.png)
Sam Nadler (00:00):
Obviously a super innovative name. Daily Check-In. I vibe-coded and replicated before I… Now instead of having to read every single one, I get a summarized form. And hopefully, it keeps me more up to date, faster.
What Clay does is it takes your leads and it allows you to write customized messages to them based on specific data and calculations.
Sam Nadler (00:22):
OpenAI released what they call OS 120B and OS 20B. These are models that have essentially zero cost to run.
Sam Nadler (00:30):
Hey everyone, and welcome to another episode of Built This Week, the podcast where we share what we're building, how we're building it, and what it means for AI and startups. I'm Sam Nadler, co-founder of Ryz Labs. And this week and every week, I’m joined by my friend, business partner, and co-host Jordan Metzner.
Jordan Metzner (00:48):
Hey Sam, how’s it going? Super excited to be back.
Sam Nadler (00:51):
Yeah, another— we say this every week, but another crazy week. We should probably stop saying that. Just to give you the docket real quick—we’re going to quickly present a tool I created this week (a little pivot from our normal routine), then we’re going into one of the tools we’ve started using—Clay—which recently raised money at a pretty large valuation. Then we’re going to cover some of the latest AI news from OpenAI.
Jordan Metzner (01:21):
Yeah, like you said, kind of a crazy week. We’ve seen releases from Anthropic, OpenAI, and Google. OpenAI has a huge press conference coming up—today, the day we’re recording. Expected GPT-5. We’ll talk about that soon. But yeah, lots to talk about this week.
Sam Nadler (01:40):
Just want to say quickly—don’t forget to like and subscribe on YouTube, Spotify, Apple Podcasts, all of that. Let’s get into it. Jordan, let’s check out what I built.
Jordan Metzner (01:50):
Yeah, let’s do it.
Sam Nadler (01:52):
One of the things I love about sharing every week is hopefully inspiring operators to use AI to build tools that quickly impact their organizations.
This week, I built something super simple. It launched Monday, and I've already made improvements based on team feedback—v2 is out.
After operating Ryz Labs for four years, we have a lot of teams to engage with. One of our practices is a quick five-minute form—basically a digital standup. Three core questions:
Two goals: get teams to think about what matters most, and surface blockers early so we don’t have to wait for meetings.
But as the team grew, reading 8–10 of these forms daily became challenging. So I built a tool to summarize them and post to Slack.
Sam Nadler (03:40):
Instead of Google Form or Typeform, I vibe-coded and replicated the form in Bolt. Summarization and posting is automated. Blockers are highlighted first.
Now I can skim summaries in Slack without reading each form. Original forms are still accessible if I want to dive in.
Jordan Metzner (04:58):
So it’s a full stack web app you built from scratch. Give a little background—how much dev experience did you have before AI?
Sam Nadler (05:07):
Not much. A bit of SQL. But now, with AI tools, I’ve built several small internal tools. I wouldn’t have been able to do this before.
It’s not revolutionary, but it saves us time. Built it with Bolt, Superbase for the DB and edge functions, GitHub for the cron job. Took maybe two hours total across a few days.
Jordan Metzner (06:00):
What’s the teammate experience like?
Sam Nadler (06:01):
Simple. They select their team, fill in their priorities, what they completed, any blockers. Hit submit.
Jordan Metzner (06:13):
Any ideas for new features?
Sam Nadler (06:14):
Yeah—permissions for teams that handle sensitive info. I’ve asked them to withhold sensitive stuff for now. Also might integrate it into our other tools.
Jordan Metzner (06:28):
What about using AI to look back over the last 7 days to see if what was prioritized got completed?
Sam Nadler (06:34):
Great idea. Could definitely check if what we prioritize is getting done. Actually, this worked yesterday—a team flagged a blocker in the summary, ran out of credits for a tool. I saw it right away and fixed it before it caused problems. That’s exactly why I built this.
Jordan Metzner (07:00):
Well, great job. Super cool.
Jordan Metzner (07:05):
Let’s talk about Clay. Just raised at a $3B+ valuation. Huge in the sales world.
Sam Nadler (07:11):
Yeah, Clay is an AI-powered prospecting and workflow automation tool. You pull in leads, enrich them with titles and data, and build custom workflows.
You can use OpenAI to personalize outreach based on criteria—like summarizing LinkedIn profiles.
For example, we used to spray-and-pray. Now we target people with specific roles posted—e.g., "I see you have two front-end engineering roles open. You could save $20K/month with Ryz Labs." Much more specific.
Jordan Metzner (08:00):
Does Clay send the emails too?
Sam Nadler (08:01):
No—you push it to your CRM or email tool. We use Instantly. Clay enriches and personalizes, then you export and send through your own system.
Jordan Metzner (08:13):
Any other thoughts?
Sam Nadler (08:14):
It’s a super powerful tool. We’re still early with it, but there’s a ton of depth. You can do more advanced personalization with LinkedIn profile mentions and more.
Jordan Metzner (08:30):
So—OpenAI released two open source models: OS 120B and OS 20B. What does that mean?
Sam Nadler (08:35):
These are models with near-zero cost to run—you only pay for server costs.
OS 20B runs on most computers. OS 120B can be hosted cheaply via Groq or AWS. You can do almost everything GPT-4 can do—at scale, for free.
Imagine a bot reviewing all your code for bugs all night. Or another doing documentation. Or another one just answering support tickets. Cost drops to near-zero = huge possibilities.
Sam Nadler (09:15):
I demoed Groq Cloud running OS 120B—it’s super fast. I asked it to write 10 marketing emails for Built This Week. It did it in ~5 seconds. Reran it with no dashes, 4.2 seconds. It’s just crazy fast.
Jordan Metzner (09:40):
Strategically, what’s the OpenAI play here?
Sam Nadler (09:42):
It raises the bar. If you're building a commercial model, it better beat their free one. It locks in their moat—“These are just our free models.” Premium models will be even better.
Jordan Metzner (10:00):
Exactly—and what’s launching today?
Sam Nadler (10:01):
Hard to say—we’re recording right before the launch. But it’s expected GPT-5 drops today. Leaks suggest multiple variants: mini, nano, a "cat-based" model (whatever that means), and a new coding model.
It’ll need to beat Claude Opus to really make waves. We might also see updates to Codex, their code model. Either way—it’s a leap forward.
Jordan Metzner (10:45):
Yeah, exciting. When was the last big OpenAI launch?
Sam Nadler (10:46):
Besides the open source drop this week, we’ve had Google’s Genesis, Anthropic’s Opus 3.0/3.1, etc. But OpenAI’s been leading—except maybe in coding. This might change that.
Jordan Metzner (11:00):
Great episode. Don’t forget to like and subscribe.
Sam Nadler (11:03):
Yep—thanks for listening. See you next Friday.
Jordan Metzner (11:06):
Great work this week. Excited to see what’s next.
Sam Nadler (11:08):
Thanks Jordan. Bye everyone.
.png)
