26 ms·

I am the founder of a niche SaaS (https://partsbox.com/ https://partsbox.com/ — software for managing electronic parts inventory and production). While I am som

by jwr 10mo ago

I am the founder of a niche SaaS (https://partsbox.com/ https://partsbox.com/ — software for managing electronic parts inventory and production). While I am somewhat worried about AI capabilities, I'm not losing too much sleep over it. The worry is that customers who do not realize the full depth of the problem will implement their own app using AI. But that happens today, too: people use spreadsheets to manage their electronic parts (please don't) and BOMs (bills of materials). The spreadsheet is my biggest competitor. I've been designing and building the software for 10 years now and most of the difficulty and complexity is not in the code. Coding is the last part, and the easiest one. The real value is in understanding the world (the processes involved) and modeling it in a way that cuts a good compromise between ease of use and complexity. Sadly, as I found out, once you spend a lot of time thinking and come up with a model, copycats will clone that (as well as they can, but superficially it will look similar).

ehnto 10mo ago
> The real value is in understanding the world (the processes involved) and modeling it in a way that cuts a good compromise between ease of use and complexity. Which I don't think can be replaced by AI in a lot of cases. I think in the software world we are used to things being shared, open and easily knowable, but a great deal of industry and enterprise domain knowledge is locked up inside in companies and will not be in the training data. That's why it's such a big deal for an enterprise to have on prem tools, to avoid leaking industry processes and "secrets" (the secrets are boring, but still secrets). A little career advice in there too I guess. At least for now, you're a bit more secure as a developer in industries that aren't themselves software, is my guess.
- jwr 10mo ago
  > Which I don't think can be replaced by AI in a lot of cases Yes. I try to visit my customers as often as I can, to learn how they work and to see the production processes on site. I consider it to be one of the most valuable things I can do for the future of my business.
a2code 10mo ago
I have two tech q about partsbox. Why Clojure? Why not CL (lack of saas related-features)?
- jwr 10mo ago
  Clojure is just better than CL in pretty much every respect. Excellent and well designed standard library, great concurrency primitives, core.async, built-in transducers (CL has SERIES which does a kind-of similar thing, but isn't as well designed and integrated) and the dominant immutability all let me write more maintainable code. Also, I can re-use model code on the client side (ClojureScript), so there is lots of code sharing, and I don't have to serialize to a crippled format (JSON), my data can pass from server to client and back intact (with sets, keywords, and other rich data types). I used to love CL and wrote quite a bit of code in it, but since Clojure came along I can't really see any reason to go back.
  - a2code 10mo ago
    I did not try Clojure so I cannot comment on how well implemented the features you quote are, when compared to CL. All I can say is that CL also provides much of the same functionality with its standard, cl-async, lparallel, parenscript, while (im)mutability is a matter of preference (IMO correct decision by CL) rather than dominance. The way I see it, is that CL is superior (opinion) due to reader macros and native compilation, rather than bytecode JVM.
    - jwr 10mo ago
      I used CL for many years before I switched to Clojure. And by "used" I mean really used it, not just for applications, but I also dove into CL implementations, for example I added return-from-frame (also known as debug-return) to CMUCL. So, I kind of know what I'm talking about :-) And I don't miss anything from CL: I honestly can't find a single reason to switch back to CL.
TeMPOraL 10mo ago
The problem IMO is simpler. You have a product, which sits between your users and what your users want. That product has an UI for users to operate. Many (most, I imagine) users would prefer to hire an assistant to operate that UI for them, since UI is not the actual value your service provides. Now, s/assistant/AI agent/ and you can see that your product turns into a tool call. So the simpler problem is that your product now becomes merely a tool call for AI agents. That's what users want. Many SaaS companies won't like that, because it removes their advertising channel and commoditizes their product. It's the same reason why API access to SaaS is usually restricted or not available for the users except biggest customers. LLMs defeat that by turning the entire human experience into an API, without explicit coding.
- MangoToupe 10mo ago
  > Many (most, I imagine) users would prefer to hire an assistant to operate that UI for them, since UI is not the actual value your service provides That's ridiculous. A good ui will improve on assistant in every way. Do assistants have some use? Sure—querying.
  - ben_w 10mo ago
    > A good ui will improve on assistant in every way. True. "Good" UI seems to be in short supply these days, even from trillion dollar corporations. But even with that, it is still not "ridiculous" for many to prefer to "hire an assistant to operate that UI for them". A lot of the complexity in UI is the balance between keeping common tasks highly visible without hiding the occasional-use stuff, allowing users to explore and learn more about what can be done without overwhelming them. If I want a spaceship in Blender and don't care which one you get — right now the spaceship models that any GenAI would give you are "pick your poison" between Diffusion models' weirdness and the 3D equivalent of the pelican-on-a-bike weirdness — the easiest UI is to say (or type) "give me a spaceship", not doing all the steps by hand. If you have some unformatted time series data and want to use it to forecast the next quarter, you could manually enter it into a spreadsheet, or you could say/type "here's a JPG of some time series data, use it to forecast the next quarter". Again, just to be clear, I agree with everyone saying current AI is only mediocre in performance, it does make mistakes and shouldn't be relied upon yet. But the error rates are going down, the task horizons they don't suck at are going up. I expect the money to run out before they get good enough to take on all SaaS, but at the same time they're already good enough to be interesting.
lonelyasacloud 10mo ago
Coding agents like Claude are just one line of AI making inroads. There are lot of nearly tasks that can be almost, but not quite, implemented effectivly with existing tools like Excel and Word. As they seek a return on their investments, are MS likely to target those nearly cases with AI in their Access, Excel, Word etc product lines?
dismalpedigree 10mo ago
Not specific to PartsBox, but we use Inventree (open source similar to PartsBox) and self host it. Over the past few months we noticed certain pain points in our workflow. Rather than looking for a new tool, we used Claude Code to write some backend services and some frontend modifications. Took 2 days of tinkering. Has easily saved that much time since we implemented it. While rolling the whole solution with an AI agent is not practical, taking a open source starting point and using AI to overcome specific workflow pain points as well as add features allows me to have a lower cost, specifically tailored solution to our needs.
- jwr 10mo ago
  I would include this in the "spreadsheet" metaphor. I do not know your use case, so please don't take this as addressed to you specifically, but I found that there is a learning/complexity problem: many people do not realize there is much more to inventory and production management than it seems. It might seem easy to AI-code something, only to find out later that things could have been done much better. This is actually a serious problem for me: my SaaS has a lot of very complex functionality under the hood, but it is not easily visible, and importantly it isn't necessarily appreciated when making a buying decision. Lot control is a good example: most people think it is only needed for coding batches of expiring products. In reality, it's an essential feature that pretty much everyone needs, because it lets you treat some inventory of the same part (e.g. a reel) differently from other inventory of this part (e.g. cut tape) and track those separately. AI-coding will help people get the features they know they need, but it won't guide them to the features they don't know they could use.
  - emanresuymsti 10mo ago
    "AI-coding will help people get the features they know they need, but it won't guide them to the features they don't know they could use." Amen to this.
- LunaSea 10mo ago
  Won't this breakdown if you need to pull new changes from the original project?
eikenberry 10mo ago
> Coding is the last part, and the easiest one. The real value is in understanding the world (the processes involved) and modeling it in a way that cuts a good compromise between ease of use and complexity. Coding and modeling are interleaved. Prototyping is basically thinking through the models you are considering. If you split the two, you'll end up with a bad model, bad software or both.

