI moved from a sidebar feature to a default expectation, the prototyping era ended with InVision, and the public failures of Apple Photos, Spotify, Sonos, and Nike taught the industry to respect the user’s primary task. 2026 is where the patterns become standardised — inline AI, generative UI, design engineers, ethical guard-rails. The teams that win are not the ones chasing every trend; they are the ones who pick the trends that pay back the user’s task and ship them with care.

If you are scoping a product redesign or a new build right now, the highest-leverage move this week is to write down the user’s primary task on every surface in one sentence each. Bring that list to a senior product+design call, and the rest of the playbook above will tell you which trends to keep and which to ignore.