Table of Contents

Auditing is for outliers

So, about those Solana articles… Well, that didn't quite work out as I expected for a number of reasons.

I did do a few more audits, actually just now, as I'm writing this, I am finishing up one audit on a Solana protocol for structured products with Cyfrin.

But over the past roughly eight to nine months since I first started doing auditing with Guardian, I've gotten maybe four or five audits. Collectively, we're looking at definitely less than 10 weeks of actual billing, which I realized is just not very feasible even if you do get high weekly rates.

And I don't get high weekly rates. It would probably take me at least one, if not three more years of learning and grinding, to get to the point where I can command the $5,000 a week, $6,000 a week, $10,000 a week that you see other people do.

I don't know if anything that I signed makes me unable to disclose my weekly rate at these firms publicly, and I'm too lazy to check, but what I can say is that I've been getting better weekly rates doing relatively normal full stack typescript work.

This is not to throw shade on these firms, just to be clear. I enjoy working with them a lot and I think they pay me fairly for the skill and reputation that I bring to the table.

Still, I have come to realize that the economic incentive just isn't there (anymore?), at least not for a relatively normal guy like me. You have to be an outlier to have a 70-80% utilization rate at a high $ rate.

The market didn’t help either

What's worse is that the market has not been particularly good for the past six months either. I hear it from lots of people in crypto: dev shops, recruiters, auditors, even meme coin launchers.

I myself got burned by a client (dev work, not auditing) for January and February, and then didn't find any paid work until the middle of April. During this period, I obviously interviewed for jobs at crypto companies as well. I have learned that companies have become far more nitpicky, and interview processes are now more like gauntlets than anything else.

Some companies still know what they want and they do a good job of interviewing people and of providing feedback, and one of them is Open Zeppelin. They've been nothing but amazing in their process. They really liked me too. They told me so. And they explained exactly why they rejected me and asked me to apply again some time.

But for every Open Zeppelin, there are a whole bunch of companies which will put you through four, five, even six rounds. At any particular point in time, when they deem that they do not care to interview anymore, they will just ghost you. Maybe, if you're lucky, they might send you some sort of an answer in two weeks, but don't expect it to contain any usable feedback.

Pivoting worked

Over the past year and a half, I've built various small AI/LLM projects. I've built Minerva, which is a RAG system aimed at people that want self-hosted systems. I've built the "job finder", which I demoed on my YouTube channel.

And over the past few months, I've applied to various jobs in the AI space. Most of them ignored me. I think there might be a stigma about having worked in crypto for so long.

And also, AI companies are less remote-friendly than crypto was three years ago, but not less remote-friendly than crypto is right now, because I think the sentiment towards remote work has changed in crypto a lot in the past six months.

Ultimately though, I found this job posting in a Slack community where this fractional CTO was looking for a full stack dev for an LLM-powered product. I applied, he liked my job finder, and he gave me a shot. After a 90-minute round, just one round, he basically hired me. Turned out he’s a pretty cool guy too.

So I guess two quick lessons from this are that side projects still work and that it's far easier to find a job in a hot industry?

TradFi is not that bad

Now, since I've moved to work with this AI company, obviously they're not paying in USDC like I've been used to. I had to set up a real local bank and take payments the old TradFi way.

I have to wait a few days for the wire to come through. I pay slightly higher fees for the various transfers and currency exchange that has to be done. And, yet… I love it.

It is a much better experience than taking payments in USDC. I don't really have to care about OPSEC. I don't have to read addresses or make sure that I signed the right thing, or remember some password. There is virtually zero potential for irreversible money-losing mistakes.

I just use Bitwarden to log into my accounting software, pick the client, put in the amount, and the email gets sent. By the end of the week, my money is in the bank and I can spend it. I can just spend it as is. No further conversions or transfers. It is beautiful.

This has really opened my eyes. If I, who have been in crypto for almost five years now as a developer, when presented with the "old school" payment rails, I am impressed and, more importantly, relieved... That says a whole lot about the state of crypto.

You should absolutely try using an AI harness

Now, as always, I do want to leave you with some knowledge, teach you something new. Another thing that has really opened my eyes during this period is seeing how these guys at this AI company work.

Previously, when I was working in crypto, obviously a lot of people used LLMs to code. Even I used them to some degree, but I never quite loved it. If I let the AI rip, so to say, it would produce garbage: garbage code, garbage git hygiene, just weird, stupid architectural choices. You have to realize I was just using Claude code as is, or before that, Cursor, or any of the other solutions in the space.

Ever since I joined this company, I saw things like skills for everything (committing, opening a PR, creating a Linear issue, etc). I saw adversarial agents each with its own role reviewing the plan or the code you (or your other agent) wrote. I've seen MCPs for everything: Github, Linear, observability. People in the AI space call this a harness.

The output of the LLM models became better, not great, but better. But it's not even that. It's just that now I have to spend a lot less time reigning the AI in, because due to the harness, a lot of times it reigns itself in.

Another really cool thing I discovered is worktrees. I can now spin up two or three agents working on different issues. While I guide or review one of them, the other is working.

This way, even if, for some types of issues I don't think that coding with LLMs makes you any faster (in terms of time elapsed from triage/idea→shipped) than coding by hand, just the fact that I can work on a few of them at a time still ends up making me more productive.

I wouldn't say all of this has made me 10x more productive like the pundits say.

But this experience made me change my opinion from: “with LLMs you can write code faster but you lose a lot more time guiding, reviewing and debugging” to “I can genuinely feel like I produce more output”.

Oh, that sweet nectar

There is one more thing I want to say, and I'm sort of circling back with this section but bear with me.

I joined crypto at the peak of the hype in December 2021. We were the new hot chick on the block. VCs had a raging hard-on for us. I remember at ETH Amsterdam, you could feel it in the air. It was intoxicating. Everybody was hyped beyond belief. Us, in particular, at Toucan felt like rockstars. A lot of people really wanted to meet us.

We, all of us, were going to change the world. Decentralized finance baby! Everything was going to be different and better because we were going to make it so. And we were going to become multi-millionaires doing it.

As an impressionable twenty-something-year-old, I half fell for it. How could you not when older more experienced people around me were drinking the kool-aid?

As you can probably tell, that did not happen. Not even close.

I still feel nostalgic though. And, in hindsight, what I've done here, with this pivot from crypto to AI, is very disloyal to the crypto industry which is the first place where I got a real (much needed) break.

This industry took me from the broke kid that I was, who was really just happy to work from home with AC, to a real professional who has real achievements, and I made good money. And I'm now kicking it while it's down.

Of course, I say all this more so metaphorically, because the industry is not a person, so there's nobody getting hurt here by me doing this.

But this sentiment has made me think how really a lot of the tech industry, not because tech people are evil people, but simply due to incentives we chase that sweet VC tit, for better or worse. The VCs chase their own incentives as well.

This creates cycle after cycle, wave after wave, a never-ending rollercoaster of industries and niches. Most of them come crashing down. Some of them win big and change the world, but make no mistake. This, all of it, is part of the game. This is what we signed up for.

I want to leave you with a poem I wrote:

Oh, that sweet nectar, that sweet golden flow.

Why toil in the dirt when the garden’s aglow?

The blooms will wilt… Tis they always do,

But others await… The bee always knew.

Keep reading