A realistic guide to shipping your own designs.Your company wants you to use more tokens. Engineers are shipping faster than you can prototype. Vibecoded slop is all over your LinkedIn feed. Agents are the future. Figma’s over. Anthropic just released Claude Design. Taste is everything. Eng just shipped another PR! Design is dead. Actually, you should use less tokens now. Wait, maybe Figma is back —did you catch that Config talk? Eng shipped something again! Your Agents need Design Skills. What even is a Design Skill? Do you still have design skills???No matter where you live in the world right now, roles across the tech sector are drastically changing. However here in San Francisco, at the epicenter of the so-called AI Bubble, we have both the privilege and burden of experiencing this change more than most. Every company in town is sprinting towards one goal: building the elusive ‘AI native’ team. The ask is direct and urgent: optimize workflows across all disciplines, 10x productivity and ship the future.But how does an AI workflow apply to something as creative and ambiguous as Design?How do you optimize craft, or 10x design thinking?These are the questions I have spent the past year trying to solve for. Candidly, I don’t have all the answers yet — but I don’t think anyone else does either. In a world where tools, roles and expectations are constantly evolving, I believe the best we can do as a design community is share our learnings, and pave a new path forward together.When I joined Amplitude a few months ago to design Agent Analytics, our new Observability and Evals platform, I had never shipped a PR in my life. Today, I pretty much own our team’s frontend.It took a lot for me to get to where I am today, and although I had a lot of support, there wasn’t exactly a playbook I could follow at the time. So hopefully, this gives you a bit more of a head start!Before I jump in, I have added a short glossary below that you might want to take a look at — just in case, like me, you are not super familiar with engineering terms.AI Design comes with new terminology.Now without further ado, here are the steps I took to become an AI Designer.Step 1 —Taking a break from FigmaMy biggest advice for any designer who wants to become AI-native is simple: try closing Figma for a week and see what happens.Say goodbye (for now!) to Figma.I know, this might sound a bit extreme. I’m not saying you have to leave Figma behind forever (although personally, I don’t use it for product design work at all anymore). But, if you really want to embrace this new way of working, you have to let go of your old routine completely.This sort of happened organically for me. I joined a 0–1 team at Amplitude, which meant we were building our product from scratch and there were no existing Figma files for me to leverage. On top of this, the engineers on my team were moving fast (ie. shipping brand new features multiple times a day) and I didn’t want design to be the bottleneck.I tried my best to speedily build out Figma screens at first. I consider myself an extremely proficient Figma-user, but there was just no way I could keep up with all the changes if I tried designing pixel-perfect, up-to-date prototypes. I tried creating rough wireframes instead, but even that was feeling slow when it came to mapping out the complex, ever-evolving workflows of our product.If I wanted to really make an impact, I had to figure out a way to influence the product direction in real time. Luckily, I had an incredible design mentor, Tyler Wanlass, who encouraged me to break away from my traditional toolstack and embrace AI head-on.Step 2 —Vibecoding interactive prototypesThis was the beginning of my first AI design breakthrough: the humble Claude-coded artifact. Instead of building out screens in Figma, I vibecoded quick click-through html prototypes in Claude.Lo-Fi wireframes made with Claude.Every day, I would share them with my team, get feedback, iterate immediately, and share them out again before EOD. If I wanted to iterate on smaller UX details, I would ask Claude to mock up 3–5 component explorations instead of building out a complete screen each time. In both cases, the ability to quickly mock up interactions was a total game changer when it came to exploring user flows in detail.Exploring smaller UI component Explorations in a chat.At first, I thought it would be impossible to accurately visualize and verbalize the changes I wanted to make to a design without manually editing a component frame by frame. Turns out, I was totally wrong. Once I actually tried to do it, I realized that I had acquired a pretty impressive subconscious database of UX patterns over the years. In no time at all, I found myself efficiently directing my agent to mock up the exact visuals I was imagining in just a few prompts.An important caveat to the success of these early wireframes was keeping them a) as simple as possible, and b) visually unrelated to our actual design system. Although it’s tempting to perfectly recreate your product’s interface, I quickly realized that perfecting visual detail is not AI’s strong suit. Getting a wireframe ninety percent of the way there is exponentially easier than trying to get the last ten percent right. Make sure to weigh up if trying to perfect the details is really worth the wasted hours of frustration.Higher fidelity HTML Product Shots.That being said, this was a few months before the release of Claude Design, which is a little better at handling visual polish — especially if you take the time to set up your design system preferences.I now work with a mix of lo-fi and hi-fi prototypes, depending on the feedback