If you’re using AI tools now, congratulations, you’re early. If you use them and don’t give up on them, congratulations, you’re going to have a ridiculous, potentially insurmountable advantage in output moving forward compared to AI “doomers”.

The doomers are a set of people who are determined to NOT use these tools for as long as possible, unless under threat of performance review at their job. And maybe they will be a good control group. Maybe they're right and we're going to do long division by hand and not have calculators to carry around with us for the rest of our lives.

I know AI-assisted heaven in the header image on the web version of this newsletter looks ridiculous, but I thought that it was hilarious that that's what Gemini came up with first.

Stick around for the end (or jump straight to it, I don’t care) for a new recurring section, “Links of the Week”!

If this sounds like a “Part 2” to my previous post, “Second and Third Order Effects of Vibe Coding”. You're not wrong. Every week of traction on this project (over 70 email subscribers!) gives me more data points (678 YouTube subscribers!) that point more and more towards what I'm predicting. I updated this section three different times throughout writing this newsletter.

I don't want to dwell on the doomers for too long, but maybe I'll do some apologetics here at least in this edition of the newsletter, and then we can move on to more interesting things next week.

Also, congratulations on the promotion to resident “AI trainer!” It may not feel like it, but you know more about the capabilities and pitfalls of AI Agents than the average person if you're here reading this.

The early adopters of anything tend to produce the training material for the next cohort. The early adopters are willing to get air dropped in the middle of a dense jungle with no map, no compass, just a vague sense of direction as to where they should go, and are willing to walk around in circles, get lost, found again, just to try to carve out a path for those who will follow them. Early adopters, to quote Ms. Frizzle:

We early adopters are seeking to get an early advantage over those who aren't willing to take the risk. We see the world in terms of effort and timing. The iron is hot right now in this particular field, at this particular time. It just so happens to be our time to strike by spending an above-average amount of time using these tools. This story is played out in every other technology and every other hype cycle. There’s always winners and losers, but I don't think we're going to go wrong in spending time learning the skills of how to best leverage Vibe Coding tools.

I’ve gotten many requests on the livestream to do a nice 10 min intro to Gas Town video where I walk through the happy path flows vs. my long, meandering series of 15 or so three-hour live streams where they may not even see a polecat get spun up. The tutorial video is coming, I just have to script it out for a project of an appropriate size. Maybe a Twitter clone?

What’s even more fundamental that I'm thinking about with training is, “how do we get people to their first ‘aha’ or ‘wow’ moment with AI coding tools as quickly as possible?” And what is the progression from there?

Steve Yegge has laid out the current authoritative eight-step leveling system in his "Welcome to Gastown" article. Level 1 is using AI tab autocomplete, working your way up to level 8 where you’re letting many agents code on your machine unchecked at once.

Yegge has also since added a couple of levels beyond this now. Certain people like Jeffery Emanuel are at a 10 out of 8. Fortunately or unfortunately, humans love hierarchies and ways to compare ourselves to each other. The desire to have a scale to measure oneself against is built into our DNA. It's adaptive, so it persists. I think hierarchies give us a structured way to see what’s possible, and something to strive for.

For example, my favorite belief-breaking fact from just this past week is that Emmanuel currently uses twenty-two, $200 per month Claude Code subscriptions to do his work. $4,400 per month in tokens to make what I'm assuming is a large multiple on that investment. Try getting that in any other investment; you can’t. That is the power of the S&ME 500.

Most people hit the limits on the $20/mo subscription and give up, or are too cheap to upgrade to the $100/month subscription. Now I'm feeling like I'm not doing enough on my $200/month subscription. Burning tokens is learning (Gene Kim and Steve Yegge may have said this first in the Vibe Coding book, I’m not sure. I don’t want to say that I made it up).

Is this how we win over doomers? How does one guy justify spending >$50k annual run rate on his own personal subscriptions to something that doomers think is fundamentally flawed? This isn't an organization spending that much forcing AI on their employees. This is just one guy.

It must be working out for him. Do they get curious about that? I think it really does help to have a real life anchor of what is possible to the absolute best in the world.

Much like knowing what the world record bench press is (782.6 lbs) and then comparing it to your mere mortal personal record (335 lbs when I was 18), you know exactly where you stand in comparison to the best of humanity in one particular attribute.

I have trained in Brazilian Jiu Jitsu for 13 years. Ten of them were in the pursuit of trying to be one of the best in the world while also pursuing a real career.

I got close. In 2022, I made it to the top 16 out of 84 competitors in the -99kg weight class at one of the "Olympic Trials" of Jiu-Jitsu. It’s not actually in the Olympics, but one of the most prestigious tournaments, ADCC, only happens every two years. There are 8 ADCC Trials Tournaments around the world every two years, so that put me in the top ~128 grapplers worldwide at my weight class for that year.

