Joe Muoio

Clarity is good. Changing how your tool works in a way that is not technically sound and breaks the work folks who love building on your tech and pushing subscriptions is exactly the reason why I left Anthropic behind months back. Big miss from Anthropic in a year of misses.

@nbaschez Same problem we always had as devs with analysis paralysis. The point of the plan and doc is to think through the problem. So you are done when you have done that and have a sufficiently detailed written plan you approve of.

My biggest challenge with vibe coding / agentic engineering lately has been getting stuck in what I call a "plan doom loop" - have AI write a plan - review myself, seems good - have AI review plan, it always finds something - repeat It drains my time and energy to determine how important the "findings" really are Who has solved this

@trashh_dev If you ain't using 1 trillion dollars of tokens per month are you even a senior dev?!

“we’re going to need to tailor back your ai usage”

@kareem_carr Super well said! This is also why it works great for Programming, too, of course. Even tighter validation loop. And exactly why all the weirdos thinking you need to track your prompts for software dev are totally off base. Code >>> fuzzy human language.

The rapid advances we see in AI-derived mathematical proofs are almost certainly not representative of science in general. A core driver of these advances is that AI-derived proofs can be translated into a highly structured human-designed verification language, which can then be checked using traditional computer programs. The AI slop-cannon can generate as many slop attempts as needed to get a proof that works, because humans already built the de-slopification engine that automates digging the diamonds out of the slop. This kind of cheap validation does not exist in data science or the empirical sciences more broadly. In fact, validation in the sciences is often orders of magnitude more expensive than all the other parts, which is why AI is going to be much less effective there.

@johncrickett I mean I think you nailed it. Curiosity and skepticism. What actually works and why you choose certain tools and techniques. How you integrate with a team.

@JoeMoCode Interesting, what does good look like to the interviewer? Like you I'd be looking for an open mind. I'd also like to see a good mixture of curiosity and skepticism. The hype is excessive, and they're still useful.

What does "AI skills" actually mean? I've seen job ads that list it as a requirement, but I'm not sure anyone's asking the same question twice. So I want to hear from the people actually in the market right now. If you're hiring engineers, are you filtering for AI skills, and if so, what does that actually look like in practice? If you're a recruiter, are clients asking you to screen for it? If you're job hunting, has it come up in interviews? I‘d love to hear your experiences. I'm more interested in what's really happening than what the headlines say.

@johncrickett *.vercel.app

Everyone talks about how good AI agents are at writing code. But where's the actual software? Share your best example below.

@kareem_carr @AntiFascistHT LOL yeah. I hate the word hallucination since it implies something is wrong with what is going on with the LLM in those situations, but it is in fact doing what it always does. Whether you think its true or not, gen AI performs same operations.

@AntiFascistHT Honestly, this is obvious from first principles and did not require a paper.

There's a toxic culture coming out of the AI industry that keeps trying to get us not to think. The message is everywhere. Don’t read the code, just vibe-code. Don’t try to understand all the text, just let AI summarize it. Don’t bother educating yourself, it’s too late. Don’t worry about the errors. Trust that everything will be fixed in the next version. The theme is the same. Don’t think too hard. Just keep swallowing the slop.

@DanielLockyer great domain name, btw. Good article. This is a big reason why big companies fail to deliver relative to their effort/resources.

From: terriblesoftware.org/2026/03/03/nob…

It's sad that we promote and encourage complexity in the tech industry I'd love to see more of a push towards simple solutions to problems

@DanielLockyer Simplicity gives you less tech debt. Own your own tech, and you own the advantages to building simply. Stop trying to impress others and instead do what is right. You'll find the right collaborators along the way.

@infinitehumanai People in tech sales or Solution Architects had to deal with this forever. They need to know and reason about the services without having had the luxury of learning through building I encountered this while doing devrel, too, but had hands on experiences prior to mitigate

@infinitehumanai yes! Good way to frame this. The more I apply AI-assisted development, the more I see the wisdom from the early software developers many decades ago. I've often thought of the code as a way to enforce constructs designed across the team which this corroborates.

