51:06Leveling Up with Eric Siu · Interview
The AI Content Machine That Drives $100M+ Leads
Alex Lieberman demos the 8-step Claude skill directory he spent 50 hours building — and shows how it cut a 30-hour post down to four.
June 3rdThe argument in one line.
The designer competitive moat is shifting from taste to systems-building: those who encode aesthetic judgment into reusable infrastructure will govern how AI expresses quality at scale.
Read if. Skip if.
READ IF YOU ARE…
- A designer wondering whether to invest time learning to code or just lean on AI-generated output.
- A design lead or manager trying to figure out what org-level AI adoption actually looks like beyond the hype.
- Someone in a leadership role who talks about AI strategy but has not opened Claude Code or Codex personally.
- A designer at a startup or mid-size company asking whether it is worth getting direct production codebase access.
- Anyone building internal AI tools or Slack agents trying to understand how other orgs handle knowledge transfer.
SKIP IF…
- You want a step-by-step tutorial on a specific AI design tool -- this is a strategic conversation, not a how-to.
- You work in a field where code and production access are not relevant to your design work.
The full version, fast.
Three practitioners from Anthropic, Every, and Ramp converge on a consistent argument: the orgs winning at AI adoption have executives in the tools every day, designers with direct production codebase access, and a culture that treats current workflows as obviously incomplete. The panel surfaces two concrete inflection points -- a 7-out-of-10 baseline that AI now reliably hits, freeing designers to focus on what genuinely requires judgment -- and a shift in where that judgment is most needed, away from the UI surface and toward structural decisions about what stays fixed and what personalization is allowed. Knowledge transfer across teams remains the unsolved problem, with Slack agents as the closest working solution.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Who's talking.
00:00hostRid
00:16guestMegan Choi
00:22guestDan Shipper
00:28guestBradley Ziffer
Where the time goes.
00:00 – 06:15
01 · Intro and milestone map
Host introduces the live panel context; Megan Choi outlines two transformation milestones: production codebase access for designers, and letting go of design control.
06:15 – 09:30
02 · Sponsor break
Dessen (prototype on production UI) and Paper (snapshot Chrome extension) sponsor reads.
09:30 – 15:00
03 · Design-engineer collaboration and the 7/10 idea
Bradley on what design-engineering collaboration really means: not labels, but care. The 7-out-of-10 framing -- AI hits the floor, humans own the ceiling.
15:00 – 21:00
04 · Time allocation and prioritization
Megan on how the Claude Code team avoids polishing things that may not exist in six months. Polish is shared responsibility now.
21:00 – 28:00
05 · Org transformation and leadership behavior
Dan Shipper: the leading indicator is CEO daily tool usage. AI working groups do not work. The late 2025 model capability jump was a real inflection point.
28:00 – 34:00
06 · What AI fluency actually looks like
Bradley and Megan on what good looks like six months from now: learning speed, belief updating, systems thinking, and reducing tool sprawl.
34:00 – 39:00
07 · Knowledge transfer and Slack agents
How individual learnings propagate org-wide. Megan on pairing sessions. Dan on Mailroom. Bradley on Ramp Cody dropping albums in Slack.
39:00 – 39:47
08 · The designer value proposition, end state
Megan personal take: models will do most fundamental design work by end of year. Designer value moves to fixed/flexible UX architecture, model harness design, identity primitives.
Lines worth screenshotting.
- Giving designers a separate sandbox repo instead of production codebase access is worse than nothing -- you maintain two codebases and miss the actual data endpoints.
- AI reliably gets any design idea to a 7 out of 10. The question is what you do with the time you just reclaimed, not how to reach 7.
- The CEO daily tool usage is the single best leading indicator of how well an org will adopt AI -- you cannot manage a team in the tool if you have never used it.
- Skills and prompts are deeply personal -- shared skill libraries always need per-person customization, which makes watching someone work for an hour more useful than downloading their files.
- Public Slack AI agents solve the knowledge-transfer problem better than any documentation system because people accidentally see each other prompting.
- Designers who thrive hold two truths simultaneously: ship excellent things today, and believe everything shipped today is wrong.
- We have solved exactly two AI use cases at scale: search and coding. Assuming we are anywhere near the end is not supported by evidence.
- The designer value proposition is moving downstack: UI layer to fixed/flexible UX decisions to model interaction harness to identity primitives.
- Pairing sessions -- watching someone else use Claude for an hour -- surface tricks neither person knew they had learned.
- Giving an AI coding agent a Gmail plus-format email address is a working multi-agent collaboration pattern available today.
- The instinct to polish everything once you can is a trap -- six hours of small tickets is rarely the highest-leverage use of a designer who can now ship.
- Multidimensional people who combine creative backgrounds with technical curiosity are disproportionately good at using AI as a creative playground.
- The worst time to polish something is when the product might not exist in six months -- prioritizing by survival probability is an underrated design skill.
- Brand and fundamental design systems will remain human-judgment-intensive because they encode subjective taste decisions that expert crafting makes perceptibly different.
The designer who thrives learns faster than the model improves.
WHAT TO LEARN
Three practitioners from companies at the frontier agree: the competitive moat is not taste alone -- it is systems-building, belief updating, and the willingness to treat every current workflow as obviously incomplete.
