About
Building multi-agent automations for healthcare
Activity
1K followers
-
Steven Gaan posted thisInteresting thing I'm learning with design systems and AI tools. We should be designing our design systems to be more easily consumed by these tools to prevent incorrect usage of components etc. This probably means that we should be simplifying them as much as possible and lessening the amount of variance within each component. This makes sense in that if you were to have a human try to understand what each component is for, it's easiest when the context is more narrow. Same goes for AI. Has anyone else felt the same/different about this?
-
Steven Gaan posted thisTried Claude Design in the last few days to see where it could really help speed up prototyping. What I found was that I loved how it could grab your design system from your repo and also from Figma. What’s not so good is that I found it to be fairly inconsistent in its current state. It would make certain design choices that had nothing to do with the design system I told it to use. On the bright side, I found that the “tweaks” feature is really cool. I used to manually make this sort of thing in Claude and I’m glad it’s just baked in. Allows for some really cool exploratory work and it brought some of the fun back into the work. The big realization here for me is that it’s again just another tool to design. But the most important thing is still the ability to explore, get inspired and to have fun.
-
Steven Gaan shared thisThe design team at Magical just submitted our first PRs 👀 Any other design teams out there doing the same? React or comment if you want a peek into how we do this.
-
Steven Gaan reposted thisSteven Gaan reposted this3 red flags in AI agent platforms 1. Never buy an AI that’s a black box. Look for full transparency in reasoning and actions. 2. Never buy a single agent that tries to do everything. Look for multi-agents for better reliability. 3. Never buy an AI that can’t learn or take feedback. Look for a clear way for humans and other agents to improve it.
-
Steven Gaan posted thisSometimes, shipping really fast means little cracks start to show. At Magical, to smooth out those rough edges, we started a new tradition: UI Polish Day. We've scheduled it to happen once-a-month for just 3 hours. There is only 1 rule: only pick up what you can finish within 3 hours. No big lifts, no scope creep, just a focused cleanup that leaves our product feeling fresh. Both design and engineering found real satisfaction in getting to tie up loose ends that usually get pushed aside. Over time, the hope is that it makes a compounding difference to the way our product feels to our customers and ourselves. Curious to learn how other teams out there make time for polish?
-
Steven Gaan reposted thisSteven Gaan reposted thisThis team is growing quickly! We're hiring an AI Deployment Startegist to join us at Magical. This is a mission-critical role that touches on every part of the business, from working with our customers day-to-day to using the Magical platform to deploy agentic automation for those customers. DM me if you or someone you know is a great fit !! https://lnkd.in/dHGu4VyV
-
Steven Gaan reposted thisSteven Gaan reposted thisWe’re looking for enterprise AEs to join us at Magical! If you or someone in your network is interested in joining an early-stage startup to help enterprise organizations solve complex problems, I'd love to chat! DM me for more information!
-
Steven Gaan reposted thisSteven Gaan reposted thisHi LinkedIn! 🚀 We're Hiring! 🚀 I’m looking for two software engineers to help build the future of experimentation at Etsy: 🔹 Staff Software Engineer I – Lead the technical direction of our experimentation platform, making it easier for teams to run fast and trustworthy experiments: https://lnkd.in/gq7czxbT 🔹 Software Engineer II – Build core services, APIs, and tooling that power experimentation across Etsy: https://lnkd.in/gjypzMg9 (UPDATE: due to the overwhelming interest in this position - thank you! - we cannot process new applications at the moment. The link will be available once again if we can intake more applicants. Until then, I apologize for the inconvenience!) These are high-impact opportunities to work on systems that influence how product decisions are made at Etsy. If you’re excited about experimentation, backend development, and solving complex problems collaboratively—or know someone who might—check out the job postings above!