Total software debt = technical debt + cognitive debt Previously, we mostly had technical debt, now with AI we have both. Peter Naur's classic 1985 essay "Programming as Theory Building" argues that a program is not its source code. A program is a shared mental construct (he uses the word theory) that lives in the minds of the people who work on it. If you lose the people, you lose the program. The code is merely a written representation of the program, and it's lossy, so you can't reconstruct a program from its code. Previously, when you built something, you accumulated technical debt but relatively little cognitive debt because we had to understand what you were building in order to build it. In other words: the theory came for free as a byproduct of the work. AI breaks that coupling. Now you can produce code without building the theory. So you're now able to accumulate both kinds of debt simultaneously - technical debt in the code and cognitive debt in yourself. And cognitive debt is arguably worse because you can fool yourself into believing it doesn't exist. Technical debt tends to show up in semi-obvious ways that we understand well as an industry. Cognitive debt is more insidious - it means you're unable to even reason about the program (because you possess no theory of it) - which is what Naur describes as the "death" of a program.

@thdxr @AIEMiami @opencode ai

@AIEMiami @opencode what should i talk about

Excited to share that Not the CEO of @OpenCode is speaking at AI Engineer Miami! Don't miss @thdxr this April! Lock in your ticket 👇

@ninja_maths @_MathAcademy_ Sick! Nice work. I love seeing these kind of systems get built and help students.

@JoeMoCode It's the knowledge graph we use @_MathAcademy_ to guide each student's learning trajectory.

For anyone wondering how a third-grader can complete six years' worth of math in a single year. This knowledge graph spans 3,000 math topics, from 4th grade to the university level, providing the perfect basis for mastery learning. Students can go as fast or far as they want! There are no restrictions whatsoever. The only requirement is that they must demonstrate mastery of each topic before moving on to the next. Kids are capable of incredible things when given that kind of freedom and support.

@justinskycak Very cool stuff. I left a question on the last tweet, but this seems to answer that it is your own knowledge graph/model/curriculum.

How to compress a grade level’s worth of learning much, much shorter than a year: 1. Identify what the student already knows 2. Overlay that on a knowledge graph to construct their personal knowledge profile 3. Teach only new topics for which they've mastered the prerequisites, their "knowledge frontier" 4. Each lesson cycles through minimum effective doses of explicitly guided instruction & active practice problems 5. Enforce mastery relentlessly: if you can't consistently solve problems correctly, then you don't move on to more advanced material that depends on it. You continue on parallel learning paths and come back to the halted one later. 6. Review previously learned material using spaced repetition & frequent broad-coverage closed-book timed quizzes 7. Review old stuff by learning new stuff. I.e., knock out as much review as possible by learning new material that exercises those review topics as subskills.

@aleattorium Go ahead and read the tweet again. It says little changed about software development and a lot changed about coding. True. Not sure why you keep fighting against "nothing changed".

@JoeMoCode I do, I actually do. Do you understand my point? It's not that code goes brrr. But that our practices around development is not prepared to code generated faster. The original tweet said nothing changed. But generating code faster we need to adapt the "rest of the stack".

The entire software engineering field evolved under a specific constraint: humans writing code at human speed. Team topologies, review rituals, abstractions, architectural boundaries were optimized for that production model.

@plainionist @Grady_Booch yep! Also to own, maintain, and fix all the AI code made by non-software devs and more junior folk that becomes important to people/businesses. Not to mention security engineers...

@Grady_Booch I believe even more software engineers will be needed in the future to build creative software systems that are unthinkable today. As LLMs are neither truly intelligent nor creative - at least today - this will remain a job for us 😉

History demonstrates that the advancement of technology destroys some jobs while simultaneously creating others. This raises two fundamental issues: the rate of change (and the societal ability to metabolize that change) and whether or not the result is net positive, neutral, or negative (with regard to human value and human costs). "AI Slop Cleaner" is a job category that I did not see coming.

@aleattorium you understand math is different than software development, right? You should read the mythical man month. Nothing fundamental has changed. Yes we can do code faster. Code is 30% of software development, at best.

We’ve seen this in math. The axioms didn’t change when computers arrived. Throughput did. Suddenly we could explore conjectures faster, test theorems at scale, and rely on computational verification. Speed alone reshaped the field.

Been seeing this around and it's really a good visual to help break me out of my echo chamber. It definitely raises more questions than gives answers, but I love the visual representation of AI adoption.