First Principles Thinking
sunilsadasivan.com
sunilsadasivan.com
An aggressive first principles approach often leads otherwise well-intentioned technologists into strategic / ideological dead-ends.
Do we do things because it's the "right thing" to do in the moment, or because of the final outcome that will eventually result?
The most ideal answer is somewhere in the middle. I am far more interested in the total area under the curve than a single instant in time.
In lieu of intentional higher order thinking, simply working backward from your customer on a regular basis will generally accomplish the same outcomes.
The problem is... which first principles? Say your goal is to get to the moon, and the principle you use is to minimize the distance, and you choose to do this in a greedy fashion. This will have you climbing ladders, trees, buildings, and mountains, and you will record that as progress justifying the approach. But you haven't reasoned far back enough to figure out just greedily minimizing your distance to the moon is never going to get you there, no matter what. Choosing the wrong principles can lead to local extremes which can be a lot harder to see in cases less obvious than going to the moon.
Reasoning backwards has its own traps as well.
First principles are not an ideology; they are more like laws of nature. Due to that, they are great both for reformulating a problem, and checking solutions.
That's the ideology.
Based on the moon example, it seems you define it as a set of principles you can base your decisions on? E.g. to get somewhere, you must minimize your distance. What makes something a first principle?
I'm asking because my understanding is that "reasoning from first principles" is just taking a step back, asking yourself what's really the goal and reasoning backwards. But it seems like you've got two different concepts (first principles and higher order thinking), which makes me think you've got it figured out more than I do.
Is it?
> In lieu of intentional higher order thinking, simply working backward from your customer on a regular basis will generally accomplish the same outcomes.
And here I thought we were discussing something important!
I was just talking with a friend the other day about how many problems and issues would be solved if people were more conditioned to asking, "...and then what?"
Then also, companies are guided by some mission / values which might prevent you from going into that direction, unless you are working for an online gambling company
Higher order: What if we accomplished X?
First principles: "We need X, how can we accomplish that?"
Higher order: "We need X, what would accomplishing that do for us?"
And who should first principles focused on? Your customer, your manager, your company, yourself? All of these are valid and often give competing answers!
"First order" thinking from my knowledge is an entirely different concept.
First principles are an illusion. Good engineers make strategic choices of the supposed "first principles" they apply, and on a particularly difficult problem, they may iterate through multiple options before discovering the "first principles" that yield a good solution.
Engineers that take first principles seriously and try to use them to guide their decisions discover that correct-sounding first principles tend to work in some contexts and fail in others. Then they go one of two ways responding to these failures.
One way is to blame the context. My favorite example of this is a project where a consultant writing a Java web service struggled for months to get Hibernate to emit queries than ran efficiently on the company's Oracle database. Pressed by management to write the queries by hand, the consultant insisted that using Hibernate was a non-negotiable "best practice." Eventually the manager grepped the logs, identified twelve distinct queries that the service needed, and tasked the Oracle admin with optimizing them, which took less than an afternoon. The consultant still pushed back against using the hand-tweaked queries, saying that if the answer emitted by Hibernate didn't run efficiently, then the database schema was wrong and needed to be fixed. This was not the right response when the database in question was the beating heart of a hundred million dollar tech company and was optimized to handle massive transactional workloads, and the consultant was struggling to write a simple web service for internal users.
The second way is to push your first principles to higher and higher levels of abstraction to make them less context-dependent. They update "Use Hibernate for database access" to "Use whatever database library or framework makes it possible to achieve the best results with the least developer effort" and eventually resort to something like "do whatever achieves the best outcome measured by its impact on what is important to you at the time."
I think a better approach is to collect a toolkit of potentially helpful principles and develop understanding of when each principle tends to be helpful.
Higher order thinking is actually substantially more common, like overwhelming so. Higher order thinking is the application of known patterns, such as frameworks and common tools. Many people often dismiss doing the right thing for matters of convenience, which is the basis of tech debt.
First principles are not elusive. State your goals and limitations first. Then proceed accordingly.
Architecture has remained so far to be one of those domains where skillset dwarfs everything by comparison.
The agent can rewrite half your codebase in one day, so architecture stops mattering for the most part.
For example is unknowingly writing a security flaw ethical, when you could have used a set of processes to reduce them before release that would have make the entire thing take longer and cost more. Seems like programmers need a lot more ethics classes as ethics are part of any large scale process.
