Small miracles

Dave Nunez is co-founder of Falconer, an early-stage startup that helps companies automatically create, organize, and maintain their knowledge. He previously led docs and knowledge management teams at Uber and Stripe, co-authored Docs for Developers (Apress), and advised companies including Anthropic, Graphite, and Metronome on docs and developer experience. 

We originally had a version of this conversation in 2025, which was published in a print resource kit for Mercury Spheres that May. Falconer launched publicly in February 2026, leading us to revisit our conversation and discuss these topics anew a year out.

Sid Orlando: You co-founded Falconer with Maxi Benedetto, but you've also co-built two consultancies with others. Was there a problem that kept showing up across your work that made you think, “This needs a product solution”?

Dave Nunez: I’ve advised a lot of fast-growing companies early, like Anthropic and Graphite. I found that the same problem kept showing up: We would experience very fast growth but knowledge transfer would break down. This problem is more acute than ever today, with exponentially more code, content, and complexity involved in building companies.

At Stripe, we solved knowledge transfer with talented people and huge time investments. These are luxuries most companies can’t afford, and a lot of them don’t know how impactful the results can be. I wanted to build a product that made the importance of solving this problem immediately evident.

When I was at Stripe, this was a human-to-human problem. Now, Falconer has to look at ways that even humans and agents share knowledge with each other, which comes with all these fun, science-fiction problems to solve. 

If we imagine a spectrum where one end is “building what the market clearly wants right now” and the other is “creating something the world doesn't yet know it needs” — where do you think Falconer falls?

We’re definitely building what the market wants right now. The industry seems to have simultaneously come to the conclusion that context is the bottleneck, and that deploying agents at scale without proper context is a surefire way to yield hallucinations.

So, yes — a great knowledge base is one of the highest leverage resources a company can have. Everyone knows that. But when the average tenure in Silicon Valley is about 2.5 years, why would someone put in the work for a knowledge base when it’s not core to their job description? 

What we’re saying is that now it is a core part of company building, even if it’s not your job. A subpar knowledge base is an immediate issue. Your agents suck without it, you spend time spoon-feeding them context, the context window blows up, and you’re back to where you started. The more your agents know about your business and intent, the smarter they become.

I’ve had this vision of autonomous documentation for 10 years before starting Falconer. I saw how good the testing harnesses were for code, and wanted to apply that to documentation. I made some cool point solutions over the years, but nothing that scaled to hundreds or thousands of documents, not until having access to AI.

...Which leads naturally into my next question. How have the advancements in AI and LLMs specifically over the past year impacted your development of Falconer?

Our whole team had bought into scaling laws. We believed that future leaps in intelligence were coming, and that they were coming soon. We pored over the frontier research and worked with early customers that had preview access to frontier models. And so, we decided to build Falconer for future intelligence.

We spent a lot of time optimizing for time-to-first-token and time-to-last-token early on. Models have gotten much faster, which has given us a huge uplift: We don’t need to spend as much time developing our own techniques to reduce latency.

Congratulations on the GA launch, by the way. Is there anything that’s surprised you about the initial reception to the product?

We were shocked that there was so much pent-up demand from enterprises for our product. We expected to earn our way upmarket through product-led growth (PLG) first. PLG is in our ex-Stripe and ex-Meta DNA, so we felt comfortable doing it. However, our first customers were enterprises, and we learned that the enterprise buyer is an entirely different animal than the standard PLG buyer. 

Enterprise buyers are feeling pressure from boards, peers, and even Wall Street to have a strong AI strategy. Educating them about AI and our product is a huge part of our process now. We went from making pixel-perfect onboarding experiences to learning how to become salespeople and build for security and scale.

What’s been the biggest misconception or finer point you find yourself spending time on when describing Falconer versus other AI-powered tools?

Falconer has two modes: fully self-driving and human review. These modes apply to all of our features. Falconer can either do something for you without asking, or it can ask you for permission first.