needed, all through Claude Design. It has the added benefit of being able to juxtapose multiple iterations of each screen on the same canvas, which was a key visual I found I was missing before. I even create HTML product shots to publish on our website.Honestly, this is the reality of AI Design —new tools are released everyday and you have to keep updating your process as you go. I find having a few high-quality resources to turn to like the AI Design Field Guide, has really helped me get a sense of how other designers in the space are operating.Step 3 — Finding your new Source of TruthThe vibecoded prototypes finally allowed me to keep up with my team’s pace and meaningfully shape what the engineers were building. I had made huge progress, but I still felt like I was failing as a designer.I was stuck on one big dilemma: I had no design ‘Source of Truth’. If someone asked me “where are the latest Agent Analytics designs?”, I had nowhere to point them to.I tried a bunch of methods, from hacky Figma-screenshot hybrids to wonky HTML exporters. All of them were either too lo-fi to be called final designs, or too hi-fi to be kept up-to-date. Even if I somehow managed to rapidly prototype super polished screens, the live product was still evolving and diverging too much for me to keep up.Around this time, I listened to Jenny Wen share how her design team at Anthropic was completely rethinking the design process. She talked about how the traditional discovery → mock → iterate workflow was designed for a world where building was slow and expensive. Now, her process isn’t centered around just prototyping in Figma, but on shaping implementation and product direction through a combination of different tools. She said that increasingly, the design Source of Truth’ was becoming the product itself.So, there it was. Prod was the Source of Truth. And if I wanted our source of truth to be designed, I had to start shipping some real code.Make the product itself your Source of Truth for designs.Step 4 — Getting access to the codebaseTo me, AI-native design is at its core the ability to ship your own designs and make them great. Up until this point, I felt like I was speeding up my design workflow but not necessarily improving it. This step was the real difference between using AI for efficient design output versus using AI for a better design output.Truly great interface design lives in the details, but we all know that even the most perfectly-annotated Figma prototype is too often desecrated after handoff. Having access to edit our real codebase has allowed me to refine the tiniest interactions and fix the smallest visual nits. These are changes that in isolation would only be caught by a designer’s eye, but when digested all together create a thoughtful, high-quality software experience that stands out to anyone. On top of the ability to bring static designs to life, shipping code has also given me the ability to add brand new design details in the form of interactivity, responsiveness and delight.With AI, making interactions delightful is easier than ever.So how does a designer start coding?In the age of AI, building is easy. I truly believe that any designer, with some practice, can learn how to make significant frontend changes and ship PRs. The hard part is getting the right support and access to ship code.I am lucky enough to be working within a super forward-thinking EPD org that has invested in resources and processes to empower non-engineers to ship code. I’m aware that everyone does not have the same access to token usage, learning resources and technical support as I do.However, even if your team isn’t there yet, don’t worry. You can be the one to push for change! In fact, I have multiple design friends who, upon learning that I was regularly shipping code, successfully advocated to do the same at their organizations. More often than not, it just takes one person to know that this workflow is possible and get the ball rolling.Once you get your first PR merged, there’s no going back!Here are a few things you can do to advocate for getting designers access to code -Share real examples of designers shipping code.I think the first barrier to entry for this stuff is simply exposure. The idea of non-engineers pushing code is still pretty new, so your team may not even have considered the idea that designers could be shipping PRs. Set up a meeting where you share real examples of companies and design teams that are successfully implementing code every day to optimise workflows (feel free to use this post as a reference 😊).Explain the benefits of the coding designer.A real (and frankly valid) concern from the Engineering side may be that designers will just create a backlog of sloppy PRs for eng to review, adding to their already busy workload. It’s important to make your case here: explain that you will actually be taking away a lot of the frontend work that engineers had to do in the past. No more pings to update colors or reduce a gap from 24px to 16px! Plus, make sure to emphasize that want to learn best practices and start small with your changes.Find your engineering Champion.Although it’s great to have buy-in from the whole engineering team, often all it takes is one Engineering partner who is willing to put in a few hours and help you get set up. I have been lucky enough to work with many lovely, talented engineers throughout my career who have generously helped me learn many of the technical skills I have today. Long story short, make friends with eng!Start out with a bot.If you are getting a lot of pushback around setting up the entire Design team to code, I suggest starting out