Why I am starting a hardcore tech company in my 50s

Or at least this is the story I tell myself about it

Twelve years ago I was in the process of selling my last funded company. It was a life-changing process that lasted for several months, filled with stress. When it was all over I mostly felt relief. It was the end of a six-year cycle that had started as a consulting gig that morphed into a company that morphed into a product startup. I had accidentally become a single founder. By the time the sale was over I was 41, and I was emotionally exhausted. I made a contract with myself: never be a single founder of a funded startup ever again. Never take funding unless you have VCs knocking on your door begging you to take their money, and because the project requires that fuel.

I have abided by that contract so far. In the past 12 years I have explored many topics and business areas, but at no point I thought “hey, I’m going to reach out to my VC friends for some seed money.” I invested in a few dozens of startups and had some decent exits, did some consulting, spent a whole year rock climbing as a fulltime job. I was pretty content. And then AI exploded. As someone with a background in information retrieval and 40 years of coding behind me, I couldn’t sit this one out. So I started experimenting with every shiny new technology that got released in the past year. I saw yet another hype cycle rise, which made me happy. Not so much because of the opportunities for making money, but because I love the energy of San Francisco when new things are happening.

In January I started playing with Langchain and later Chroma, which led me to a couple consulting gigs and hackhathons. I tried every public model that got released on HuggingFace. Eventually I picked up a pattern: everyone is moving absurdly fast, fueled by absurd amounts of VC money driven by FOMO. No company that I know is focusing very much on stability or reliability. That will come later, for now it’s all about getting attention and one-upping the competition with features. This of course would be fine if we were talking about an entertainment-driven business, such as social networks or VR. The problem is that the holy grail of AI are agents that will do work for us. The current state of agents is very misleading, because on one hand language models present an interface that tricks us into thinking that we are interacting with a competent human. And on the other side, there is a bumbling idiot who is driven by random chains of causality found on internet text. What makes matters worse is that companies are arming these systems with all sorts of peripheral tools so they can interact with the world. Few people realize how dangerous this is.

As an aside, there is a whole movement of AI existential risk that has a very detrimental effect on the immediate issue I’m referring to. Over the past several decades the field of information security has strengthened software against the most malicious actors in the world: nation states, armies of hackers who want to steal large sums of crypto, political organizations trying to steer elections. And now we are integrating extremely fragile language models into the flow of business applications. It’s only a matter of time until we see some epic catastrophes caused by gullible language models with unreasonable access to critical resources.

This brings me to my own personal motivations. I am 53, but I am full of energy. I spent the past couple of days bouldering v8-v9 problems in Lake Tahoe, which is something that not many twentysomethings can do. I don’t want to become a VC, or continue being a consultant. What makes me tick is not the desire for more money, I really have all I need. The happiest times of my professional life happened when I was in a room with people I liked and respected, thinking about solving problems that mattered to us. So when I started having these conversations with infosec people ealier this year, I felt that excitement. And after a few weeks of meetings with Ariel Futoransky, we launched Gradient Defense. So what is it, and what is the thesis behind it?

In the late nineties I was very connected to some historical figures of information security who launched a company called Core Security. I had always been interested in security and cryptography, starting with Bruce Schneier’s books. I wanted to join them but they were a close group of friends, and I wasn’t qualified anyway. So instead I picked the path of internet search and information retrieval. Now with LLMs hastily patched into business operations, the two fields intersect so this is my chance!

There are many ways in which you can start a company related to AI today. If you have a track record, you talk to some VCs with a plausible angle and get a few million in seed funding. No thanks. Or you could go all in with a product hacked in a few weeks, collect a few thousand github stars and raise a lot more money. Again, not for me. These are not even the MySpace days of AI, it’s more like the times of Friendster if you remember that. So our plan is to start as a services provider. Right now we are helping customers build resilient systems that use both private LLMs such as OpenAI’s, as well as self-hosted variants like Llama 2 and its ilk. We don’t need money, because we can charge enough for these services to keep the lights on. There is a war for machine learning talent, and the quickness with which people accept our rates make me think we should be charging significantly more! What is more interesting, we are starting to see patterns that will likely lead to the creation of a product. This is exactly how it panned out for my last company: years of consulting led to a repeated demand from customers, which made us build a product that everyone was asking for.

In this context, does it matter that I am in my 50s? Not really, but I thought it would make for a good hook in the title. If I can remain healthy and work sane hours (which is not hard when you had good partners and know how to delegate), I can see myself doing this while my brain remains sufficiently sharp. And I know enough bright people in their 70s and 80s to know that this is a reasonable goal. The important thing for me to remember is that this is not a vehicle to become rich, or part of my identity. It’s more like a team sport that I enjoy playing. It’s no harder on the body than rock climbing, and it certainly beats golf.

Old Hacker, credit https://www.deviantart.com/avataart/art/Old-Hacker-by-Midjourney-925085770