01Intro and milestone map
- Production codebase access is a prerequisite for design relevance, not a privilege to earn. A sandbox that mirrors prod is always out of date and misses the real data endpoints that shape actual product behavior.
- Letting go of design control is the symmetric milestone: just as designers need to touch code, engineers need to ship UI without a designer in the loop on every feature.
03Design-engineer collaboration and the 7/10 idea
- AI reliably reaches a 7-out-of-10 quality floor on almost any design idea. The question is not how to reach 7 -- it is whether you are being intentional about what you do with the time you just recovered.
- Care is still the differentiator. AI lowers the floor, not the ceiling, and the extra time should go toward the details that matter most to users, not toward compulsive polishing of everything.
04Time allocation and prioritization
- The hardest thing to internalize when you can suddenly do everything is that doing everything is not the right move. Prioritize work that survives the next model update.
- Polish is now a shared responsibility. It is no longer only on the designer just because the designer can now ship it.
05Org transformation and leadership behavior
- Executive daily tool usage is the single most reliable leading indicator of whether an org will actually change. Delegation to an AI working group predicts failure.
- The model capability inflection of late 2025 was real and sudden. Teams that tolerated the weirdness before that moment built compound intuition advantage.
06What AI fluency actually looks like
- The best practitioners operate every day as though the current state is obviously incomplete and could be obsolete by the next model drop. That belief drives exploration rather than defensive attachment.
- Systems thinking -- reducing 15 tools solving 15 problems to 4 tools solving 30 -- is what AI fluency looks like at the org level, and it requires deep product knowledge that AI cannot substitute.
07Knowledge transfer and Slack agents
- Public Slack AI agents solve the knowledge-transfer problem better than any documentation system because people accidentally see each other prompting, seeding ideas faster than any meeting.
- Skills and prompts are personal enough that shared libraries always need per-person customization, which means watching someone else work for an hour teaches more than downloading their skill files.
08The designer value proposition, end state
- The designer value proposition is moving downstack: from UI surface decisions toward structural choices about fixed versus flexible UI, then toward model interaction harness design, and eventually toward the identity primitives that determine what a product feels like.
Terms worth knowing.
- 7 out of 10 baseline
- The panel shorthand for the quality floor AI now delivers automatically -- good enough that anyone can ship it, freeing expert designers to focus on the gap that requires genuine taste.
- Production codebase access
- Giving designers direct access to the live codebase rather than a sandbox fork, so they work with real data endpoints, actual org tooling, and changes that users immediately experience.
- Slack agent
- An AI model connected to a Slack workspace that reads channel context and responds autonomously -- used as a knowledge-transfer mechanism and debugging tool across teams.
- AI fluency
- Operating in Claude Code or Codex for most of the day and continuously updating your workflow as models improve, rather than occasional use or surface-level prompting.
- Fixed vs. flexible UI
- A structural UX decision about which elements remain constant versus which can adapt or be personalized -- increasingly the central design question as AI handles more surface-level generation.
- Harness layer
- The infrastructure wrapping a model inside a product -- prompt routing, context management, constraints -- distinct from the UI layer and increasingly where design expertise is needed.
- Mailroom
- A personal tool that gives an AI coding agent a Gmail plus-format email address, allowing Slack agents to route tasks directly to a Codex instance that handles them autonomously.
Things they pointed at.
41:00toolInflight (feedback tool, open beta)
31:20toolThe Victor (Slack agent platform)
31:10toolMailroom (custom Codex email routing)
37:30toolRamp Inspect (Slack bot)
38:00toolRamp Cody (Slack AI agent)
Lines you could clip.
02:28
“Let your designers get access to your production codebase. That is the starting point of this conversation.”
07:10
“Then you are maintaining two repositories. If you ask any engineer if they ever want to fork their codebase, they will always tell you that is a terrible idea.”
26:10
“The main thing I always look at is what is the CEO doing? Because it is not outsourceable.”
33:00
“Every pixel I push could be obsolete next week depending on the next model that comes out.”
33:20
“We have really only solved two use cases: search and coding. We are so far away from the end.”
37:00
“I felt silly because I was talking to a bot for a really long time, and his name is Cody. People here are so nice.”
Where the conversation goes.
00:00 – 09:30denseProduction access and org milestones
09:30 – 21:00denseThe 7/10 benchmark and time allocation
21:00 – 28:00denseOrg transformation and executive behavior
28:00 – 34:00denseAI fluency definitions and systems thinking
34:00 – 39:00steadyKnowledge transfer and Slack agents
39:00 – 39:47denseDesigner value proposition, future state
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Welcome to Dive Club. My name is Rid, and this is where designers never stop learning. This week's episode is a special one because it's actually a live recording of a panel discussion that happened last week in New York City.
00:14We get to hear from three incredible guests. The first is Megan Choi, who's the design lead behind Claude Code and Cowork.
00:22The second is Dan Shipper, who's the CEO of Every and one of my favorite thinkers when it comes to AI. And the third is Bradley Ziffer, who's a design engineer at Ramp. And the entire point of this discussion is to do a deep dive into not only the future of design, but what AI workflows look like at some of these prolific companies.
00:44So without further ado, let's dive in. Welcome back, everybody.
00:53I've always wanted to start talking after a live applause. That felt pretty good.
00:58I'm not gonna lie. That felt pretty good. I think I could do that.