Steven Gaan commented on a post
1mo
“It’s hip to be dumb” - Huey Lewis and The News
Steven Gaan replied to a comment
3mo
Nate Patel totally! I know it will only improve over time. Right now it still takes some manual effort or takes time to process requests
-
Steven Gaan liked thisSteven Gaan liked thisHiring? Does it suck? Tell me about it! Seriously. I'm Zach, director of product design at LinkedIn for SMB hiring products. I'd like to hear what's working and what's not in your hiring processes. I have five 20min slots left in the next 2 weeks: https://lnkd.in/bookwithzach Your feedback will help us make hiring great people, better.
-
Steven Gaan reacted on thisSteven Gaan reacted on thisA few years ago I set up a revocable living trust. The legal part was fine. Banking it was a different story. I looked into what it would take: in-person visits, coordinating trustees, multiple trips, and gave up before I started. So the trust sat there, unfunded, for years. Today we're launching trust accounts on Mercury Personal. Most trust accounts are dead ends. You open one, park cash in it, and interact with it as little as possible because the tooling is so bad. We built ours to work the opposite way. Your trust account on Mercury Personal earns yield. You can set autotransfer rules that move money into it automatically on whatever schedule you want. It has a debit card. You can use Command to understand how your trust is growing over time: track yield, see what's moved, and take action to optimize it. And each trustee, your CPA, your advisor all get exactly the right level of access. Spending, viewing, or managing, without extra coordination or separate logins. A trust is one of the most important financial structures you can have. It should be as easy to manage as everything else in your financial life. If you have a trust, or you're forming one, you can open an account entirely online. No in-person visit, no paper forms. https://lnkd.in/gjxzyQ3c *Mercury is a fintech company, not an FDIC-insured bank. Banking services and debit cards provided through Choice Financial Group and Column N.A., Members FDIC, pursuant to licenses from Mastercard®.
-
Steven Gaan reacted on thisSteven Gaan reacted on thisLast week I was watching Pupstruction with my small children. It’s a show about a dog construction crew with an incredibly catchy theme song called: “It’s Time to Build.” Turns out, those pups are onto something. And this week, I joined Lovable to lead Strategic Partnerships & Channel. What pulled me to Lovable is a simple belief: everyone should be able to build software that people love. I want every consultant to be able to solve customer problems with custom apps instead of another spreadsheet. And I want every partner to be able to create solutions that simply weren’t practical to build before. What used to take 12 weeks can now take 12 hours, or even 12 minutes. That changes who gets to build, what gets built, and how quickly ideas can become real products. That’s the ecosystem I came here to help create and shout out to Matt Norton , Cecilia Stallsmith and Ryan Meadows for laying an amazing foundation. I’m deeply grateful to everyone at Axon for an unforgettable chapter protecting life and scaling enterprise partnerships. The mission, teammates, customers, and partners made me better every day. Special thanks to Mike Shore, Delucia Lepore, Anna Spallino, David Zyck you are amazing. Now my friends, it’s time to build.
-
Steven Gaan reacted on thisSteven Gaan reacted on thisFriday marked my last day leading the WorkRamp team after an almost 11-year journey. After successfully integrating our team and product into Learning Pool over the last 9 months post-acquisition, I wanted to say goodbye to our customers, teammates, and investors. To our customers, thank you for believing in us. I will forever remember our earliest customers, like Russ Thau and Kristen Buhler who were our first two customers when we had little more than a functioning website. Then amazing believers like John Ley and Frank Traditi, who took so many reference calls to help close early customers and investors. Then more champions like Julissa Perez, Dr. Claudia Igbrude, Ryan Giordano, and Lish Barber, who were huge advocates of WorkRamp in the early days. I can’t name all of our 700+ customers, but I want to say thank you to each and every one of you for trusting us to support your enablement, L&D, and customer education programs. To our team of WorkRampers who walked our in-person and remote halls, thank you for every contribution you made to our company and the culture you built. During our exit interviews, so many people said the #1 reason they worked with WorkRamp for so long was because of the other team members they could work with day in and day out. Among a sea of options, I truly believe we were an amazing place to work because of the environment you all created. It was an honor and a privilege to lead you over the years. To our investors and board, thank you for being partners with us through the years. I never had an ask turned down, from joining our all-hands and customer webinars to making sure we had a backstop during the SVB crisis. Through ~6 years of board meetings, ~75 investor updates, and unwavering commitment from our Seed Round through Series C, you were an amazing group supporting our journey. Lastly, to Arsh Mand, you were the best co-founder anyone could ask for. From our first boba tea together discussing WorkRamp to integrating WorkRamp into Learning Pool, you were WorkRamp’s rock behind the scenes all these years. As for what’s next, I’ll be back at the drawing board, starting something new again soon. There’s never been a more exciting time than now to start a startup, so I can’t wait to get back in the arena for another run.