Besides, ethics come from upbringing and social influences, not from attending a mandatory ethics class.
This is not true for everybody or for every project. Sometimes it’s about the code and not about what it does. Sometimes it’s about learning. Sometimes it’s about the fun of creating. “Only the end product matters” is a narrow view of the world of software development.
But why would you care when an AI can just rewrite it? Yes, but can it rewrite it to a good architecture? Or just to a different one?
Does a good architecture make code easier for an AI to maintain? I don't know, but I think it's at least not proven that it doesn't.
Some apps don't care about scalability and performance but many do. Ignoring architecture all but guarantees inefficient, wasteful software.
I'll create a simple "framework" of what I know works. After that's there the LLM is fantastic.
LLMs benefit from abstractions for the same reasons that humans do. More information in the same amount of text. Fewer working parts to juggle so fewer ways to make mistakes.
Similar to the output of a compiler. Nobody (with very few exceptions) reviews its machine code output. No reason to do that. If you want to change it, just recompile.
(I don't share this view, but I think a substantial and growing fraction of folks does.)
But if the codebase is large enough, and if you don't have tests for everything, you'll likely lose stuff along the way. So you'll then spend at least another day tidying up the rewrite, just getting it back to where you were two days ago.
Does that sound like fun? Wouldn't it be easier to just plan it correctly the first time?
I'd be sad working with someone who approached their work like this.
I feel the opposite, like if I go to an agent without first knowing what I want to build, I'll never figure out what I'm doing or why and it'll run away from me.
I pretty much always go back and forth and have the agent write out a plan to a file and review it myself in my text editor. I still sometimes end up with surprises I disagree with, but I don't really find it to be true that the LLM ends up trying to "drive all of the thinking".
When I'm thinking through an architecture, I not only instruct it to refrain from writing any code, I don't even necessarily tell the agent what I'm trying to build.
I start talking to it while doing menial tasks like cleaning or doing the laundry, and I discuss the architectual decisions and options until I come to some resemblance of a plan.
Good side of this approach is that I can't just "skim over" or "copy paste" things - either I understood them and can repeat them myself, or I can't. It takes more time than /grill-me and similar approaches, but it's the only approach that doesn't make me want to claw my brain out.
It's very enlightening, though it requires being comfortable with being alone with ones thoughts. It seems to me many people are not and need constant distractions and/or dopamine kicks.
AI just serves as a substitute - an artificial conversational partner.
I feel like you consider yourself more virtuous for not using AI for discussion. Would you consider people who don't talk to colleagues at all, but do all the work themselves more virtuous in an analogous manner?
I walk to work and home with ChatGPT Voice and AirPods. I ask it to be Socratic and I just start rambling the top of thing on my mind. After 20 mins of back-and-forth it's usually teased an answer out of me or I've teased an answer out of it.
And do you feel comfortable with the setting that a corporation has a detailed log of your deepest thoughts?
Other times, I let the agent create a black box with a well-defined interface contract. I don't care about the architecture inside the black box.
"Couldn't this be stateless?" "Do you have a plan to be able to shard this?" - We will almost always pivot from the agents initial design but the agent is able to easily understand the reasons and benefits and align quickly.
If I was writing the code myself, I'd often have to make compromises between the ideal architecture and the level of effort required to implement it - now I can just always have the ideal architecture.
Now you have what you think is the ideal architecture. Since you didn't implement it, you didn't discover it was not the ideal one mid way into the implementation.
May be it is too complicated, but you wouldn't know, because LLM is doing the implementation. If you did implement it yourself, you might have spotted a critical point that might simplify the whole thing...
This is the "Jodie Foster in Contact listening for the SETI signal with headphones" theory of how production software systems work.
But then, one small detail changed everything. And it wasn't something I'd have thought of until I was writing the code.
I shudder thinking about the number of these issues hiding in LLM-generated code.
There is a difference between LLM-generated code that somebody merged essentially unseen and LLM-generated code which a human then looked at carefully and massaged until they were happy with it.
The first version can hide a lot of crap. All you see is the end-to-end functionality and there can be corner cases or scenarios which you have not tried and in which it is terrible or plain incorrect.
The second version hides just as many issues as human written code.
Unfortunately, the debate that our field has about LLM-generated code mixes both quite freely even though they are quite distinct.
Another perspective is that this isn’t a new problem. Every dev manager or team lead with very strong direct reports has this problem. They don’t lose the ability to reason, they strengthen their ability to reason more quickly, and know when to trust the expert and when to poke holes.
