8 ms·

Back To Atoms: Why we can stop building SaaS and build the future instead.

60 points by carllippert 2y ago

keeptrying 2y ago
How do you build Stripe payments with LLMs. Only very shallow SaaS may be killed with LLMs but not 98% of em.
- winrid 2y ago
  I can't even get an LLM to accurately demonstrate how to use the Stripe API except for very basic scenarios.
  - danpalmer 2y ago
    Having built a relatively standard, not particularly complex Stripe integration with 3DSecure, there is no way I'd trust code written by an AI for it. Hell, I set up code review sessions just to walk through the code and discuss. Transactionality, idempotency, what happens if a customer drops out of the flow at the 3DSecure prompt, how do you re-engage them, what happens if they want to continue, etc. It's hard both in terms of having a detailed understanding of the technology components, and in terms of understanding the UX requirements, customer support requirements, etc. Taking a payment with Stripe is a 5 minute job. Taking thousands of payments a day, without ever double charging someone, and without ever losing an order, while optimising the UX and allowing a customer support team to provide good service, that's a huge task. (Also no criticism of Stripe, having looked at Braintree circa 2019 I expect it would be significantly harder with them).
  - youainti 2y ago
    One change I expect to see over the next few years in API documentation is a series of complicated examples. This isn't for people to review, but for coding assistants to use as to inform RAG or something similar.
- aitchnyu 2y ago
  Did we figure out GenAI thats beyond a mashup of Conway's Game of Life tutorials and Stack Overflow answers, which humans created hoping to be credited?
rnewme 2y ago
Yea but business needs are constantly evolving, most of the time one step ahead of the accurate understanding of what needs to be done. We get paid, in theory, to build that, but main expense is not actually building, it's finding what needs to be done and then maintaining it. No ai can address that imho
- chaos_emergent 2y ago
  If you don’t think an AI could do that then I think you need more imagination and to follow the OOMs
  - packetlost 2y ago
    I think "could" is substantially different from "practically will in a time-frame that matters" and the former is being used as a shortcut for the latter.
  - elicksaur 2y ago
    “OOMs”? Not sure what that means, but I’m guessing you don’t mean out of memory errors.
    - chaos_emergent 2y ago
      orders of magnitude
- dontchooseanick 2y ago
  Came exactly to say this. We often don't solve technical problems or write algorithm. We're here to understand a client who doesn't know what he wants. Good luck for the general purpose AI to solve this [1] "If Architects had to work like Programmers" [1] https://archive.is/r4l6C https://archive.is/r4l6C
danpalmer 2y ago
Let's assume that AI is human level (it's not, but humour me). Let's assume that businesses can buy an off-the-shelf black box of AI that will run whatever business IT process they want to run. Customer data goes in, actions for humans to perform come out. It's basically an automated call centre full of humans. Now something goes wrong. A customer isn't getting what they ordered. What do you do about it? Do you just drop the customer? Well that's lost revenue, or an easy win for another company, you don't want to do that. Maybe you tell the AI to fix the customer in a particular way and hope it will do so. If it does, all it takes is having a human available to fix it (a customer support dept, even if tier 2), but if it doesn't fix it now you still need all the software you needed before to fix it. I've built line-of-business software before, and I've over-automated things before. Being able to inspect the state of systems, and being able to manually change things in software, is incredibly valuable. This is why spreadsheets run businesses. My bet is that companies will still mostly buy the software they were buying before, but that some of the interactions with that software that humans would have done before, AI tools will action instead. All the same CMSs, CRMs, ERPs, and other TLAs will exist, they'll just be augmented with AI. They'll probably do some more interesting things with that AI, but they'll still exist in fundamentally the same way. Companies are not going to entrust their business to a black box that sometimes says no.
- Harmohit 2y ago
  I completely agree. I do not believe that AI is a disruptive technology like computers and the internet was. It will at best be a + 25 % efficiency for business processes - which is awesome and I am all for it. But it is not a game changer that will fundamentally change how things are done.
  - danpalmer 2y ago
    As with all disruptive technologies. Uber was 10x better than taxis when it was being pumped full of VC cash, now it's better enough that it's my go-to choice, but not by a 2x margin. AirBnB was 10x better than hotels when it was being pumped full of cash, hadn't realised that people might trash places, and chancers hadn't filled the platform with crap properties, now it's a wash and only another option alongside hotels. Bitcoin, well, lol. AI looks 10x better than the alternative, while it's being pumped full of VC cash and while we haven't realised all the downsides (because so far we've only seen idealised marketing promises). It'll work out to be an expensive and not much better option.
FredPret 2y ago
Software is eating the world, but according to [0] there's only 27 million of us in the whole world. We need all the tools we can get to amplify our efforts. Further, I find that AI makes it much easier to dabble outside my wheelhouse and to interface with other technical people much quicker and better. Where I used to have to read endlessly to understand the context of unknown-to-me technology, now I can get that in seconds. Previously this luxury was only available to large corporates who can hire a wide variety of experts. I think we'll see dramatically more 1-10 person software businesses in future. [0] https://en.wikipedia.org/wiki/Software_engineering_demographics https://en.wikipedia.org/wiki/Software_engineering_demograph...