01:00Three amazing guests who've all given wonderful workflows, and we're gonna go a little bit deeper here and talk about the different things that people are observing in the industry, how the way that we work is changing some of the best practices that are still kinda actively being discovered inside of our orgs.
01:17And I'm gonna try something a little bit weird. Haven't really prepped them at all at all. And I'm gonna ask us to put on these hypothetical consultant hats.
01:27Not so hard for you, Dan, but, you know, maybe maybe over here, a little bit new. And we're gonna just help people think about different avenues for your transformation.
01:38Right? It's like if we wanna grow as builders, as orgs who are leveraging AI to, you know, the fullest capabilities and really rethinking from first principles how we operate, what is that going to entail?
01:51Right? And so talking, Megan, with you earlier, something that I learned is that Megan does a lot of on sites with teams, teaching these design orgs how to make the most of Claude and building and using AI.
02:04And so I'd like to start with you and we can put on again this not so hypothetical consultant hat and think about you know, you talked about, like, this two zero one level course almost that you're putting people through. So when you think about the transformation that you're hoping to have design org see, what are some of the main mile markers on that journey that you would look for?
02:28Yeah. I think I met a few folks today actually who attended, like, some previous talks back when Cloud Code was less popular. And I think the theme of 2025 and, like, for orgs first getting into this is, like, let your designers get access to your production code base.
02:42That's, like, the starting point of this conversation. It is been a very gate cap and for a lot of good reasons, for security, for privacy, for, like, your ability to understand that you're making these changes. It's been segregated for a while, but it's so, so, so important now that these tools are available.
02:59And, like, the more information access they have, the better you'll be able to build to get access to that production code, to go through the process process of shipping to production because the closer you are to what your end user see, the closer you have to, like, influence the product. The second, and this one is deeply uncomfortable for a lot of designers, is in the same way that we're asking our engineers to let us in and help us code, you need to be more comfortable letting go of design.
03:24And that means that a lot of features can go out without you. A lot of the time, it's actually very possible for people to get a v one, a v two, even a v three out there and have it be pretty good and have it be pretty aligned with your design system. And, you know, you build the automations in place so that that's the right checks and balances, but I think it, like, goes both ways.
03:43And so those are, like, the two big milestones that I often see. And I think it's crazy if you think about it that, like, we all want access to that production code base.
03:52But when it comes to having our engineers designed for us, the initial reaction is like, my god. I don't wanna do that. But it's so important, I think, to, like, shift your mindset a little bit and, like, recognize that we're all kind of have, like, this new ability
04:03to build. And let me push on that for a second because I've talked to quite a few, especially larger teams, but not even massive teams where they're very intimidated by getting designers into prod, and there's this story that you can tell around, well, if we just make a separate repo and a playground, then designers are coding and that's, you know, the eighty twenty value.
04:23What would you say to that leader? I was just like, then you're maintaining two repositories. If you ask any engineer if they ever wanna fork their codebase and maintain two versions of the same thing, they will always tell you that's a terrible idea because you have to maintain two things, and they're always gonna be out of dates.
04:37Also, part of being in the production code base is having access to all the tools that your organization is building into that code base because you bet your engineers are also on this journey learning with us, and they're scaling themselves. It also gives you access to the actual data endpoints so that when you're logging and you're doing data queries, you can actually see what's happening.
04:53There's so much built into a real code base that you just can't replicate in a sandbox that it's just you just really need access to it to really experience the power of what it's like to build a product that your users end up using.
05:06I've anticipated Dessen's recent release for a full year. It's called Surfaces, and it enables you to design and prototype directly on top of an existing production interface.
05:19All of the pages and flows that you need are preloaded inside of Dessen as starting points. So you can easily make changes, explore ideas, and then when you're ready, you can share your prototype in a single click.
05:33Dessen is the only design tool that allows you to design on existing pages and flows pulled directly from prod, which is a pretty big deal for teams, and you can get started today, just head to dive.club/dessen. That's dessn.
05:50Never in a million years did I think my design workflow would change so drastically in the last few months, and a big part of it is Paper's new snapshot tool. It's a Chrome extension that lets you copy any component or element from your live website and then paste it directly into Paper as editable layers. So all the time, I'll grab something from prod, have Claude immediately spin it up six different ways on Paper's canvas, and then when I'm ready to send a concept back to Claude, it's seamless because Paper's canvas uses real HTML and CSS.
06:23That workflow feels a lot like the future of design to me, and you can start doing this today. Just head to dive.club/paper to try it out.
06:32Now on to the episode. Alright, Bradley. I'm gonna go to you for a second here as the resident design engineer, this person who's been kind of bridging and seeing both sides of the table for a while now.
06:42In this world where many more designers are shipping, touching production, doing things in GitHub that they may or may not fully understand, how is this shaping what you think really solid design engineering collaboration looks like?
06:59I don't know if it's so much that we all need to become
07:03design engineers or designers or whatever box or label you want to put on things.
07:10I think it's more about what does it mean to put care and intention into something.
07:20Like Megan saying, we can probably get to a place where seven out of 10. We're at a seven out of 10, like, of the box, a thing you wanna do, an idea that you have, we can get it seven out of 10, and anyone can do it.
07:33And that's actually great, but that just means you have more room to care and put love and intention and thought into the things that you build, and I think that that that's what it in my mind, it's about.
