“Defense tech” should not be a capitalist for-profit venture. The incentives are all wrong. The business model is basically “milk the government for as much money as you can get, over the longest possible time period”, which is never a good business model for incentivizing efficiency or innovation. That is how American defense manufacturing got so slow and costly. It’s why America struggles to produce something as simple as cheap artillery shells at scale: simply not profitable enough. Much more profitable to sell fancy tech at $3m a pop.
Yup and this is the reason I just use Apple devices. Google and Microsoft products are design-by-committee garbage, and most open source projects eventually devolve into a mess, unless it is tightly controlled by a single author (such as SQLite or Redis or even Laravel).
People forget that Nature's default is to devolve to the chaos of the jungle (beautiful, but does not work toward any human purpose). The absolute best human designs are those controlled by benevolent dictatorships with a ruthless focus on simplicity. Nature's default is for every garden to overgrow with weeds, unless a gardener keeps at it relentlessly. Eventually the weeds become a beautiful jungle, but the absolute worst gardens are those designed my multiple competing gardeners. Either have 1 gardener, or 0.
See also the incompetent clusterfuck that is the European Union.
No we don’t, and I’m working on solving this problem. There’s no reason for a pure SaaS company to pay these middlemen if the only thing they buy and sell is software. Software vendors can settle accounts directly like banks do, we just need something like an ACH for SaaS. And I’ve been thinking a lot about how to do this while having the equivalent of a petrodollar to create underlying value based on a commodity. For example a meta currency for compute credits that is cloud agnostic.
I think the Flywheel analogy is great. One of the big differences between working a regular job and running your own product/content business is that it's a flywheel - it works even you don't. Your creative/productive output might be very sporadic and jerky, but the output tends to be smooth and compounded over years.
One of my "revelations" when I founded my own SaaS project a while ago is when I realized there are a bunch of abandoned, yes, completely abandoned, apps on various App Stores and marketplaces that still make a few thousand $ per month, years later, even though they don't even even work. The founder clearly put in a bunch of effort to build it initially, and then didn't carry on for whatever reason, yet it still gets a steady trickling stream of actual customers paying real money.
It's also one of the reasons it is counterproductive to burn out or over-hire or over-raise when building a bootstrapped SaaS. Just don't die, keep pushing the flywheel occasionally when you have some creative inspiration, and eventually you may build something truly great. Or maybe you won't, but at least it will keep you going. But have a flywheel, something that stores your energy and continues output, don't just waste energy trying a bunch of different unreleased projects.
You can guess from the number of reviews and downloads as they change over time. Everyone says it "doesn't work" yet people still download and try it, which means they paid for it. Granted some might chargeback and refund, but not all. Also my personal experience as the owner of such listings is that you get about 1 review per hundred customers (unless you have a very pushy review process). I've made thousands of sales with only a dozen reviews. Also I'll admit to once having an app on a marketplace that was never updated and barely worked still making hundreds of dollars a month almost a decade later.
Given how easy it is to get a refund for an app purchase, I would be really surprised if this results in a significant revenue stream, but I don't actually know, of course.
Warren Buffett has a term for this, it's called "float". At my current (active, not abandoned) startup, my guess is somewhere around 30% of active paying customers are also non-users. Lots of industries have this. The majority of gym membership sales (50%+) and even a significant number of airline tickets are unused.
Ooh, I love these points of Conventional Faith presenting itself as Reason. I would argue that most true Knowledge is actually knowing what is not true, yet commonly believed. Actually, if used wisely, you can execute arbitrages on financial markets for profit, aside from other social benefits, at the cost of appearing insane in the beginning. Peter Thiel says that almost every successful person has "secrets", i.e. things that they know about the world that fly in the face of Conventional Faith and that they therefore exploit to their advantage until it's a fait accompli. Personally, I have quite a few, which if I posted here, would get me downvoted like mad, even on HackerNews, but here are some of my LESS controversial ones:
* Cloud services like AWS actually take more time, resources, and headache to maintain than baremetal servers, whether at the scale of a few backoffice users or millions of consumers. Related: the vast majority (95%+) of apps could run on a single baremetal server and be more reliable than whatever cloud shit you use.
* Most SPAs are garbage. Unless you are building something like Figma, write backend code only, don't use a frontend framework like React, and when reactivity is needed, tie it to the backend with something like Livewire. Actually even Figma uses WASM.
* Using a framework like Laravel that makes the business logic easy is far more important than choice of a language based on its technical characteristics.
* People who are outwardly racist are often the easiest to work with, as a minority. And vice-versa, people who outwardly appear anti-racist are often masking a deeply inbred superiority complex and patronizing attitude (hello Democrats), and are the most difficult to work with unless you submit to their authority. Allowing open racism actually makes it easier for a minority to live, and I say that as a minority.
This problem is as old as time itself and a part of human nature that monotheistic wisdom (or even Buddhism) has always cautioned against. People attributing love, protection, adoration, etc. to fake images construed in their imagination is also known as "idolatry", or "polytheism".
YC was always the dream for me, and I've applied about 25 times over the last 10 years. My work has been all about startups/side projects since I was about 14 years old. I live and breathe startups. But since then I've learned there are downsides to even applying for YC, let alone actually joining.
