Members-Only
Recent Talks & Demos are for members only
You must be an AI Tinkerers active member to view these talks and demos.
January 26, 2026
·
Dublin
How building with AI showed me where all the design decisions were hiding
A non-coder designer built an AI SaaS platform using Claude Code, Codex, and Supabase, detailing agent orchestration, feedback loops, and product decisions.
Overview
I’m a product designer who can’t code - but I’ve built an AI-powered SaaS platform using Claude Code, Codex, and Supabase. I’ll walk through the actual implementation: agent orchestration, feedback loops, memory systems. Where I let AI apply judgement vs kept tight constraints. And the product decisions I didn’t know existed until I owned the full build.
Video
Transcript
Generated 4 months ago
Summary
Generating a talk summary...
View full transcript
Joe, yeah, nice to meet all of you. I'm Tiernan. I'll give a bit of background about myself before jumping into the project. I'm a product designer. I've been doing product design, unlike traditional product teams and in consulting, for about 8 years now.
And the last 2 years, I've been kind of exploring my in my own side projects and kind of changing the way that I work, on client projects and that type of thing just to kind of feel out what this new way of prototyping and building out Tiernan prototypes is. And and naturally, that brings you closer and closer towards the development process, and I've found myself kinda getting to the point time and time again. Like, well, why stop here? I can go that little bit further. I can go that little bit further.
And I have no technical background. I've been playing around with the AI tools, for software development for for a couple of years Joe. But, like, I I can't write any Joe. And I think that's just important saying because I'm about to speak by crabs for a moment. Haugh anyone heard of the term and by the way, this isn't slides, that's a code base.
Has anyone heard of the term, carcinization? Yes. What is it? Everything becomes a crab. Everything becomes a crab.
Exactly. Yes. And that's the vibe. That's the vibe. And we're all about vibes up here today.
Joe so, yeah, karystinization is what we know as a crab. There are actually many different evolutionary paths that lead to that crab. So I think there's 5 distinct paths, and nature has decided that each of them did for 6. I know. Right?
Yeah. I'm gonna be the sixth. Nature has decided that all of them need to turn into this thing. And I think the the point that we're at right now in software development is we're all kind of starting from different starting points, and I think we're merging towards a crab somewhere. So I'm gonna speak a bit about my my journey to to becoming a crab and how I've been working as a designer in these tools.
And I'd love to hear more about your guys' experiences if you think you're also becoming crabs. See, there's not a term for it yet. So crabs. So so the way that I work, I think because because I can't write code, it means I need a lot of process around creating the code. And because I'm a product designer and, like, from a research and design and product testing background, I like being involved in a lot of the decisions and stuff.
I've gone through a lot of different iterations in in, how I work. You don't need to be able to read what's on the screen here. I'll I'll just speak to to what's on it. What I have found being most effective for creating production quality code is using the CLI tools is incredibly useful. Rosinskyi this terminal I've got Claude Joe with Opus 4.5 And then over in this terminal, I've got codex with 5.2 high.
I use, Claude Code as, like, my daily driver. I think of it kind of as my product team, and I'm working through, like, a new feature with Claude Code. I then build a ticket with Claude Code, and I create an actual markdown file. And then I'll go over to codex, and I've given that basically a a a role or a prompt around, being, like, the the lead engineer. So it will come in, and it will review the ticket, and, and it will provide feedback.
So so that's the stage that I'm at here is, getting feedback from my lead engineer and basically copying and pasting it across here and as input, to work on the ticket. And then I'll go into the planning stage and, again, do that review process. So I'll be very much involved conversationally with Claude Joe over here. And then when we get to the point of, getting into the more technical detail, that's when I'm I'm bringing in my quote, unquote engineering lead. I'll come to your question at the end, but that's alright.
Because I also want to show you a bit about, just do a bit a bit of a demo and kind of the tech stack and show that you can build quite complex stuff in this way. So I'll quickly do an an onboarding of the platform. So it's called Many Futures, and, like, the the short version is, it allows you to you you create an account on it and it spins up some research agents and it creates a daily briefing for you about your context. So if you're working in a certain industry and you need to stay on top of change and technology, it will serve up kind of like a a daily substack newsletter for you about that change. So so I I like, old version of me, like, the product experience that I'd be designing there, it would be in in Figma, maybe doing some flows in Miro, but kinda limited around where I can actually make design decisions.
Because something like this, the experience is in the quality of the content. Right? Like, what it actually looks like on a page doesn't matter so much. And so when we create an account, you'll see I I want to make it as easy as possible to, like, get as much context about the user as I can. And so I'm just gonna drop some stuff in here so I can show you the actual platform.
I'm saying here that I'm working on a train at a train operating company and basically looking at what the future state might look like. And then a bit more, it doesn't really matter, this can be anything. The first thing that happens here is I'm using, has anyone come across Parallel? It's kind of like Exa AI. Joe it's got this feature called monitors, which essentially sets up a monitor based on a search query.
And the cool thing about it is that it when new content hits that search query, it hits my platform with a webhook. And so essentially it is like like a search API. Yeah. Essentially. Yeah.
Yeah. Yeah. I wouldn't have been able to answer that a few months ago, but yes. So here we can see what those kind of monitors are getting set up. Again, I'm using what what what was really interesting going going building out this experience is, like, where do you put in I'm building an agentic platform, but where do you put in the the LLM and and the AI stuff?
And already here, just in the onboarding process, it's doing quick Haugh goals and to Claude HiQ to to, populate what these jobs might be. This is actually this account has already been set up. I just cleared the, kind of, onboarding through, so that we can then come in and see a mature account. What yeah. When it how it works and says it doesn't I had I didn't even know that was gonna come up.
But but you'll see here it it's, it's captured stuff for a feed and then, the agent builds out a newsletter based on the the content that has appeared over the last x amount of time. When when I was building this out, I I came across, like, a a few different, kind of concepts that I had to get familiar with. So so the first thing that I realized was the the, the, like, news items that would get returned by monitor, sometimes they weren't that good. Sometimes the it was maybe like a paywall it would get hit with or the the content just wasn't that relevant. And so instead, what I have now is those items those links populate a database inside SuperBase.
Shout out to our sponsor. And and then another another CronJob sets off a second, API call that gets parallel. Again, it it fetches the content from the site, and then Haugh reviews all of the content and, writes up, like an evidence review. Is it usable? Is it not?
So you'll see that here. And then a little annotation. So what what what I'm doing here is kinda cleaning up the data to pass over to the agent, which is like the the writer, kinda modeled on the newsroom. And then I'll just jump into my railway logs for a moment. You'll see the cron jobs here, at the time that the user is chosen.
So, like, 06:30AM, a cron job hits and, it sets off the the briefing. And so then I'm I'm using the Claude agent SDK, which is essentially Claude code but deployed in the cloud. And so that gets, okay. What are the latest news items from this user? What is the user's context?
What is my role as, like, a writer or editor? And then it creates the content and pushes it to the web. Joe, yeah, that that's that's kind of all that I wanted to show. I think I think the interesting thing for me there's there's 2 bits. There's, like, I think not knowing how to code is kind of an advantage because you need to build up the process around it.
That's my story and I'm sticking with it. And then also there are so many design and product decisions kind of litterate that kind of full product development flow, which normally, like, I wouldn't be close to at all. So it's been, it's been super interesting kinda owning that end to end process. Thank you.
Tech stack
Finding related talks...
Compose Email
Sending...
Email preview
Loading recent emails...