07:46Sure. It's like, you should understand this this animation you're putting out and what it means to be performant on the web and what it means to make me rage quit your app because you have like a menu and then another menu, but when I go a little bit outside the boundary of that second menu, it like closes the menu and I'm like, you should care about those things, but that is all just care.
08:12Just you care. It's just you caring and now you have a maybe a hopefully a little bit more time to care. Okay.
08:18I wanna touch on the little menu thing because I think it's exists in this
08:23box of responsibilities that all of a sudden have kind of fallen potentially on designers plates.
08:29Like, even speaking from my own experience, that's something that previously I would make a linear ticket. I would add polish, and then I would make a comment and be like, please, this is really important.
08:38Right? And then mostly get ignored for the next, like, quarter. And now I don't even make the linear ticket.
08:42I just open up a new work tree and start working on it. And it's empowering, but also the slope is slippery because I can find myself just polishing and polishing and polishing and polishing.
08:53And so given the new capabilities, I'm curious, how do you think about allocating your time as a designer who all of a sudden may or may not feel like you can with the deploying code and pushing the PRs and stuff like that? Sure.
09:08In the beginning, it takes a long time to to do these things and to put that care and craft in because you haven't done it before.
09:15You've never done it before and maybe you just stumbled upon like John Famos' tweets and you're like, wow. Shout out.
09:23I should I should do that. And then you take that tweet and you give it to Claude.
09:28Come on, man. This is hitting too close to home. You give it to Claude and
09:32then you maybe you just hope it works or maybe you ask why does it work?
09:39And, hey, next time, can you make a scale out of this? Can you make a scale? And and you don't even write the scale.
09:43It just makes a scale. So I don't know. I don't think it's a maybe in the beginning, it takes a long time.
09:49We start to spend a lot of time polishing things, but I I think that that what are you doing? If you can get to seven out of 10 like this, you probably got a lot of time back.
10:01You just maybe don't realize it.
10:03There's a bunch of things I'm gonna put a pin in. One is the seven out of 10. One, Dan, I'm coming to you for more organizational transformation.
10:09But really quickly, I'd love to hear from you this idea again. I'm like, everybody can code as much as they see fit.
10:18You have an engineering background and, yeah, you're leading design of, you know, coding products. Like, do you think about the allocation of your work and maybe have you seen a shift over the last even couple models?
10:30Yeah. We're a very lean team on the Claude code side, I would say. And so I think I just, like, I really trust the designers on my team to prioritize themselves and transparently.
10:43Sometimes that looks like not prioritizing polish, and I think that's okay. There's a few truths that I hold in my head as we're working that I think really help guide how I see prioritization should ship. Part of it is because we're working at a lab, and so our job is to be, like, researching at the frontier.
10:58And so one thing I like to encourage the team to think about, and this is, like, an ethos in the Cloud Code and team in, like, the product team at Anthropic is, like, is it worth polishing something that's not gonna be here six months from now? Like, we're so early in this journey right now that we actually don't know what the final shape of these products are gonna be.
11:18And so is your time better spent on that future looking thing where it requires deep thought and kind of, like, a lot of thinking and, like, Cloud can't do that yet and requires, like, a really big campus to explore, or is it better spent on, like, those 30 tickets? I actually got some great feedback at one point from my engineering team that I was not spending my time wisely pushing these PRs.
11:38They're like, yeah. You're helping the polished work. But right now, Claude isn't good at design, so I am doing a lot of manual look over it.
11:45And it wasn't just, like, a great use of my time. And so I think it's hard to internalize that as a designer because it feels like now you can do it all, and it feels like it's all on your shoulders and your responsibility to maintain that polish.
11:55But we like, in the same way that now that we can do it, also your engineers should be able to do more too. Like, it's a shared responsibility to care about that polish and about that care.
12:05And so it's not all on you. You can still bug your engineer to do polish on features in the same way that your engineer can now bug you to implement something on the front end. Like, it's it's all of us and we're all in it together.
12:14Yeah. I like that a lot. I've felt that myself where it's just so easy to knock out a little task, feel good about it, and then immediately just jump to the next polished ticket, and you can kind of all of a sudden, like, six hours are gone, and things look great.
12:27But then you're kinda like, did I work on the highest leverage thing? And I think it hits on this idea of, you know, you kinda really have to be self aware around not only how you're working as an individual, but how the org is operating.
12:39And so maybe, Dan, if I can toss it to you. When an org comes to you and says, Dan, we know AI is a big deal.
12:48We don't feel like we are evolving as quickly as maybe we should. Can you just help us get to that next milestone? Where are some of the main opportunities that you look for or signals that you would use to get a sense of like where someone's at?
13:04Like, if you're gonna get this org to the next level, what does that typically look like when you're wearing that consulting hat? The main thing that I always look at is
13:13what is the CEO doing? And maybe more broadly, what is the executive team doing? Because I think for the first three years of AI, there was a lot of lip service paid to, yeah, we got to have AI or whatever.
13:28But it was always like, and we'll do an AI working group. And generally, that has not worked.
13:35And the organizations that seem to do the best are the ones where the leadership team is like in the tool all day. And that's the only way that you'll you'll have an intuition for how to manage a team who's in the tool all day.
13:49And it's not outsourceable. So that's we spend a lot of time with leadership teams actually just being like, okay, literally open Cloud Code and make something, and they they love it. But a lot of them are kind of afraid or don't really wanna do it.
