Seth Watson is a self-described YouTube guy who gets extremely curious about things. Like when he came across a video about a new dinosaur species found in the Sahara Desert and went down a midnight rabbit hole using his own product to learn everything about how scientists dig up dinosaurs. What tools do they use? What's the process? Within 30 seconds, he'd extracted that information from 50+ papers.
That product is Moara — a platform that helps researchers organize and synthesize academic data. Cofounders Seth and John built it out of their economics research firm when they realized their demand didn't match their tools. They were literature review experts helping Fortune 500 strategy teams with questions like "what's the willingness to pay for one ton reduction in greenhouse gas emissions?" But they couldn't keep up.
When I talked with Seth, what stood out was how they built something so valuable they had to sunset a profitable business to pursue it. Here are three things I learned about building AI that actually solves real problems.
Librarians Rip You to Shreds And That's Exactly What You Need
When Seth talks about selling into academia, there are two groups: librarians and the users librarians serve (researchers). Librarians are extremely skeptical of AI, rightfully so; their whole expertise is understanding what data you can trust, how to trust it, keeping everything clean, and finding the data.
Researchers, on the other hand, are less skeptical than Seth expected. They care about producing sound and trustworthy research because their reputation is on the line, but they're open to tools that help them work faster.
Seth said librarians "constantly rip us to shreds." He loves it because Moara's team are technologists who want to push the needle forward and innovate. And librarians come back saying, "Look, we've been through this before. Here's the lessons from the past."
They take those lessons and build guardrails so end users have value and are also safe in the process. That tension between innovation and caution is where the best products are built.
The Hardest Part Isn't Building AI; It's Packaging Data So Humans and AI Can Collaborate
One of Seth's biggest challenges is packaging up data so it can efficiently be collaborated on by both humans and AI, with identifiers throughout the process to ensure there's no breakage.
When you ask ChatGPT to "remember that thing we talked about last week," sometimes it remembers, sometimes it doesn't. Moara solves for that context window problem using tools like Leta and working directly with Anthropic's startup team. Each feature is different. Each objective requires trial and error to figure out what available options will build what users actually need.
Their backend uses proprietary and open models. They have workers to batch jobs when users upload thousands of PDFs. They connect to external APIs and data vendors for metadata validation. Every time a paper gets added, they run something they call the "DOI and metadata finder" to validate citation counts, correct journal names, and verify author names.
It's not sexy infrastructure work, but it's what makes the tool actually trustworthy for researchers whose reputations depend on the accuracy of their data.
Listen to What People Using Your Tool Are Telling You, Even If It Means Killing Your Business
Seth and John had a profitable economics research firm where they could have kept hiring people to handle demand. But they couldn't ignore what their friends and colleagues were telling them when they started sharing their internal tool: "This really helps me. This saves me. A task that used to take 12 weeks now takes one week."
The click moment came when a professor at MIT named Nancy Rose had been trying to finish a literature review of 100-200 papers for two years. She kept passing it to students who never finished it. She used Moara and completed it in minutes.
That's when Seth and John realized: This is a valuable tool. This is something we should pursue. So they sunset their first business and went all-in on Moara.
In hindsight, it’s crazy to completely drop what was working but they couldn't ignore the signal. Even in their own work, Moara was helping them. And now they're one of the early examples of what people call "verticalized AI" — deeply specialized tools solving specific problems really well.
Building For Real Problems
After talking with Seth, I'm thinking differently about what it takes to build AI products that people actually want. It's not about having the flashiest features. It's about solving a real problem so well that people say "this saves me weeks of work." It's about building guardrails informed by the people who've seen every cautionary tale. It's about having the guts to bet on something when the people using it tell you it matters, even if it means walking away from something profitable to chase something meaningful.
Listen to the full episode of Actually Intelligent to hear more from Seth Watson about how Cursor changed his development workflow overnight, why he's experimenting with Claude's computer control on a Mac Mini, and his take on whether the "software engineer" title will be obsolete by the end of the year.
LEAVE A COMMENT