The biggest misconception is that you have to choose one mode over the other. The power users want to be in the loop on all changes, and people who never have time for knowledge management just want to put it on self-driving. In reality, it’s not a binary choice. In some realms, the same user will want to let Falconer drive on its own, and elsewhere they’ll want to be very plugged-in and manage everything.

What’s a skill from your consulting days that’s proving unexpectedly valuable in building a product company?

 When you’re a consultant, the customer is not buying software that they can see and test. They’re buying access to you. They know they have a problem, and you need to make it very clear how you will solve that problem. I learned how to solve pain rather than simply selling solutions. Being clear and simple with pricing and milestones have proved extremely valuable in selling software.

How about a new skill you’ve had to develop almost overnight?

I’ve had to learn how to develop a repeatable product sales motion that we can refine and eventually scale for others on the team to execute.

In general, consulting sales cycles take 1–3 days before you have a clear yes or no. Enterprise sales can take 1–3 months before you get a clear answer.

When you come in with structure and repeatability with an ideal customer in mind, you will find about 80% repeatability, with maybe 20% improvisation.

Who’s in your “founder support system” — and what conversation do you find yourself having with them most often?

My co-founder, Maxi, has been a huge boost. We share all the wins and all the losses. We’ve been at this for a year and we’ve already learned so much. 

There are also a few founders who are further along than me who I can text or call when I have a big decision to make. Hearing their perspective and their challenges helps the journey feel more palatable.

Some of our investors have also been a key part of our support system. They’ve made us feel comfortable asking dumb questions, and they’re quick to offer expertise, act as a helpful resource, or make an introduction to someone useful.

From today’s vantage point, what does “making it” look like?

I’ve realized it takes a bunch of small miracles to get on the path to success. 

To me, success means reaching profitability at scale. We want to give every company their own knowledge factory, which means we have to build a powerful enough platform to solve all the pieces involved with this vision. And we have to do it sustainably, so we’re not just subsidizing tokens on behalf of the labs. 

When you think about the founders you admire, what’s a specific decision they made that you’ve deliberately chosen to emulate in your own journey? Conversely, is there a common “founder trap” you’re actively trying to avoid?

Making bold, quick decisions can replace lots of time stressing or second guessing yourself. I saw this up close with Travis Kalanick while at Uber. The best founders I’ve seen don’t give in to external or internal pressure. The famous story is Netflix’s Reed Hastings disrupting his own business by going hard on streaming and killing his successful DVD delivery service. It’s a ruthless approach. It might mean cutting a project you’ve put a lot of time into, firing someone who isn’t a fit, or reversing a decision you made when it’s clear it’s not working.

I’m trying to avoid the common founder trap of reinventing everything. Founders start something because they want to right a wrong. They want to solve a problem. But it can be hard to turn that instinct off. I try to pick my battles, and to not get nerdsniped into solving every problem I see. Most of the time, the best move is to go with the simplest pricing model. Or to just go with the simplest messaging that’s going to resonate with customers. Spending too much time on decisions you can always change later is a tempting trap I’ve seen many others fall into.

What’s been the most humbling moment so far in this startup journey? 

Realizing there’s a lot of tedium in building a company, and it’s not all stimulating and fun.

So much company setup has nothing to do with product-building: from incorporation to cap tables to payroll to taxes. It’s draining to do this as a first-time founder. It’s important, but takes you away from the fun stuff of spending time with customers and building the product.

How about the most happy-making?

Customer feedback has been like oxygen for us. We have a channel called #customer-love that brings the whole team together whenever it lights up. One customer even made a custom animated emoji of a hand petting an eagle, which they use when Falconer’s Slack agent gives a great answer.

There are two pieces of customer feedback I think about a lot: “the amount of time that I've spent answering teammates’ questions is just amazingly gone” and “Falconer has knocked my socks off.”

We’ve had two separate customers who’ve told us they went to company offsites and teammates kept coming up to them, saying “Thanks for bringing Falconer to us!”

We built for engineers first and foremost, but it’s even more thrilling to hear from non-engineers, like finance, sales, and HR teams who now have a tool to do their jobs better. The best AI tools require technical depth, but Falconer is dead simple for anyone at the company to access knowledge like an engineer.