14:01But I think that's that's one of the big leading indicators to me.
14:04You're a pretty good prototype for what it looks like to be the leader who lives with the tools and has hands in the clay pretty much all day is what I'm assuming. So maybe we could zoom in on your workflow for a second. And if you kind of put, you know, you today up and then over here is you maybe, I don't know, three, six months ago, something like that.
14:22I'm curious in terms of how you are interfacing with the models on some of these more, you know, ambiguous creative opportunities. Where do you see the biggest deltas in terms of how you are working, what language you're using, tools, skills, anything like that? I mean, I think there there was this big moment that happened
14:42in, I wanna say, November or December 2025 when Opus four five and GPT five three came out or yeah.
14:55Five three. It was like this moment that I think a lot of us for us at every like, we had been using Cloud Code for about we've been using it for about a year and and had not been have not been looking at code for about a year. So at that point, it was still pretty risky.
15:10And people are are you fucking kidding me? You're not looking at the code? But I think we could see that that's where it was going, and we were willing to deal with some of the weirdness of that.
15:22And for me, because I'm writing and I'm in meetings and I'm doing all this stuff, like to vibe code, but I'm not going to do any serious engineering.
15:33But when that stuff came out, when Opus Opus four point five and five point three came out, I was like shipping PRs to our products.
15:44And I was like, holy fucking shit, this is crazy.
15:48So I think there that that has been a big shift because the models are just much smarter. They're much more independent.
15:57They don't do as many dumb things. And I think we've also figured out like Cloud Code, you guys created this.
16:06I think a lot of people had models of what does an agent that does work for you look like? And prior to Cloud Code, there was like the Devons of the world or even Codex.
16:17The original Codex was it was like you have a sandbox agent in the cloud. And what what was really special about Cloud Code when I first tried it is like, no, no, it's just on your computer. It has access to all the things that you have access to.
16:29And that just opens up this whole new territory. And I think as that has started to catch on and as the models have gotten better, it it has become this new work operating system in a way that I think is really, really special.
16:43And so so the big power ups are, yeah, I can you know, that that app proof. Like, I just made that in between all the other stuff that I do.
16:54I have shipped PRs to, like, pretty much every one of our products, but don't really know the code base. Sometimes, me shipping PRs is actually really bad. So it can be it can be annoying.
17:07Your mileage may vary. Like I said earlier, I I struggle mightily to respond to emails on time, and I'm now just perfect.
17:17And it's it's great. That's really crazy. Like, all the things that I would normally procrastinate on, I just it just gets done.
17:25And, you know, I have a lot of I have a lot of thoughts on, you know, just generally what happens to experts and expert workflows and and where where to spend time.
17:34But I I think the my whole day is very different because I'm just in Codex or Cloud Code all day, and I use I do all my work in there. I wanna talk about this phrase AI fluency,
17:45which feels like it's probably the next buzzword buzz phrase. And let's just, like, increase the fidelity of what that could look like, what good looks like inside of an org.
17:56And I'm gonna start with you two because you shamelessly plugged your designer roles. And let's imagine that somebody is somehow inspired by your pitch and they wanna join.
18:07Right? And maybe we fast forward six months, and my question is, like, what is good look like?
18:16Right? Like, what is a behavior that they would be doing on a regular basis that would get you to point at them and say, yeah, that that is what AI fluency looks like.
18:25That is what we were looking for when we were putting this design role out there. If in six months,
18:32I can look at you and say that you have found a way to cut out all of the noise and truly think about what what matters most.
18:46Right? It's learning. How fast can you learn?
18:48What is the iteration of you learning and then acting upon that? And then what what do you do with that learning, and how do you proliferate it throughout an organization that is now at, like, terminal velocity where everyone is doing things and putting things out?
19:02Um, there's this this this idea that, um, you know, we can't, uh, alignment doesn't exist. I agree with this, by the way.
19:10But the the cost of alignment is is quite high, and the cost of redundancy is low because we can kinda clean stuff up. But I think if in six months, you are the one who is able to understand across your your whole product very deeply.
19:29Right? How you can simplify all the way down from you know, we have 15 things that solve 15 problems to we have four things that solve 30 problems.
19:41I think to me, that means that you're leveraging the tools because I think the reason that we don't get to that today or we do now, but we've never gotten to that is because we spend a lot of time doing administrative work or figuring out how to set up calls with researchers.
19:59And since we did shamelessly plug, uh, RAMP, my time I've I've been at RAMP for four months now. And, um, my first week, I met with the research team, and I've never scheduled a research call my whole time at Ramp.
20:11Every week, four of them pop on my calendar, and they tell me exactly what I need to talk about, and they know what I'm already working on. And I just go, and I just show up, and I just get to learn from customers. And I get to hear about what they do and what they care about and what resonates in their world.
20:25Um, and then I get to apply that across the entire organization because, um, I don't have to read all those docs. I can just synthesize them.
20:34And and I can understand and and look very deeply at I don't know what comes next. Where are we going?
20:44And I think it's AI fluency to me. Just just carve out more. Be be be selfish.
20:49Be selfish. Okay. So we have systems thinking, the importance of internal workflows, something I've been feeling a lot recently too.
20:57Dan, I'm curious to hear from you. You know, you've let's say you find the senior product designer in this room. Six months from now, what would get you to the point where you're like 10 out of 10 higher?