That result fulfilled what I felt like I needed to do. I wanted to prove to myself that I could compete with the best in the world, and I did. Training with that competition in mind for 3 years made me understand what it takes to try to be one of the best in the world at something. Every aspect of your life has to be aligned and pointed towards the goal.

What really lit a fire for me and showed me what was possible was a guy named Jon “Thor” Blank. He’s about my same age, was a full-time electrician, and he placed 4th at the ADCC World Championships in 2019. I was there live and watched him take on guys who trained professionally full-time. And yet, he still did it. Maybe not to the degree that he personally wanted (everyone who competes believes they can win the whole thing), but he still won something for the rest of us: Belief [1].

I'm using the fact of Emmanuel’s 22x $200/mo Claude Code subscriptions as evidence of what's possible, and I just don't know how to do that, yet. It's not something completely inconceivable outside of my abilities, it's just the thing I do currently, multiplied by 22.

That sounds like a lot, but I could probably say that’s the same amount of progression that happened to me between the time I watched Jon Blank in 2019 until I stepped in the ADCC Trials arena in 2022.

Skill progression is not linear. It may appear linear because of linear time, but as you learn more and do more, you can learn more and do more, faster. The better you are, the rate of improvement, improves. The actual level of skill is at minimum, quadratic over time, at least raised to some exponent. In physics terms, the acceleration (second derivative) is positive. The version of my Jiu Jitsu that I competed with in 2022 could probably make the 2019 version of me submit 22 out of 22 times without a doubt.

What does all of this mean to tech nerds who are just trying to keep a job while raising a family and having other obligations? You need to follow at minimum, the following people just to be aware of what's possible and do your best to use the tools and techniques that they recommend to not get left too far behind.

My current set of belief breakers are: Steve Yegge, Geoffrey Huntley (Ralph Wiggum loop guy), and Jeffery Emmanuel.

I'm sure there's more brilliant people out there that I have yet to discover, but I'm also just trying to keep my focus narrow and not be obsessed with collecting “knowledge about” vibe coding, as much as I should be collecting “knowledge of” it through doing the thing itself. My alma mater, Cal Poly’s motto is “Learn by Doing”, and it could not have been more accurate and still informs my perspective of the world nearly 13 years after graduating.

If doomers aren’t convinced by a guy putting his own money where his mouth is, I don’t know what to tell them. Just know that you all who are asking for a Gas Town tutorial, you’re waaaaay ahead of the curve, by at least a quarter, if not two. Since I’ve been actively working in this space, it does feel like an arms race, though:

Andrei Jikh, a finance YouTuber I like, recently put out a video about, “You have 5 years left to get rich”. In it, he describes what some people expect the economy to look like over the next few years given the rise of AI + robots and the impact it is going to have on traditional labor jobs if the cost of manual labor goes to zero.

I think we're going to have a K-shaped economy in software engineering as well. Those who fully embrace this vibe coding revolution will “take off” and will have an upwardly mobile career path, indefinitely. Those who only want to artisanally hand craft every line of code, outside of very specific novel edge cases where AIs are poorly trained (which shrink every version), will be relegated to the same status in the economy as other hand-crafted items: a few real winners, but mostly “arts and crafts” Etsy-level hobbyists doing it for the sake of it.

Steve Yegge calls vibe coding with multi-agent orchestration, "factory farming code." We are now figuring out how to mass-produce code, with quality, at scale. We are now trying to figure out how to deal with less than 100% yields, but with a much larger acreage available and able to be farmed. Doomers are subsistence hunter-gatherers looking at a John Deere combine harvester and complaining that 20-40% of the wheat gets crushed underneath the wheels and is wasted. They are missing the point that one person can now harvest many more times than what was even thought possible just six months ago.

How this translates to the average person in the corporate world won't be felt just yet. There won’t be as big of a spread in this performance review cycle in the next couple months because we just got agents in the last six months and we are still figuring out how to best use them. But a performance divergence is coming. Don’t worry if you’re reading this, you’ll be fine.

If you couldn’t tell, my very conversational writing style is heavily inspired by Taleb. Writing style isn't the only thing I try to emulate from him. I'm taking this opportunity at this technological juncture to add some anti-fragility to my career.

I am creating my own distribution by putting out content (this article, “knowledge about”) and demonstrating what I can build with my technical skills (roxas.ai , “knowledge of”). I recommend everyone do this to the degree that you can in this new era of layoffs, increased competition, and “ghost jobs”.

It's almost a daily headline that the job market is terrible for nearly everyone, but especially for software engineers. Even more especially for entry-level, junior software engineers.

There's stories every day of people applying to hundreds and thousands of jobs. Each submission is a roll of the dice hoping their resume makes it past whatever filters there are to at least be read by a human, and the funnel just decreases from there. There's studies that show from application to offer is a 0.1% to 2% success rate.

