Can't go alone. Give me a sword, axe or a bow.

Dawid Laszuk published on
4 min, 788 words

A downbeat this one is.

It officially started a few months ago. The execution plan was formed over a year ago. The expected long term result is with me for circa 20 years. Though the rest might sound like I'm giving up, I certainly am not. Just following internet customs of shouting out one's intentions, thus avoiding prolonged gossips by squeezing them into a burst. Ay, clever, ain't it?

The good

Agents Bar is growing nicely. The driving idea, or the closest milestone, is to be able to train hundreds of agents by having them compete against each other in multiple environments, all online. The difficult part is tested but I'm a few weeks shy from publishing this as a service. So far Agents Bar allows managing agents and (in a couple of days) environments. It will shortly allow creating experiments, i.e. run online training on co-located agents and environments, and later support for league, i.e. evolve agents based on their performance - clone best, prune rest.

Development is super enjoyable. I'm applying Sid Meier's Civilization development strategy of 1/3rds. Between their series they keep 1/3 of features, improve 1/3 and the remaining 1/3 is completely different. In my case, that's 1/3 of tech that I've been familiar for long time (e.g. Python, Docker, Linux, reinforcement learning), 1/3 that I used a bit for some time and wanted more (e.g. PostgreSQL, PyTorch, GitLab, RabbitMQ) and 1/3 that is completely new (e.g. Kubernetes, Terraform, Vue, Golang). Ok, it isn't exactly 1/3 each but it's to say that depending on the will-power I can either jump to things I can just do or learn something new.

The Bad

Agents Bar is growing in vacuum. Right now it is shaping as a MLOps for Deep Reinforcement Learning but whether that's something needed, or a basis for a business, it is questionable. Reinforcement learning (RL) is quite specific and often it is a solution to something that can be approximately solved with other methods much quicker. I need it for my eSport team trainings and IoT swarms but I don't know if anyone else needs it and need it now. Reaching out to people and asking is both time consuming and demotivating. Assuming that 1% of people who could use RL are actually using it that means 99% people won't, and those companies are in minority of all companies. All in all, hunt is more than one person task with an unknown result.

When I left Amazon I wasn't planning on starting anything. I thought about it but I wasn't planning. Only after discussing with some folks and having a few "maybes" on joining forces, I thought that it might be worth a shot. Unfortunately, probabilities of independent events aren't additive and rare events do happen. No hard feelings towards anyone :) And I'm to blame as well as I'm in for the idea/goal rather than trying to start a successful company just for the sake.

The Ugly

Agents Bar requires more work. Physics 101: Work is the sum of all forces (pushes) over distance. A big work can be done with short but forceful push, or stretched along the way many smaller pushes. I've done a big push already but there's much more required and going like this isn't sustainable. I'm missing a lot ability to collaborate, the feeling of working people around me, and having a common goal. These lacks won't break the system but they do make the whole much more leaky. To fix work env feeling I might join a co-work - hopefully there are some with people who work, not just wannabe big shots. But, to collaborate, at this point, I can either hire someone, find cofounders or be hired. Although I've secured enough for myself, paying someone needs a solid reason. And, as mentioned earlier, finding a cofounder is also difficult.

That's a long way to say that I might need to find a job. Yes, I could start a promising company with a few clever people but not now. If I have to dedicate a lot of my time to something which isn't not my goal then I'd rather be compensated for it with more than hopes and wishes.

Do you think I might be worth recruiting? Do you know a good role for me? Do you want to buy or collaborate on Agents Bar? Let me know. Here is my CV and contact details.