I’m a software engineer and technical leader with a multi-faceted career spanning management, leadership, education, sales, recruiting, finance, but most of all software engineering. I'm passionate about helping engineers maximize their professional impact and earnings.

Alright hear me out maybe these ram and gpu prices are worth it after all

Victor Moreno

Senior Grug @AWS | 🗣 web systems & career advice

I fear dev jobs of 2027 and beyond are going to have way less WLB than ever

You know how directors are sort of expected to have their phones on them 24/7 and answer a slack even at 10pm?

I think with ai coding, devs are going to be expected to burn the candle at both ends. 

You get to spend a lot of time "waiting" throughout the day for your ralph loop to get stuck so you can nudge it along

I think the expectation is gonna be that you do this for 14, 16 hours a day.

Weave your personal life in between.

Like a day trader that trades US and european markets

Or like an executive of a worldwide company who answers emails at all hours

We used to be limited by the amount of exhaustion that deep work generates. But we won't need deep work to produce software any more.

I have an idea for an interview question in the AI-first world. Tell me what you think.

The idea is to come up with a small app. Something obscure. Not a url shortener but something different. Maybe an email marketing SaaS or an AWS systems manager clone.

You’d flesh out all the product requirements (spend enough time to do a good job at this). Then ask ai to do a system design based on the product requirements.

It’ll inevitably be very bad as AI always does a bad job at everything on the first shot. 

So, the question would be like so:

Show the candidate the requirements and the ai-generated design. Ask the candidate to critique it and defend their opinions.

Wdyt?

Two tips on actually achieving increased productivity with AI:

1. Practice the skill of knowing AI is ngmi. If you gain back an hour from doing a bunch of little things with AI, it's real easy to lose that hour back by continually insisting that ai do something it can't really do. Developing that intuition for which tasks require a detailed breakdown and which ones can be one-shot is super important.

2. Set it to work in a ralph loop at the end of the day. Spend some of your productive hours accumulating a backlog of ralph loops and always have at least 2-3 loops in reserve. When you sign off every day, set the AI to work on the loop. It's like getting an extra couple of hours of work in every day.

A broader way to look at #2 is this: don't measure productivity in terms of output per hour you're at the computer. Instead, measure productivity in terms of output per wall clock day. Ai can extend your work hours (and thus your output) without additional effort, if you plan properly.

There's a trope about "the same year of experience repeated ten times."

I think there's a lot of truth to the trope. A lot of people fall into that trap. I'd go as far as to say the majority.

There are two broad approaches to avoid this trap, and one psychological component.

The psychological component is just intensity. Demand a lot of yourself. Be always hungry to improve. Be hungry to read and learn and try and fail. Make sure to plead your case and take some opportunities for yourself.

The behavioral approaches are (not in any order of preference):

1. Switch companies to a bigger and better one with more scale (this is about getting more depth)
2. Switch lanes. For example backend to frontend, to embedded, product to platform, to consulting, etc (this is about getting more breadth)

I think you should do both of those approaches, at different points in time. 

In the beginning of my career I did a lot of the second approach. I did mobile, frontend, backend, full stack, worked in consultancy and worked in long-term company. You learn different things from all of those.

In the latter part of my career I've stuck with tech companies but I've been going from smaller, lower-stake systems to high traffic systems with a maximal demand for correctness.

How soon should you switch? Depends on the emotional side. 

The less intense you are, in general the faster you need to switch. For some people, after joining one place and being there for 18 months, they get 90% of everything they were gonna get out of it. This can happen for many reasons which I'm lumping under the bad name of "lack of intensity." But it can just be shyness and not knowing how to assert yourself to be given more scope. Or it can be having made a mistake early on and not having recovered from having created a negative first impression. Or it can be actual lack of intensity/drive.

Either way, when you reach the point where your learning significantly slows down, you should move. If you already have decent breath, then move to a company that does similar work but has higher quality engineering. If you already have decent depth, change lanes entirely and get some breadth.

The more intensity you have, the longer you can last without switching. I'm at 3.5 years in my current team and while it's no longer a firehose it's quite a lot of daily learning still.

Why all this? Because ultimately, the only thing that comes close to guaranteeing career longevity is being good. And the only ways to be good are either having a tremendous amount of breadth, or having decent breadth with a couple of areas of real depth. 

And when it comes to building real depth, nothing replaces using the thing for years and years and testing it at its limits of performance, scale, or whatever other vector is important for the particular technology.

Tutorials and docs are the surface. 

Building gives you familiarity.

But using it for years and taking it to its limits is what gives you mastery.

Controversial economics take you probably haven't heard before: the terminal state of free market competition in a winner-take-all market is a market-destroying monopoly.

(Socialism is still much worse btw, but that's a topic for another day)

The vast majority of economic philosophy around free markets was developed in a time where there were few if any winner-take-all markets with network effects.

In those times, when the most relevant economic units were farms, restaurants, printing companies, and even car manufacturing plants, there were no winner-take-all markets in a global scale. You couldn't ship potatoes from idaho to siberia due to spoilage, and even shipping a car from detroit to nepal was too expensive and would eat into margins too much. 

And so, there was a forcing function there to keep competition alive. If shipping your Ford to nepal means having to mark it up by 50%, that opens up room to compete; a local company can make the same car and get it there chaper.