- bruce511 2y ago
  There's no shortage of technical info on the internet. If you want to learn about networks or printing or Active Directory or databases or Windows APIs or Web Servers or security or whatever you like, it's sll there. Yes, the AI let's you ask questions and get answers. It's a more interactive approach, and it helps to clarify thinking, but having a solid foundation will always lead to better understanding. As someone who has always worked in a small company, and as someone who has by necessity had to learn many different parts of IT, up and down the IT stack, it's always been possible. Equally I work with businesses, the vast majority of whom are less than 10 people - and those small businesses produce code every day which is used all over the place. Turns out, if you knew a domain really well, could code, and don't have 3 layers of management, good things can get done really quickly.
  - wruza 2y ago
    There's no shortage of technical info on the internet Right, but last time I checked it was behind pages of seo bullshit. I suspect many of us still have that inertia since our learning age, so knowing where to get information feels natural, while in reality it’s knowledge aquired with years or decades. Imagine forgetting all you know and trying to figure out the learning meta of that medium article. Multiply it by what google “search” is today. You’re basically screwed.
  - FredPret 2y ago
jongjong 2y ago
My experience tells me that the world of atoms requires a perfectly level playing field in order to support reliable businesses. The world of atoms is inherently stochastic with many unexpected events. We cannot plan for the future if the government intervenes in the markets. It's hard enough to plan for 'natural' events. We shouldn't also have to worry about the thousands of ways in which the government could destroy the incentives and dynamics that are vital to your business. In the world of atoms, failure incurs high costs. Whenever you opt for value creation as a business, you put yourself at the mercy of government policies because the government, through monetary policy, can distort markets in extremely harmful, unpredictable ways.
- ehnto 2y ago
  This is true for even regular software APIs. By all accounts, we could have a beautifully interwoven quilt of standardized APIs that connect any system to any other system with ease. But business realities mean that just cannot exist, it actually doesn't survive first contact with business logic let alone all the auth* issues. An atomic level of interoperability will hit this issue immediately. If they are seeing AI as a sort of fuzzy connector between components, it is a non-starter from my perspective. Even if you had squishy brains doing the work, it wouldn't pan out.
altdataseller 2y ago
“ We are regularly building and launching rockets into space. We are fighting climate change with science and industry” 99% of the people who were building SaaS probably arent skilled enough to launch rockets or fight climate change with science. Now if youre talking about kids still in school, who have yet to decide on what to major, then yes perhaps most of them will forgo learning how to build software and pursue hard tech fields instead
- intelVISA 2y ago
  Excuse me it takes at least 3 neurons to npm run dev
croes 2y ago
>Humans are good at fixing problems mostly because we are good at making them. Are we? Just look at climate change. >We are fighting climate change with science and industry. We are also fighting fighting climate change with pseudo science and industry money. We aren't good at fixing things, we are good at surviving even if we don't fix things. Maybe lots of people die, maybe we repeat the same mistakes over and over again but at least some will survive to start all over again. Every high culture went down the hill and a new emerged from the rubbles of the previous one. But that's not fixing, that surviving.
torginus 2y ago
> You can even get VC money to build military defense hardware. How do you create exponential growth when you are building military hardware?
- perilunar 2y ago
  Get your politicians to start a few wars?
torginus 2y ago
The world of atoms is being disrupted as well. Spurned by the anxiety of AI replacing my codemonkeying skills, I decided to get knJust recently I bought a 3d printer, and after spending a couple weeks learning CAD, it's insane how much of the act of physically making things has turned into just modelling stuff on the computer and hitting print. I've designed 'complex' (for me haha) mechanisms with levers, gears and springs, the manufacturing of which would've required a machine shop, multiple trips to the hardware store, tons of hands-on experience and lots of careful measuring and swearing, if not for the miracle of CAD software and my printer. I just modelled the thing, hit print, and after 2-3 iterations of fixing dumb mistakes, I had a product that worked, and fit to tighter tolerances than what I could've realistically made by hand. Of course not everything can be made out of cheap thermoplastic, but the jump to higher grade materials is available even today for the everyman.
var_cw 2y ago
imo most of the software play is going to be an infra play where the max of value is being provided/captured by the big tech there will def be some opportunities for mid-tier software(not necessarily saas) companies where they end up leveraging either some niche distribution or some neat packaging. another ray of hope is - distribution takes time, there is a long cycle of adoption for any new tech. currently we are in "the chasm" and some of us(us being software builders) will end up benefitting from a long adoption curves. but i have my doubts if it's a vc backed thing as it will be a slow cycle.
jenniferw011 2y ago
[flagged]
ashleygh 2y ago
[flagged]

