When I published "A Little Background About AI and Why You Shouldn't Be Afraid of It" in 2023, large‑language models still felt like experimental toys that might - or might not - fit into a professional design toolkit. At the time, I had zero opportunity to use generative AI at work; security and confidentiality rules inside corporate banking made any external tool off‑limits. My only hands‑on practice happened after hours, on side projects and hackathons. Paradoxically, that restriction turned into a gift: it forced me to explore AI on my own terms, free from legacy workflows and approval chains, so I could later implement proven tactics directly in new roles. During that period, I also completed the Interaction Design Foundation (IDF) course on AI for Designers, grounding my experiments in established UX research methods.

The AI landscape from 2023 to 2025

The biggest shift during the past two years has been speed and specificity. Second‑generation language models such as GPT‑4o (OpenAI, May 2024) and Claude 3‑Opus (Anthropic, February 2025) read images, listen to audio, and process context windows the length of small novels, all while replying in near real time. Instead of visiting a standalone chatbot, designers now meet AI at the point of work. Figma's AI beta (September 2024) turns plain‑language prompts into fully‑linked wireframes. Framer AI v2 (May 2025) can draft a production‑ready marketing site complete with React components and CMS hooks. Canva's July 2025 integration with Claude lets you draft long‑form copy and push it straight into a visual layout. Another breakout platform, Lovable (Stockholm, 2024), turns plain‑language prompts into fully‑functional web apps and internal tools, stitching UI components, database schemas, and hosting in a single chat flow. The tool rocketed to 500,000 daily active users and US$10 million ARR within two months, making it a go‑to for lightning‑fast MVPs and design‑system prototypes when engineering bandwidth is scarce. AI Job Search rewrites your résumé and surfaces roles that match your real skill graph - not just keyword soup.

How AI fits my UX workflow today

Because my corporate contracts once banned generative tools, I had to weave AI into freelance gigs, community mentorship with Women Go Tech, and, most recently, my own job search. Here's how that looks in practice:

Problem framing now begins with a dump of raw interview notes pasted into Claude 3 alongside the prompt: "Cluster these statements by Jobs‑to‑Be‑Done intent and return JSON." Ten seconds later, I have a structured backlog instead of an unreadable wall of sticky notes.

For research synthesis, ChatGPT's Code Interpreter ingests survey spreadsheets, builds pivot tables, and visualises sentiment trends - work that used to steal half a day. Concept exploration happens inside the canvas: I type "Generate three mobile onboarding flows for a savings app" in Figma AI and receive interactive drafts that jump‑start discussion with stakeholders. Micro‑copy refinement is equally tight‑loop; I highlight a paragraph, hit Rewrite, specify "friendly, B2 reading level," and compare alt‑versions in seconds. Even accessibility is faster: GPT‑4 Vision scans exported screens and flags contrast violations or missing alt‑text before the QA team sees them. When deliverables are due, Canva × Claude drafts speaker notes that mirror my voice, freeing me to polish visuals instead of staring at a blinking cursor.

From usable to lovable

AI's power is not just in automating rote tasks but in crafting moments of delight at scale. By learning from tone, context, and real‑time feedback loops, tools like GPT‑4o, Framer AI, and Adobe Firefly propose playful micro‑copy, celebratory animations, and hyper‑personalised nudges that turn a flow from merely usable to genuinely lovable. In one recent fintech side‑project, Claude suggested a confetti burst and a milestone message tailored to each user's savings goal. A/B tests lifted task‑completion by 9 % and bumped NPS six points. Next time you review an AI‑generated screen, ask yourself: Would I smile while using this? If the answer is yes, you just shipped lovable UX.

AI in the job hunt

Generative tools have also shrunk my application funnel. LinkedIn's AI Match suggests roles that align with user‑research depth and hybrid flexibility in Berlin - two filters that a basic Boolean search often mangles. A quick ChatGPT browse command - "Summarise the last three N26 earnings calls with a focus on UX risks" - produces a cheat sheet that would have taken an evening to assemble. Claude rewrites individual résumé bullets to echo the metrics in a job description, increasing recruiter response rates from roughly ten percent to more than twenty. I rehearse interviews with GPT‑4o's voice mode, receive a transcript and immediate feedback, then follow up with AI‑drafted thank‑you notes that reference each team's OKRs. The numbers are telling: over three months, I sent fourteen tailored applications, landed six recruiter calls and reached three full panel interviews. Pre‑AI, I needed roughly ten generic applications for a single screening call.

Pro tips for AI‑curious UXers

The most common question I hear from mentees is "Where do I even start?" My answer is start with the task, not the technology. Write down the objective in human language before ever touching a prompt. Work in short loops: draft, tweak, regenerate. Supply concrete examples - good and bad - to anchor the model's style. When the data is sensitive, strip personal details or spin up a local model instance. Version your prompts the way you version code, and let a second model critique the first. Above all, remember that AI amplifies good judgment; it never replaces it.

Ethics & governance

Freedom to use AI does not absolve us from responsibility. Bias hides inside both prompts and outputs, so audit regularly and test across demographics. Be transparent in design review about what was AI‑generated and invite human feedback early. And never dump raw research transcripts into a public model - upload only what the prompt truly needs.

Looking ahead

By 2026, I expect most design systems to ship with a built‑in "AI sidekick", a context‑aware co‑pilot that suggests components and copy based on the screen you are editing. That does not erase UX roles; it elevates them. The designers who thrive will be those who treat AI like a junior teammate - fast, tireless, occasionally wrong, and always in need of guidance.

Your move: pick one workflow from this article, test it this week, and share your learnings.