I think this is a temporary, transitory situation. Right now, companies are realizing that they can “do more with less” by augmenting their existing engineers with AI for $100s of dollars per month per headcount. Once they fully realize the economics of what is possible with their AI-augmented existing engineers, then they can be more assured in investing in hiring more engineers to “do more with more”. Then it will not only be an arms race for tokens but for people who can effectively wield massive amounts of those tokens.

I see technical interviews changing drastically in the near future as well. It will no longer be about what can you Leetcode yourself, but what can you demonstrate that you've built or can build by yourself. All this can be publicly verifiable via commits on GitHub. I'm doubly publicly verifying this by doing it on YouTube livestream. I truly hope that I've done my last Leetcode problem.

Since I've been putting out vibe coding content over the last two months (just two months!), I've gotten a lot of inbound interest. This is the exact dynamic I hoped that would happen. It changes the relationship when someone reaches out to you vs. you sending hundreds to thousands of copies of your resume into the void, hoping for a response. Hope is not a strategy.

I believe everyone should aspire to reverse their relationship with employment. It should be obvious that you have valuable skills to everyone, and people come out of the woodwork to ask you for help.

It takes effort on our part to demonstrate these skills out in the open. It is hard work that is above and beyond your day-to-day job. It also takes a lot of hard work to search for a job when you've been laid off. Choose your hard.

To be very clear, I'm happy with where I'm working (in case you're my co-worker or manager and you're reading this), but creating optionality and antifragility for myself is a safety parachute that is giving me reassurance in this era of seemingly random, “AI-driven” layoffs.

For those who are camera shy, it takes even more effort to make your letters of text stand out from everyone else's on the internet.

Your face and voice are uniquely your own, and they stand out by themselves. I suggest to try to plaster your face all over the internet, for the exact reason that it is uncomfortable. That discomfort prevents nearly all engineers from doing it because they tend to be introverted, just want to code and not talk about it. You have an advantage if zig where others zag, for now, at least.

Humans still want to work with other humans. Your face and voice are the most human things about you. Use them.

Every day, I see the same guy on LinkedIn post a very heartfelt message about being out of a permanent full-time job for about two years. You may have seen his posts. They get great engagement and LinkedIn probably suggests them to me more often because the algorithm knows that I stop and look at them every time I see them.

I’m guessing he’s towards the end of his career and ageism is rearing its ugly head. This guy figured out how to grow his own distribution, but not in a productive way where anyone is offering him a job out of pity.

I’m sorry that’s not how the world works. I wish we could just take care of people and make sure their needs are met. I would love to vote for somebody who could make that happen, but until then we need to put on a professional facade of “everything’s going great, and I’m the expert and you should listen to me”.

You can prevent ageism in 20-30 years by posting content now and demonstrating your skillset over time. You can create a cohort of peers who follow your work throughout your career and have people who look up to you as a mentor parasocially. Letting people know about what you can do with your skills is at least 50% of our jobs as engineers.

The best time to start posting about your professional and technical skills was 10 years ago when a lot fewer people were on LinkedIn and social media in general and it wasn’t so saturated. The next best time is now.

Nearly all the videos I post on LinkedIn get very little engagement, yet, I still post them anyway. My newsletter posts do a little bit better, and I’ve seen some traction and results from them. I’m only two months into this thing and I found it to be well worthwhile to continue in some form indefinitely.

You can do this thing too.

I started by just banging out short-form tweets and posting them everywhere with Buffer. Here is my referral link but feel free not to use it. I just care that you're getting your ideas out into the world.

A very time-efficient way to post content is to live stream or otherwise just yap for a couple hours while doing something technical and record it. You can then automatically clip it up into short form videos with Opus Clip. Here's my referral link. Again, feel free not to use it

A new recurring segment in the newsletter: links of the week!

On my live stream, people suggest a bunch of interesting new projects and technologies that I can’t get through them all, but I want to share an interesting selected portion with you guys.

Unfortunately on newsletters to not get caught in spam filters, I have to limit the number of links I use in total in the whole article to just six.

So, this first link I share is selfishly that we started a Discord server! As soon as I posted the Discord link in my YouTube live stream, 14 people joined. We’ve had some great discussion so far and we would love for you to join us.

Join the Discord server here

In the server, we have a channel dedicated to interesting links that we share with the group. I'll be posting the rest of my links for the week in there.

[1] Before watching that performance, I was just kind of “meh” about this thing in the back of my head about trying to compete at the highest level. I broke my hand the second match of my senior year of high school in wrestling. Back then, I wanted to qualify for the California state tournament, which was the highest, most badass level possible. But then I couldn't. A competitive desire went dormant, yet lingered in me until I started doing Jiu-Jitsu in 2012. I proceeded to earn my blue belt in a year or so and then be a blue belt for six years, inconsistently training due to various life circumstances.