60 points by carllippert 2y ago

keeptrying 2y ago

How do you build Stripe payments with LLMs. Only very shallow SaaS may be killed with LLMs but not 98% of em.

winrid 2y ago
I can't even get an LLM to accurately demonstrate how to use the Stripe API except for very basic scenarios.
- danpalmer 2y ago
  Having built a relatively standard, not particularly complex Stripe integration with 3DSecure, there is no way I'd trust code written by an AI for it. Hell, I set up code review sessions just to walk through the code and discuss. Transactionality, idempotency, what happens if a customer drops out of the flow at the 3DSecure prompt, how do you re-engage them, what happens if they want to continue, etc. It's hard both in terms of having a detailed understanding of the technology components, and in terms of understanding the UX requirements, customer support requirements, etc. Taking a payment with Stripe is a 5 minute job. Taking thousands of payments a day, without ever double charging someone, and without ever losing an order, while optimising the UX and allowing a customer support team to provide good service, that's a huge task. (Also no criticism of Stripe, having looked at Braintree circa 2019 I expect it would be significantly harder with them).
- youainti 2y ago
  One change I expect to see over the next few years in API documentation is a series of complicated examples. This isn't for people to review, but for coding assistants to use as to inform RAG or something similar.
aitchnyu 2y ago
Did we figure out GenAI thats beyond a mashup of Conway's Game of Life tutorials and Stack Overflow answers, which humans created hoping to be credited?