21:05Overall, when I think about people who are who are experts at at what they do and in in AI, I I sort of split up,
21:12and and we can say maybe designer specifically. I split it up, but I think this applies just as well to engineers or product managers or whatever.
21:19I split it up into two buckets. One is because these tools make competence cheap for everyone, there's huge glut of people doing design work in every that are not designers.
21:33So a good designer is going to figure out how do we how do we make systems to harness that design work and use it rather than like be like, no, that sucks or we we can't use it.
21:46Right? But that's actually hard. There's a lot of systems to build on that side.
21:51And then the other thing, the other bucket is, how do I use these tools to make something that no one has ever made before? And the first one is, you know, there's all this there's all this routine work that I can I can sort of automate and I can harness the routine work that all the other people in the organization are doing?
22:09And then there's all this other stuff I can do that was just would just never be possible otherwise. So and and I think the kinds of people that like to do that are very curious.
22:20They're very playful. The the kinds of people I love to work with are multidimensional.
22:25Everyone inside of Every is like, you know, Kieran who who runs Quora, our email agent was like he's, you know, super technical, but he was a composer and a baker before this, like professionally.
22:38And I think that kind of person, all these tools are just this amazing playground. And you can what I love is when the design team's like, yeah, I'll just go I'll just go make a little app for that.
22:50And like, you know, and often it's an it's a little internal tool or something like that. But sometimes it's like, you know, the that that graphic I showed you with the the Zenos Paradox graphic.
23:02Like, I just pulled Daniel, one of our designers in, like, the day before because it was something got fucked up and I I just needed someone to do it. And like, we just got it done.
23:11And it's really good. So there's real there's like there's something about working super fast on new things and being like really excited about that and using the tools to harness what is now possible and like push it to a level that wouldn't be possible otherwise.
23:31One thing that Dive Club has made abundantly clear to me over the last year is that the practice of design is changing and the old process of getting feedback just doesn't quite cut it in today's world. That's why I'm excited to announce that in flight is officially in open beta. It's the feedback tool that I've always wanted, and it's built for a world that moves at the speed of AI.
23:53So I can share my prototypes, give context and video walkthroughs, and in flight makes it easy to get the exact feedback that I need to move forward, whether it's voting on directions or maybe even getting the green light to ship a new idea. And all of this is available in a single link that I can drop into Slack or maybe even share with power users to test out a new prototype.
24:15I use in flight every day and it's totally transformed the way that I share work. So I'm excited for you to try the product and if you ever wanna jam about it, just email me at rid@inflight.co. I wanna return to something that Bradley said, but maybe I'm gonna have you talk to it because I I think that you've been thinking a lot about how to learn as an individual with the models because a lot of what you're talking about is, like, know, being curious, trying things, making sure that you're continuously improving as an individual.
24:42And so I'm curious when you think about your journey as somebody who's constantly tinkering with these models, how have you been able to grow the muscle of making sure that you are learning as you use these tools?
24:56We're also hiring, in case that wasn't clear.
25:00But I think, like, we are so early.
25:06I cannot emphasize this enough to everyone here. We have really only solved two use cases, and the use cases of everything that can be solved right now, search and coding.
25:16There's so much else out there that to assume or to think that we're, like, anywhere close to the end right now is, like, we're just not. We're just so far away from that. And so, like, I operate every day imagining, like, thinking in the world, like, this could all be obsolete.
25:31Like, every pixel I push, everything I'm trying today could be obsolete next week depending on the next model that comes out, depending on, like, the next innovation that comes out. And I think that belief that, like, there's nothing really that I that you need to, like, hold too dear, and you need be willing to update your belief system constantly lets you be very exploratory in the different directions that you're going, but also, like, leads to you wanting to learn more different directions.
25:56Like, I think that's truly how we need to be living today because we're just so early on. And so once you, like, embody the idea that we're, like like, we're so not near the endpoint right now, we're, like, 1%, and we all have the power to shape the next 99.
26:15It becomes, like, an opportunity. It becomes exciting, actually. It becomes fun.
26:18And, like, it be and, like, everyone's next 99 is gonna look so different from each other because the power of these tools lets you be so customized that I think it just gives you the opportunity to learn and focus on the areas that you want to learn and focus on. And so it just it just becomes fun and interesting.
26:37Like, learning for the sake of joy, I guess, is part of it. Can you talk a little bit more about what it's like in your position where
26:44you you really are having to operate in the future? Right? Like, you're designing for maybe the next model release or the next six months set of use cases.
26:54How does that change your daily, weekly practice of design? Probably in two fundamental ways. The first is that I think because we are, of course, still shipping products today and we want our users to have a good time, we need to be able to hold two truths in our heads simultaneously.
27:10The first is that we want the products that we ship today to be excellent and as good as possible. We wanna serve people, and then we also have to believe that the products we have today are wrong, and we can rethink them from the ground up.
27:19And so you flip back and forth in between these two all the time. And so when I'm looking at the future, I constantly imagine what is annoying to me today. What do I not wanna do anymore if I'm building for myself as much as I'm building for you all?
27:33And what does my team complain about? What like, when I talk to people, what are things that they don't wanna be doing more of? And also, what are the things that they do wanna be doing more of?