-
Steven Gaan reacted on thisSteven Gaan reacted on thisAfter an incredible chapter, I've closed out my time Toboggan Labs last week. I'm leaving with a lot of gratitude. I had the chance to work on hard problems with people who genuinely care about building thoughtful products. The work challenged me, the team made me better, and I'll carry those lessons into whatever comes next. Thank you to everyone I had the privilege of working with. Thanks to Belinda Darcey for the small big world that reconnected us and the stellar team that continues to grow at Toboggan! I'm excited to see what you all build next. If you're ever in Toronto area (or just want to catch up) don't hesitate to reach out and say hi!
-
Steven Gaan reacted on thisSteven Gaan reacted on thisMost AI companies ask you to trust their benchmarks. But we got KLAS Research to ask our customers instead. The 2026 KLAS Emerging Company Spotlight just dropped, and Magical received an A+ Likelihood to Recommend, with every interviewed organization saying they'd highly recommend us. When we started Magical, the thesis was simple: healthcare's back office runs on human expertise that never gets captured, never scales, and walks out the door when people leave. We built a platform to change that. What these KLAS scores reflect isn't just product quality. It's what happens when AI is actually trained by the people who know the work, and then deployed into the systems where the work happens. 100% key functionality adoption. Live in weeks. No APIs required. The verdict is in. We're just getting started.
-
Steven Gaan liked thisSteven Gaan liked thisThis week, I discovered a more efficient way to create animations using Claude Code. Traditionally, producing this type of animation involved using After Effects or Rive, followed by converting it to Lottie for production, which can be quite time-consuming. The new workflow I explored in Claude Code allows for direct output of code that can be integrated straight into a web page or product, eliminating the conversion step. Here’s the workflow I followed: 1. Draw vectors in Figma, set more keyframes for complex motion, and ensure layers are grouped and named consistently. 2. Connect to Claude Code via Figma MCP, have it read the file, and explain back what it understood before adding animation direction. 3. To add animation, use Bezier curves or specify a preset like "cubic-bezier ease in" for easing. 4. For audio, while Claude can generate sound, I prefer to use my own audio for this animation. 5. Convert fiddly parameters (color, duration) into live sliders instead of hardcoding them. A few takeaways from this experience: - Claude excels at non-deforming motion, translation, rotation, and scale where the logic of motion is visually clear. However, for deformation or more abstract animations, the time spent explaining can rival coding time, with uncertain payoffs. - The most significant benefit is not just time saved, but a new design-to-production path that outputs usable code from the start, bypassing the AE-to-Lottie round trip. - If the animation complexity increases to the point of needing to recreate more native AE features within Claude, it may be counterproductive. This approach seems best suited for mid-complexity animations; beyond that, After Effects remains the superior tool. #VibeCoding #CreativeCoding #AnimationDesign #FigmaMCP #ClaudeCode
Experience
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
Education
Volunteer Experience
-
Mentor
adplist.org
- Present 10 months
Helping designers with any questions they have around career, portfolios, and AI
Honors & Awards
-
Hack to the Future, Winner
Box
Won the "best hack" by a non-engineer for the Box Career Center. Built a site to improve and promote internal career development.
Recommendations received
2 people have recommended Steven
Join now to viewOther similar profiles
Explore more posts
Explore top content on LinkedIn
Find curated posts and insights for relevant topics all in one place.
View top content