This is how you end up with unnecessary complexity [1]. The best engineers don't aim for "designing something ambitious", instead they come up with the simplest possible design. They take something that seems complex and make it simple.
Unfortunately that's not how engineers are evaluated [2].
[2] https://terriblesoftware.org/2026/03/03/nobody-gets-promoted...
These are orthogonal.
A lot of the dissonance on HN appears to come from two groups of people talking past each other:
a) developers working at some corporation they hate vs.
b) developers working for themselves or somewhere they don't hate
As an example, heliocentrism was profoundly ambitious yet devastatingly simple compared to geocentric formulae.
But because of time, we all have to consider the value of our own time. AI forces you to look in the mirror, and reflect.
Don't be afraid to change: use first principles thinking and broaden your perspective about yourself. Your station, your position in your own mind has changed due to AI. It's easier to think AI will just go away: and our egos and personas can remain unchanged. We'll be back to where we were, and everything we knew about value will be right where it was.
It usually doesn't work that way, and you usually don't do yourselves any favors in being so rigid and stuck in the old way of thinking about yourself. Keep up with the times, but balance your health and happiness.
While I think there are better introductions to "first principles thinking" than this post, the idea of "...When I step back and ask what we’re actually trying to do, why it matters, and how the pieces connect, I usually find more ways forward than I expected."
That's an important skill to have.
However, I think sometimes we overvalue thinking from first principles when it isn't warranted. I've seen times in my career where first principles thinking led to a solution that ignored key non-technical constraints. (For example, it would require a full re-architecture of the system and require deferring all feature work for a year. Another example: the proposed solution breaks Conways law in a way that would require a reorg that would break other organizational constraints.)
Sometimes, we need to recognize our constraints. Spend the time to question them when appropriate, but realize that there are other tools that are more appropriate in some cases -- such as anthropological thinking.
The problem isn't the thinking though, but the lack of grounding.
You don't have a problem with people coming up with a technically better solution. Your problem is with the people not realizing that they exist in reality and not in a vacuum.
Please don't question the act of thinking itself. It is the wrong target.
Fundamentally software we create exists to solve problems, if bad code solves the problem is it bad code? Counter to that is we are engineers and it's our job to design systems that mean problems can be solved safely and effectively.
Like most things in life the truth lies in the middle
When you say something isn't warranted because it takes too much time, it isn't valuable: you're still thinking about what it is you're doing (working), what tools you have (your team, people), to create value (your product or service).
It's the measurement of value part that's tricky. Who determines value? Sometimes it's the customer, sometimes it's just you.
It's about perspective, and seeing where you fit in the picture. Apply the principles from there. Nobody says you have to be a jerk about it, that's up to you and how you wish to apply the output of the first principled thinking.
You are working, and you are building things, to create value.
What are the tools you use to move about your environment and create your product, your value? Your knowledge, your training, your time.
If you examine the first principles of whatever it is you're doing, you should realize: you must keep your toolbox open and ready to swap new tools in and out in order to keep up with the time and manage yourself effectively.
AI is a tool, learn to use it as a tool and stop 'fighting the times' so much. All of your criticisms are valid, and you are correct: but everyone still wants to save time, and so you must keep up.
I think they're describing an approach to decision-making in the context of agents.
Engineers encountering LLMs for the first time think the problems they encounter are entirely novel, but really we are rehashing old lessons which Fred Brooks wrote about half a century ago, but in new jargon.
The essential difficulty of specifying what you actually want remains the same and nothing I've seen from LLMs will change that.
I find that I need to invest a whole lot in high level design myself to get Codex to create a suitable architecture and make the right tradeoffs. It's more like I am designing and Codex is reviewing and occasionally we brainstorm. When I tried having it design based on requirements, it went wild with an unsustainable design / architecture.
And I use things like plan mode etc. My experience is unlike what I read in most vibe coding exploits.
I wonder if I am doing something wrong. Is there a good canonical example of a project built with Codex or even Claude Code that shows how the human and AI interact that I could use as a reference
Then generalize + standardize that process, get an agent to replace you as high level designer so you can manage a team of high level designers, etc etc
Seriously, how do you brainstorm with an entity that would 180 if pushed a bit..How can you take anything it say at face value?
You put the customer first, that's first principles thinking. You adjust the way you spend your time, according to your customer's needs.