with a coding agent bot in your company’s messaging workspace. Once the eng team sets up the bot, there’s zero setup for designers — no GitHub, repos or dev environments, you can just ping bots to make changes to real code. It’s a nice, low-friction entry point for designers — and it encourages smaller changes like copy fixes or color tweaks, while everyone familiarizes themselves with vibecoding.I shipped my first PR using a Cursor Bot in Slack.If your company still isn’t up for it — do it yourself!Don’t feel like you have to wait to be at an AI native company to become an AI native designer. If you’re getting no support at work, invest in yourself. Sign up for your vibecoding tool of choice and ship your own projects. You could start off with easy-entry tools like V0 and Lovable, then work your way up to the likes of Cursor and Claude Code. There are endless free online resources to teach you how! Then, once you have your own builder’s portfolio, I can guarantee that many AI native companies will be excited to have you.Step 5— Owning your product’s frontendTo be clear, when I say “owning” the frontend, I don’t mean to say that you should be building out your product’s whole interface. I certainly am not (and could not) do that.But, you should become the lead on frontend decisions. On my team, I review all frontend PRs, collaborate directly with engineers on UX overhauls and implement a large portion of UI polish myself.My GitHub Contributions since joining Amplitude.Getting to this place isn’t easy, there’s a lot you will have to learn and unlearn. The key is to start small, ask questions and study up! Begin with tiny UI changes to build up confidence, and then go bigger. Figure out the lines between frontend and backend changes. Understand when something is worth your time, and when you should hand off tasks to Eng.Once you get the basics, start leveling up your vibecoding game:Build design skills that make everyone’s code better — I highly recommend checking out John Phamous’ article about Teaching agents product design at Vercel.Communicate your reasoning behind design updates, even if you feel like the reasoning is obvious. In Details That Make Interfaces Feel Better, Interfere’s Jakub Krehel beautifully illustrates the delicate nuances that can totally transform a UI.Sprinkle in more advanced interactions throughout your designs. I suggest Emil Kowalski’s 7 Practical Animation Tips to get started.It definitely took some time before I truly built up enough trust and skill to make meaningful changes. I had to prove that I was willing to do the work and educate myself. I had to partner closely with my engineers and PM to rethink our old handoffs. Over time, I got more and more comfortable shipping PRs. I started building out more refined interfaces, working on multiple designs simultaneously and ultimately becoming the go-to person for UI/UX decisions.Step 6 — Embracing the chaos ?Don’t lose your craft amidst the vibes.When I finally felt like I had achieved my coveted ‘AI Designer’ status, I couldn’t wait to share my new workflow with other designers. I had been through so many ups and downs, learned so many new skills, and I felt like I had finally reached a satisfactory, sustainable destination. Here I was, spinning up several agents at a time, efficiently vibecoding entire pages in a few prompts and comfortably shipping multiple PRs a day. Honestly, I thought I had finally figured it all out.Then… the novelty of shipping my own designs began to fade. Jumping between tabs and constantly switching context became mentally exhausting. This new ability to quickly make changes, which had at first seemed like so powerful, started to make my designs feel transient and meaningless.And after months of embracing the chaos, re-inventing my routines and learning new tools, I feel conflicted. Last year, I read a truly amazing article by Amelia Wattenberger titled Our Interfaces Have Lost Their Senses. It is a thoughtful and creative exploration of how the loss of friction in our experiences has led to a loss in overall human satisfaction. As AI continues to takeover the world, this article resonates with me more and more.My new AI-powered workflow has allowed me to build out beautiful, complex designs at a totally unprecedented speed and scale. I have added delight and refinement and consistency to the interfaces my users are actually using and that impact feels super tangible and right.But I must also acknowledge that it has come with an unprecedented detachment from my craft. When I compare my new daily routine of deliberating with agents, jumping between prompts and babysitting PRs, with my old days spent perfecting Figma frames in a total creative flow state, I feel stuck.It feels like I have created a workflow that is optimized enough that I cannot go backwards, but not rewarding enough for me to stay where I’m at. I guess my only option is to move forward, embrace the chaos, and eventually, somehow, figure out a better final destination.In conclusionI hope that this article was able to provide you with some real insight into what one AI native Designer’s workflow looks like right now. I’m sure that in a month’s time, my routine and tool stack and opinions will have totally changed again — but I think this is an accurate record of where I am at right now. I know we’re all figuring this out together, so please share any new tools or workflows that are working for you in the comments!FAQQuestions I get asked (or have asked myself) about AI Design —Q. How do I pick which AI tools are the best?Trying to navigate the sea of AI coding and design tools can be super overwhelming. While I