27:42And not just the people who are talking and showing it, because I think it's very easy to talk and show. But when you observe people's lived experiences of how they're using these tools, like, what do they look like they're enjoying doing, and what are the parts that, like, aren't fun?
27:54Like, if you were to tell me last year that people were, like, having four terminal windows open, I would have been like, that's crazy. Like, who who wants to live that way? And then we saw our team do that.
28:02And so I think a lot of it is, like, through your observation and your experience and then building, like, based off of, like, the peers around you. So it's actually a lot of fundamentals.
28:13Like, you you hear this like, wow. That's design. Well, yeah, because these core design skills are still extraordinarily important in our day and age and, you can keep building on that foundation.
28:20You just can test new things faster now. Okay. So there's this thing that I'm experiencing where
28:26I'm interfacing with these models, I'm figuring out the thing that I don't want to happen twice. Right?
28:32And I have Claude or Codex, make a skill. Right? And then you just kinda dump it into this black box and you hope that it helps everybody and the entire process of collaborating on top of the learnings that we're all having as individuals has been largely opaque for me.
28:48And I know that you are operating in orgs where you're you're taking that set of problems pretty seriously. And so maybe, really quickly, can we go down the line? I'm kind of curious.
28:58What are you seeing or doing or experimenting or having success with in terms of making sure that the learnings that you're having as individuals are propagating to the rest of the org and benefiting the larger team? I do think it's very isolating working with these models.
29:11Let's just all admit it. Like, if you spend eight hours of your day talking to Claude, you haven't talked to another person and it can kind of be a little strange. We're also, as I mentioned, very early in this journey, so a practice that we have that we're doing on the Claude Co.
29:23Team is to pair for a while. What a concept. Pairing with another person.
29:29Is that even legal? Yeah. Like, how often do you wish you could work with another designer on a project and you can't?
29:34Well, what we do is, uh, we'll schedule an hour. We'll shadow each other.
29:39You'll just literally work on a thing that you're already working on. You'll see what another person does because it's actually really hard to explain your workflow to someone. When I was trying to do this demo, it's, like, really hard for me to come up with what I'd talk about.
29:48But if you just watch them in their actual workflow, they'll do things that, like, you don't even realize that that they don't even realize are special, that they're learning because they just did it so many times now. And then, uh, we do those every month.
30:00We cycle through the design team. It's still a little bit of a newer practice, but the engineering team has been doing it for a while because pair programming is pretty common. And it just helps you learn from each other.
30:09Like, we are all experts to learn from right now. And it also builds, like, a really important camaraderie of working on a team. So that's something that we've done, and it's been quite successful actually.
30:17There's no great solutions yet. Like, we have we have
30:22an internal skill library, for example. But then it's always like, is this skill up to date?
30:27And should I download it? And or should I just make my own? And skills turn out to be pretty personal in a lot of ways.
30:33And so you can download it, but then you have to customize it for yourself anyway. And there's a there's there's something that needs to be solved there, think. The The thing that feels the most solved right now for this kind of thing is actually just Slack agents.
30:47We have a lot of those. Some of them are things that we build internally, and it's just like a Claude code Mac Mini, like connected to Slack.
30:57Or we also use this thing called the Victor, which just raised a bunch of money. We also have our own agent agent that we're building.
31:06And the the really interesting thing about doing that, especially if your organization enforces and we we don't enforce this, we highly encourage people to use it in public channels is you get to see other people prompting.
31:19And that is a surprisingly intimate thing right now. You're kind of like, this is this should be between me and my AI. So I think that's a really interesting way to spread stuff because what you do is you get people will have ideas.
31:38Like you said, people have ideas that you never would have thought of. And they will do things that you never would expect. And then that just seeds something in your brain like, oh, I'm going to do that.
31:46So I think that's really cool. And also, if you have multiple agents in your workspace, they actually share things really easily and really quickly.
31:56And that's also really cool. So one of the things I have is I built this little thing called Mailroom that gives my Codex an email address.
32:06And it's but it's it's it's on my email. So you know how Gmail has like the plus format? So it's like Dan plus Codex and there's like a little string.
32:14And Codex just knows to check that email. And then the agents that we have in Slack know that that's my Codex's email. So they just like email stuff that I need to do to that Codex.
32:22And then it just does it and I get it. And that's like another interesting agents are so fast at sharing things.
32:29And I think that that solves a lot of this problem once you start getting it to work. What about you, Bradley? What about collaboration,
32:35knowledge transfer, things like that? I mean, I think I I would harp on the the the Slack bots. When I when I joined Ramp, like, seeing people that was only four months ago.
32:47Seeing people use Inspect in public channels, in every public channel.
32:52Something's broken, immediately somebody's on it. They just tag Inspect.
32:58In fact, like, there's even a funny funny instances where, like, someone will say, like, oh, this thing should be different. Something's wrong.
33:07And then, uh, there's there's, like, a a pattern where someone will come in and, like, drop a link to something. In this in this case, it's like they just add inspect, they don't say anything else.
33:18And it just reads the context of the thread, and you see you see it work. And, um, that's interesting. When I joined Ramp, um, I I kind of felt silly because I was talking to a bot for a really long time, and his name's Cody.
33:34And I was like, wow. People here are so nice. He's so generous with his time.
33:39Yeah. I just I so, like, he just anonymous PFP, and I was like, okay, cool.