rnewme 2y ago

Yea but business needs are constantly evolving, most of the time one step ahead of the accurate understanding of what needs to be done. We get paid, in theory, to build that, but main expense is not actually building, it's finding what needs to be done and then maintaining it. No ai can address that imho

chaos_emergent 2y ago
If you don’t think an AI could do that then I think you need more imagination and to follow the OOMs
- packetlost 2y ago
  I think "could" is substantially different from "practically will in a time-frame that matters" and the former is being used as a shortcut for the latter.
- elicksaur 2y ago
  “OOMs”? Not sure what that means, but I’m guessing you don’t mean out of memory errors.
  - chaos_emergent 2y ago
    orders of magnitude
dontchooseanick 2y ago
Came exactly to say this. We often don't solve technical problems or write algorithm. We're here to understand a client who doesn't know what he wants. Good luck for the general purpose AI to solve this [1] "If Architects had to work like Programmers" [1] https://archive.is/r4l6C https://archive.is/r4l6C

danpalmer 2y ago

Let's assume that AI is human level (it's not, but humour me). Let's assume that businesses can buy an off-the-shelf black box of AI that will run whatever business IT process they want to run. Customer data goes in, actions for humans to perform come out. It's basically an automated call centre full of humans. Now something goes wrong. A customer isn't getting what they ordered. What do you do about it? Do you just drop the customer? Well that's lost revenue, or an easy win for another company, you don't want to do that. Maybe you tell the AI to fix the customer in a particular way and hope it will do so. If it does, all it takes is having a human available to fix it (a customer support dept, even if tier 2), but if it doesn't fix it now you still need all the software you needed before to fix it. I've built line-of-business software before, and I've over-automated things before. Being able to inspect the state of systems, and being able to manually change things in software, is incredibly valuable. This is why spreadsheets run businesses. My bet is that companies will still mostly buy the software they were buying before, but that some of the interactions with that software that humans would have done before, AI tools will action instead. All the same CMSs, CRMs, ERPs, and other TLAs will exist, they'll just be augmented with AI. They'll probably do some more interesting things with that AI, but they'll still exist in fundamentally the same way. Companies are not going to entrust their business to a black box that sometimes says no.

Harmohit 2y ago
I completely agree. I do not believe that AI is a disruptive technology like computers and the internet was. It will at best be a + 25 % efficiency for business processes - which is awesome and I am all for it. But it is not a game changer that will fundamentally change how things are done.
- danpalmer 2y ago
  As with all disruptive technologies. Uber was 10x better than taxis when it was being pumped full of VC cash, now it's better enough that it's my go-to choice, but not by a 2x margin. AirBnB was 10x better than hotels when it was being pumped full of cash, hadn't realised that people might trash places, and chancers hadn't filled the platform with crap properties, now it's a wash and only another option alongside hotels. Bitcoin, well, lol. AI looks 10x better than the alternative, while it's being pumped full of VC cash and while we haven't realised all the downsides (because so far we've only seen idealised marketing promises). It'll work out to be an expensive and not much better option.

FredPret 2y ago

Software is eating the world, but according to [0] there's only 27 million of us in the whole world. We need all the tools we can get to amplify our efforts. Further, I find that AI makes it much easier to dabble outside my wheelhouse and to interface with other technical people much quicker and better. Where I used to have to read endlessly to understand the context of unknown-to-me technology, now I can get that in seconds. Previously this luxury was only available to large corporates who can hire a wide variety of experts. I think we'll see dramatically more 1-10 person software businesses in future. [0] https://en.wikipedia.org/wiki/Software_engineering_demographics https://en.wikipedia.org/wiki/Software_engineering_demograph...

bruce511 2y ago
There's no shortage of technical info on the internet. If you want to learn about networks or printing or Active Directory or databases or Windows APIs or Web Servers or security or whatever you like, it's sll there. Yes, the AI let's you ask questions and get answers. It's a more interactive approach, and it helps to clarify thinking, but having a solid foundation will always lead to better understanding. As someone who has always worked in a small company, and as someone who has by necessity had to learn many different parts of IT, up and down the IT stack, it's always been possible. Equally I work with businesses, the vast majority of whom are less than 10 people - and those small businesses produce code every day which is used all over the place. Turns out, if you knew a domain really well, could code, and don't have 3 layers of management, good things can get done really quickly.
- wruza 2y ago
  There's no shortage of technical info on the internet Right, but last time I checked it was behind pages of seo bullshit. I suspect many of us still have that inertia since our learning age, so knowing where to get information feels natural, while in reality it’s knowledge aquired with years or decades. Imagine forgetting all you know and trying to figure out the learning meta of that medium article. Multiply it by what google “search” is today. You’re basically screwed.
- FredPret 2y ago