of course cannot definitively say what the best tools are, I can share my own toolstack.These are the AI tools I use every day:Claude Chat: For a mix of research, diagram mockups, smaller UI explorations, technical debugging and general brainstorming.ChatGPT: For help with in-product copy, feature naming and other language-based tasks.Claude Design: For prototyping many iterations side-by-side in one canvas and for creating exportable html for marketing-style product visuals.Cursor: For vibecoding in prod and shipping PRs.Linear: For tracking design tickets, team goals and other to-do list items. Not really AI, but I do use their slackbot a lot.Notion AI: For summarizing docs and customer call transcripts.Q. How do I stop wasting hours on one PR?The simplest way to do this is to set a limit on your “repeat” prompts . If you find yourself trying to get your agent to fix the same issue with more than 3 prompts, pause. Either move on to a new PR and come back later (sometimes I feel like the model is just having a bad day), or hand it off to an engineer to finish up. If you find yourself fixing the same issue in multiple PRs, that’s when it’s time to update your SKILL.md file.Q. How do I ship full feature designs, not tiny changes?Honestly, practice makes perfect! Be patient, keep asking questions and start small. Over time, you’ll inherently develop a better sense of the possibilities and limitations of vibecoding. One other concrete recommendation here is to build out a prototype you’re happy with in a tool like Claude Design first, then use that as a reference for your coding Agent when updating Prod. It’s much more effective than going in and trying to make a bunch of incremental changes directly on the real UI.Q. Why does it feel like people are questioning my design choices more?AI has made the lines between all EPD roles blurry — Engineers are taking customer calls, PMs are making wireframes and Designers are vibecoding. As responsibilities and skillsets start to overlap, this can lead to a bit of friction. When I first started shipping code, I got lots of comments on my PRs questioning my UI/UX decisions. At first, I was frustrated. I felt like I was the design lead, and that was the end of it. But in this new era of work, we need to take the time to redefine our roles, share our knowledge and work together to build trust. I realized that if I was expecting engineers to take on the work of helping me learn to code, the least I could do was take a few seconds to explain my design rationale. Nowadays, I feel like pretty much all my UI/UX decisions are trusted.Q. Why am I spending all my time fixing UI nits?AI isn’t always great at UI — even though it’s getting better and better every day. If you find yourself, like I did, wasting several hours a day polishing your UI, here are some things that might help.Improve Your Design System: Work with the DS team to improve your foundations to speed up vibe-designing in the long run.Work on Design Skills: Continuously try to improve your design skills, updating them every time you notice a repeat problem. Then encourage your engineers to use them…Set up a Polish Week: One side effect of designers being able to edit the real UI, is that we can become the team’s go-to “cleanup” person. To avoid this, I pushed my team to set up a monthly Polish Week, where we set up daily bug bashes and assign out UI nit tickets across everyone. It’s been great!Q. How do I optimize Craft and 10x Design Thinking?Short answer: I just don’t think you can.You can optimize your workflow and 10x your output. You can create endless iterations and fully-interactive prototypes and really cool animations. You can ship the most refined, pixel-perfect interfaces of your career. But, you cannot speed up creativity. You cannot simulate Taste. You cannot use an LLM to replace what time and experience and art bring to the table.I had really hoped to have a better answer to this question by the end of this article, but like with any change, there is improvement and there is compromise.Q. How do I stay connected to Design with all this change?Obviously, as per Step 6, I don’t have this all figured out yet. However, there is one thing that has really helped me stay in touch with Design: connecting with other designers. It can be easy to forget that we’re all going through this chaos together, and sharing that experience can be really helpful. I would encourage you to check in with your design colleagues, reach out to your design friends and go to more design events!ReferencesOn Rethinking the Design ProcessWen, J. (2026). The Design Process Is Dead. Here’s What’s Replacing It. Lenny’s Newsletter. https://www.lennysnewsletter.com/p/the-design-process-is-dead-heres-whats-replacing-itWattenberger, A. (2025). Our Interfaces Have Lost Their Senses. https://wattenberger.com/thoughts/our-interfaces-have-lost-their-sensesMather, R., Chen, R., & Villa, F. (2026). AI Design Field Guide. https://www.aidesignfieldguide.com/Amplitude Design Team (2026). Design at Amplitude. amplitude.designOn Learning to Build with AIWanlass, T. Principles. https://tyler.cv/Krehel, J. (2026). Details That Make Interfaces Feel Better. https://jakub.kr/writing/details-that-make-interfaces-feel-betterKowalski, E. (2026). 7 Practical Animation Tipshttps://emilkowal.ski/ui/7-practical-animation-tipsPhamous, J. (2026). Teaching Agents Product Design at Vercel. Vercel. https://vercel.com/blog/teaching-agents-product-design-at-vercelArticle written by Maya Brennan, Lead Designer at Amplitude + Statsighttps://www.mayabrennan.io/How to become an AI Designer was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.