33:47He tell me everything, and he he's he's he's speaks naturally. He's all lowercase.
33:55And and then and then when I realized he was a bot, oh, I felt like a crisis. But I I I then started to see him evolve actually in front of my eyes.
34:05I started to see him have a blog. I started to see him teach other agents how to to work and then how to teach their humans how to work.
34:15I started to see him innovate on patterns where it was like, oh my gosh. We have now everybody wants a Slack bot. Everybody wants Cody.
34:22Oh, that's kinda unsafe, uh, by the way. Uh, and Cody's like, oh, don't worry.
34:28I got it. I saw this, and I'm gonna fix it. And then he helps us fix it.
34:33And then, um, this is what's crazy to me. Um, I learned something. I learned something great from from Cody, um, and from the people building him.
34:42Uh, and I touched on this. There's a lot of noise. Right?
34:45And and one of the most important things I learned was, now we have new ways to cut through that. I saw Cody drop an album in our Slack about everything he was doing.
34:56I wish I could play it for you. It's actually pretty good. And it is all about the things that we are learning and need to do maybe differently.
35:06And and then he started posting videos. Somebody gave him Gemini and Nano Banana and oh my gosh.
35:17Every day, how do you do this? Don't worry. Let me make a song for you.
35:22And it's kinda catchy. I don't know. It's it's it's yeah.
35:25I think it's the Slack bots for me. And then we have a fantastic design programming, right, where, you know, learning is is not required, heavily encouraged, but it's starting to shape many different shapes and forms, and it's, like, very applicable.
35:44It's like, go learn a thing over here, but you don't actually ever use it. That's not that helpful. We understand, like, where things are difficult and where we're having trouble.
35:51So let's, like, sit in a room. We actually sit in sometimes in here. Sometimes we have, like, a smaller version of this, if you can believe it.
35:59And we all sit in there. And Elizabeth here shares her screen.
36:03And just we we just see it happen. Or I see somebody do something, and I walk up to them with my little microphone. And I go up to their desk, and I I I know they use looms.
36:12I turn the loom on. And then I connect my Bluetooth, and then I I connect the microphone, and then I ask you, can you show me your prompt?
36:18And can you walk us through it? And then we post it in the channel and everybody sees it. It's interesting how and this is kind of in a trend even on the podcast where there's so much opportunity right now for people who are problem finders, designers, to go and fix things internally.
36:33Like, the the internal workflows are so ripe, so many opportunities for little tools, solving problems, eliminating inefficiencies. That being said, now for the final question here, I wanna return to the seven out of 10 idea because it's not just internal tools. Fact is, the models are not there yet.
36:49They're not amazing at design, but every once in a while, Claude will do something on the paper canvas where I'm just like, dang. That was pretty good, you know, almost uncomfortably good.
37:00And you can kinda see where this is going. And so given a world where the models are playing an increasingly large part of our process when we're designing the interfaces used by real people on our real product, how do you, Megan, think about the way that the value proposition of design is shifting?
37:23My very personal opinion, and by personal, I mean, it's truly just me. This is, like, not a representation of the company that I think our models will be good enough by the end of the year or the AI industry in general to do design, most of what we consider very, like, fundamental design work.
37:37Um, and so I think there's a few different kind of directions I see this evolving. The first is that fundamental systems and brand, those are all subjective tastes that you need to be able to guide and, like, establish.
37:55And those have a lot to do with, like, the automations and the systems that you're building. Those are established. Like, you can prompt into them, but, like, when it's really, really an expert hand, like, crafting it, you can really tell, and I think that will still be very important.
38:07And so a lot of work will go into building those systems that help models design and help everyone get access to these. Uh, the second one I think is that we're gonna enter an era where personalization and customization is the name of the game.
38:19Uh, you can already see it in all the demos that you saw today where people really love building custom tools for themselves, But I have a very strong hypothesis that there's limits to how much people wanna customize. You always need a great canvas.
38:32You always need a great starting point, and then we need to have a flexible framework to show people what should be fixed. Like, you don't want your login screen to change every single time you log in, but you might want your dashboard to be flexible as well. And the decision of what to keep fixed, what to be allowed to customize, how you guide people through that, that's, like, fundamental UX.
38:51So I still think there's gonna be a lot in there in the decision making, but we're abstracting away from just designing the UI layer into the structural layers of, like, what is fixed and flexible UI, into the harness layer of exactly how you interact with the model, to, like, the primitive layer of, like, what does it mean for a model to have an identity or the product you have to have an identity.
39:13And I think what we're gonna see is that, uh, builders, designers, engineers in general will end up going deeper and deeper down as, like, more and more of these these higher levels are solved. I hope I can speak for everyone here. Like, it's a fun time to be a designer, and there's just so much to learn.
39:28There's so much to evolve
39:30and dive into, and I really, really appreciate all three of you joining. Uh, do you know, look up to all of you and the impact you have on the industry, and it means a lot to be able to pepper you with questions. Let's just give them a round of applause.
39:42Thanks, everyone.
The Hook
The bait, then the rug-pull.
A live audience at Ramp HQ in New York City. Three practitioners from Anthropic, Every, and Ramp seated in armchairs on a stage with microphones. The host has not prepped his guests. What follows is forty minutes of unscripted thinking from people who are actually building the tools, shipping the code, and living the org transformation that most design panels only theorize about.
How they asked for the click.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