jongjong 2y ago

My experience tells me that the world of atoms requires a perfectly level playing field in order to support reliable businesses. The world of atoms is inherently stochastic with many unexpected events. We cannot plan for the future if the government intervenes in the markets. It's hard enough to plan for 'natural' events. We shouldn't also have to worry about the thousands of ways in which the government could destroy the incentives and dynamics that are vital to your business. In the world of atoms, failure incurs high costs. Whenever you opt for value creation as a business, you put yourself at the mercy of government policies because the government, through monetary policy, can distort markets in extremely harmful, unpredictable ways.

ehnto 2y ago
This is true for even regular software APIs. By all accounts, we could have a beautifully interwoven quilt of standardized APIs that connect any system to any other system with ease. But business realities mean that just cannot exist, it actually doesn't survive first contact with business logic let alone all the auth* issues. An atomic level of interoperability will hit this issue immediately. If they are seeing AI as a sort of fuzzy connector between components, it is a non-starter from my perspective. Even if you had squishy brains doing the work, it wouldn't pan out.

altdataseller 2y ago

“ We are regularly building and launching rockets into space. We are fighting climate change with science and industry” 99% of the people who were building SaaS probably arent skilled enough to launch rockets or fight climate change with science. Now if youre talking about kids still in school, who have yet to decide on what to major, then yes perhaps most of them will forgo learning how to build software and pursue hard tech fields instead

intelVISA 2y ago
Excuse me it takes at least 3 neurons to npm run dev

croes 2y ago

>Humans are good at fixing problems mostly because we are good at making them. Are we? Just look at climate change. >We are fighting climate change with science and industry. We are also fighting fighting climate change with pseudo science and industry money. We aren't good at fixing things, we are good at surviving even if we don't fix things. Maybe lots of people die, maybe we repeat the same mistakes over and over again but at least some will survive to start all over again. Every high culture went down the hill and a new emerged from the rubbles of the previous one. But that's not fixing, that surviving.

torginus 2y ago

> You can even get VC money to build military defense hardware. How do you create exponential growth when you are building military hardware?

perilunar 2y ago
Get your politicians to start a few wars?

torginus 2y ago

The world of atoms is being disrupted as well. Spurned by the anxiety of AI replacing my codemonkeying skills, I decided to get knJust recently I bought a 3d printer, and after spending a couple weeks learning CAD, it's insane how much of the act of physically making things has turned into just modelling stuff on the computer and hitting print. I've designed 'complex' (for me haha) mechanisms with levers, gears and springs, the manufacturing of which would've required a machine shop, multiple trips to the hardware store, tons of hands-on experience and lots of careful measuring and swearing, if not for the miracle of CAD software and my printer. I just modelled the thing, hit print, and after 2-3 iterations of fixing dumb mistakes, I had a product that worked, and fit to tighter tolerances than what I could've realistically made by hand. Of course not everything can be made out of cheap thermoplastic, but the jump to higher grade materials is available even today for the everyman.

var_cw 2y ago

imo most of the software play is going to be an infra play where the max of value is being provided/captured by the big tech there will def be some opportunities for mid-tier software(not necessarily saas) companies where they end up leveraging either some niche distribution or some neat packaging. another ray of hope is - distribution takes time, there is a long cycle of adoption for any new tech. currently we are in "the chasm" and some of us(us being software builders) will end up benefitting from a long adoption curves. but i have my doubts if it's a vc backed thing as it will be a slow cycle.

jenniferw011 2y ago

[flagged]

ashleygh 2y ago

[flagged]