With digital goods, this is not the case. And as we have seen with mag7 valuations, digital goods are ever-increasing in their dominance of economic activity.

What happens when the only cost to distribute your social media app in vietnam is the cost of translating some strings? 

It creates a winner-take-all dynamic in the market for social media.

Same goes for ERP's, database management systems, you name it. Most software.

Once an institution has ossified its chokehold on a market, they have an easy time squashing competition through M&A or even through darker tactics like sabotage and disinformation.

The end result is a monopoly where a giant, bureaucratic company has too much financial and political power to ever be out-competed.

When you have full economic freedom, that guarantees exactly one thing: that the strongest competitor will win.

"Strongest" can mean smartest or most innovative, but once an incumbent reaches a certain level of political and economic power, wits and innovation are not enough to defeat it.

A de-facto, unbreakable monopoly is created.

And for those of you who don't know, once a company achieves monopolistic status, they gain price-setting ability. They are no longer subject to market forces and so they set prices at a level to maximize economic profit rather than be subject to supply and demand forces.

This is not necessarily a problem in and of itself. It's only a problem when paired with another attribute of human nature: greed.

If monopolies were always fully benevolent, they would invest their surplus economic profits into R&D to stay on top forever and widen the gap between themselves and budding new startups.

But history shows this only happens for so long. After a while, the temptation to invest some of that economic surplus in lobbying and tampering with democratic processes becomes too great. 

And so they gain political power. 

I don't think this problem has a solution. Make sure you own land.

In case you bump into a paper claiming to mathematically prove the limits of AI

Formal Methods | Software Engineering | Software History

A paper proving the "mathematical limits" of AI has been sweeping all my social media feeds, with raving coverage by Wired, Gizmodo, and Futurism.

So I went and read the actual paper (link in comments). Unsurprisingly, it's atrociously bad.

The core argument is that the time an LLM takes to produce output scales quadratically with the input string, so it's mathematically impossible to do any task that scales faster than O(n^2). For example, it wouldn't be able to complete a task like "count to a billion."

This completely misses how LLMs work in practice, which is to produce an output, and then REPEAT generation on the output + input. So even if the task isn't O(n^2), LLMs output part of the solution in one iteration, then a bit more in a second iteration, until eventually the whole thing is done.

(An analogy: "CPUs can't add big numbers because a CPU does O(1) work in a clock cycle and addition is O(|digits|))

The paper also states that if a task is really hard to solve, it must be really hard to check a solution. This is a basic misunderstanding of computational complexity. NP-Complete problems, for example, (probably) have exponential solving times but (definitely) have polynomial checking times. This immediately ruins the latter half of the paper.

The biggest problem with the paper, though?

The lead author was a high school student with no background in computer science.

I don't blame the author for writing it (We were all like that in high school), but I do blame the journalists for not disclosing that the author of said paper was not remotely qualified to write it. Readers aren't gonna Kagi the authors on their own!

And, as always: don't believe articles just because you like what they say. That goes for both pro-AI and anti-AI people.

Link to "paper" in comments.

Has anyone worked on a login system where, if it's the first time you see a particular email, you just trust that the customer owns it?

You could ask the user to put in their email, and if you've never seen that email before, don't force them to verify it or pick a password. 

If you have seen that email before, you obviously do need to force the user to authenticate in order to protect the data that is associated with that email.

But if you haven't seen the user before you can just create a session id and associate it with the current device wit an httpOnly cookie. 

Let the user proceed and make a purchase without the friction of verifying email. 

Then, in the end, tell them they need to verify an email in order to keep access to their purchase data or digital product; and the 2fa code can be in the receipt email.

I've never seen this done. There's gotta be a reason why? Do you know?

A basic SEO audit is now 1 claude code prompt away

The implementation is like 3 claude code prompts

Crazy times

I like this take. The penguin is the Wright brothers…

The penguin is George Washington

The penguin is Simon Bolivar 

The penguin is Magellan and Columbus, and everyone who ever took a trans Atlantic trip before we discovered how to measure longitude

The penguin is Grace Hopper

The penguin is Henry Ford

The penguin is Nicola Tesla

The penguin is Nayib Bukele

The penguin is Katherine Johnson

The penguin is a Eddie Hall, Roger Bannister, and everyone who ever shattered limits of the human body that were theoretically impossible

The penguin is Marie Curie

The penguin is Satoshi Nakamoto

LinkedIn look at this amazing new feature you recently shipped

Delete a character when editing the post - lose all your line breaks

Someone forgot there's \n and \r\n line breaks?

taplio

Victor Moreno's Linkedin Analytics

Victor Moreno

Performance ScoreVictor Moreno's LinkedIn Performance

Victor Moreno's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?

taplio

.css-6ytdur{background-image:linear-gradient(284deg, rgba(39, 43, 72, 1) 0%, rgba(33, 47, 183, 1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Victor Moreno's Linkedin Analytics

Victor Moreno

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Victor Moreno's LinkedIn Performance

Victor Moreno's Best Posts .css-1emoo84{font-size:var(--chakra-fontSizes-md);font-weight:400;color:var(--chakra-colors-gray-600);}(last 30 days)

Ready to grow your LinkedIn presence?

Victor Moreno's Linkedin Analytics

Performance ScoreVictor Moreno's LinkedIn Performance

Victor Moreno's Best Posts (last 30 days)