This is so true. Every senior engineer I've worked with knew what needs to be done regardless how big a particular initiative/project was. Most of the time they even had a hunch or a vague idea on how to do it as well, but they were so good at breaking things into smaller parts and get the important things done first.
Recently with AI I've realized how easy it has become to figure out how to do something as well. I can just play around and explore a solution and ask the LLM to code it in the code base and then I can actually go through the solution on high level and see how it would work/not work without doing brain gymnastics trying to figure it out all in my head because I'm obviously too lazy to write any code just to experiment.
Finding key projects that impact your organization; being the first to dive into toughest operational challenges and finding non-linear solutions are also great ways to grow in your career.
I’m not sure why would you want to think about how an AI would help. If you understand what you need to do and you understand the capabilities of the tools available you run the coding / learning loop. If you think you should use AI then you are solving the how to use AI problem, which just adds to the complexity. Orthogonal to first principle thinking, just a workflow issue.
Now about this article, idk, maybe i'm just jaded on what engineering has become. My gut reaction is agentic engineering feels wrong.
Getting that on-call incident for a piece of code I own, when i've lost the mental map of what the black box does, it keeps me up at night.
Is the solution really going to be, "use ai for that!". Ugh.
To actually do it is different and usually comes from having to wrestle with a problem.
Sadly people from the academia and the public at large has a hard time understanding what this even means.
They equate it with exam based memorization or delegation to authority. Funnily they even think first principles reasoning is an improved version of doing the same.
But this is a blessing in disguise as it gives those who wrestle with real problems a unique skillset that can be advantageous.
In what departments and at what universities? The term "first principles" comes from academia. I think you'd be hard pressed to find a faculty member in any philosophy department doesn't understand what reasoning from first principles is. I'd be surprised if any working mathematician thinks of "memorization" or delegating to authority rather than axiomatization. What experiences led you to say this?
If this comes as a shocker to you, then I should be the one to enquire as to how you managed to stay blind in the face of the obvious.
Just strike up a discussion on some complex topic, and you can see many people resort to "because the author here in this book said" or "we are taught so and so".
A first principles reasoning can show you the steps that lead to a specific conclusion without invoking any author, teacher or course.
Cheers. Hope it helps.
What's missing in your claim is evidence that people in academia often mischaracterize this style of discussion from accumulated knowledge and shared context as "thinking from first principles". I don't see any plausible rationale for why they would.
Otherwise, using their typical mode of interaction from outside observations to infer they misunderstand first principles is not a standard anyone doing specialized work inside or outside academia would ever be able to meet.
The private sector would grind to a halt if discussing a specific IEEE 802.11 protocol implementation with a fellow SME required a lengthy preamble of networking first principles before answering in order to be epistemologically sound.
Similarly, using a conversation overheard at a conference to infer the experts lack first principles thinking would not be reasonable just because they appealed to IEEE documents instead of rearticulating the underlying decisions made by the standards committee.
You can always breakdown one layer on demand. And you only need to drop down as far as the other party insists.
My example was about striking up direct conversation with someone and not overhearing.
If someone pointed to IEEE protocol and I asked "but X, so why?" and if the answer is still IEEE protocol again, then the person does not have "first principles".
That is the test. It exposes the shorthand user from the legit first-principles capable user.
And as you said, pointing to literature just "addresses" a topic. Obtaining it is a different matter altogether.
I think you're right that lots of people with tons of training, expertise, or experience are (sometimes very disappointingly) disinclined to engage in first-principles thinking. But I would expect most academicians to at least know what it is.
Complex topics are built on history, on those who came before us, on attempts and failures and successes. Ignoring it is garden variety stupid.
Your first principles reasoning must then be deference to authority and course. Good to know.
> Complex topics are built on history, on those who came before us, on attempts and failures and successes
Tell that to Alan Turing, Godel, Tarski, Aristotle and a lot of other first principles thinkers.
> Ignoring it is garden variety stupid.
Bro, What about ignoring the original comment.
Also, in my limited personal experience so far, the people who relentlessly talk about and evangelize first principles thinking (mostly highly self-confident tech bros and big-tech veterans) are often the ones who fail most absurdly at actually doing it. It reminds me of the exchange attributed to Robert Surcouf: when an English officer supposedly told him, “You French fight for money, while we English fight for honour,” Surcouf replied, “Each man fights for what he lacks.”
P.S I don’t have any stance on the French-English conflict in the anecdote above.