1. People will steal some of your good ideas. I interviewed with YC back in 2021 where I discussed some unique marketing strategy that had not yet been done in my space. Very suspiciously, I was rejected, yet some large YC companies working in adjacent spaces immediately started implementing that exact specific idea over the next few months. It's possible that was a coincidence, but the timing was very suspect, especially since I was explicitly told during the interview "you have good ideas, but a large team could do this better". And so I'm pretty sure they stole my ideas and gave them to larger teams on the YC portfolio. Beware!
2. It's just a massive distraction if you can actually go it alone. Sam Altman famously said earlier this year, it's only a matter of time before there is a billion dollar company run by a single person with AI. I used to say that back in 2021 even before ChatGPT was released. For many types of SaaS companies which could effectively just be a single person, there is no need for funding, networking, or anything else except writing code and talking to customers. Everything else is just a distraction. YC advice is free on the internet. Vinod Khosla said many investors bring negative value to the companies they invest it.
3. The Silicon Valley groupthink bubble. I laugh at some of the nonsense that comes out of SV, but it doesn't matter to them because there is so much money sloshing around SV that as long as you can attract money by playing their high school popularity contest, you are validated. No matter if there is no substance to your idea. YC companies should count their actual revenue as not including money that comes from other YC companies because it is suspicious if your “product” only matters to people with VC money to splurge. For me, the ultimate test of a good idea is if people are willing to give you their OWN hard-earned money.
4. I, personally, don't do well under intense competition and social pressure. My best ideas and most productive spurts of my life have been when I don't feel pressure to grow, but I work out of a sense of enjoyment in the craft and wanting to solve people's problems that I care about. I have a feeling that YC is one intense pressure cooker, and I'd most likely die there.
I keep hearing this comment everywhere Claude is mentioned, as if there is a coordinated PR boost on social media. My personal experience with Claude 3.5 however is, meh. I don't see much difference compared to GPT-4 and I use AI to help me code every day.
Yeah they really like to mention it everywhere, like yeah it's good but imo not as good as some people make it out to be. I have used it recently for libgdx on kotlin and there are things where it struggles, and the code it sometime gives it's not really "good" kotlin but it takes a good programmer to know what is good and what is not
For me, Copilot and its sisters have been like self-driving cars: a slightly more advanced IDE autocomplete, analogous to a slightly more advanced cruise control. But it's far from Level 5 self driving and it's not obvious whether we will ever reach that.
You still have to keep your hands on the wheel and you need driving expertise. But since writing uncommitted code has less disastrous potential consequences, it is much more usable.
I still don't believe the people who claim they are using AI to write or rewrite entire codebases. Maybe for the first version of toy projects. But I've yet to see anyone using AI to automatically write entire features that span an enterprise software codebase.
I’m the author of aider, an ai pair programming tool. I use aider to develop aider, and it keeps track of how much of its own code that it writes.
Aider wrote 58% of the code in the last release, and >40% of the previous few.
The release history page [0] plots this stat for each release over the last 12+ months. The overall trend is pretty cool, especially since Claude 3.5 Sonnet.
It’s not an enterprise code base, but it’s not a toy code base either.
I think what many people miss, is how long context LLMs get better (needle in a needle stack) and how Context is very important.
With github copilot or with continue for VS Code the main issue lies in how they decide which context to give:
Ideally it's a graph of all function calls and instantiations of classes so whenever your cursor is in a particular spot you could hop through the graph and get all important code pieces as context.
Currently continue uses vector search of chunks which is just a crutch. I am not sure what copilot or Aider does, but the right context is key.
Another way to improve a coding assistant is to move away from simple paradigms to agent based Workflow that can deploy sub agents and break down tasks. All in the back autonomously while you code and then it surfaces suggestions or changes. This will get possible with increasing inference speeds (groq) and better agent frameworks.
Most devs at our company say LLMs are useless for coding and github copilot is a glorified autocomolete costing 20USD. I think the tech and ecosystem will improve a lot over time and they underestimate LLM abilities due to their bias.
To your point: check out Supermaven. I have no affiliation. But it has way more context than Copilot and is way better at suggestions that make sense in the codebase.
I kinda wonder what the percentage would be for my use of autocomplete. I only ever write out variable names once, the other times they are completed. Like I write "wacz" and it sees it matches with WArehouseZoneController and inputs that.
I probably type less than half the code I commit, even without AI assistance?
Certainly. That statistic is misleading without knowing what exactly it measures. The author no doubt knows it. I am surprised you're the only person on HN pointing it out.
Ok you’ve convinced me to give Aider a try. But is there a way to feed it documentation of a particular library or API? I want it to read the Stripe docs for me for example and then implement a feature.
A team in our company did I an actual refactoring for a customer using some generative AI. They claimed it gave them like 10-15% estimated speed boost - but then again, people who’d try this out probably also buy into the hype and will say they had a speed boost regardless. Also obviously no control to compare to, so not much gained there.
I’m not dismissing their achievement, but estimates is the most complicated thing in programming, almost as difficult to do as solving P=NP so I would be very cautious when someone says this.