by jwr 10mo ago

ehnto 10mo ago

> The real value is in understanding the world (the processes involved) and modeling it in a way that cuts a good compromise between ease of use and complexity. Which I don't think can be replaced by AI in a lot of cases. I think in the software world we are used to things being shared, open and easily knowable, but a great deal of industry and enterprise domain knowledge is locked up inside in companies and will not be in the training data. That's why it's such a big deal for an enterprise to have on prem tools, to avoid leaking industry processes and "secrets" (the secrets are boring, but still secrets). A little career advice in there too I guess. At least for now, you're a bit more secure as a developer in industries that aren't themselves software, is my guess.

jwr 10mo ago
> Which I don't think can be replaced by AI in a lot of cases Yes. I try to visit my customers as often as I can, to learn how they work and to see the production processes on site. I consider it to be one of the most valuable things I can do for the future of my business.

a2code 10mo ago

I have two tech q about partsbox. Why Clojure? Why not CL (lack of saas related-features)?

jwr 10mo ago
Clojure is just better than CL in pretty much every respect. Excellent and well designed standard library, great concurrency primitives, core.async, built-in transducers (CL has SERIES which does a kind-of similar thing, but isn't as well designed and integrated) and the dominant immutability all let me write more maintainable code. Also, I can re-use model code on the client side (ClojureScript), so there is lots of code sharing, and I don't have to serialize to a crippled format (JSON), my data can pass from server to client and back intact (with sets, keywords, and other rich data types). I used to love CL and wrote quite a bit of code in it, but since Clojure came along I can't really see any reason to go back.
- a2code 10mo ago
  I did not try Clojure so I cannot comment on how well implemented the features you quote are, when compared to CL. All I can say is that CL also provides much of the same functionality with its standard, cl-async, lparallel, parenscript, while (im)mutability is a matter of preference (IMO correct decision by CL) rather than dominance. The way I see it, is that CL is superior (opinion) due to reader macros and native compilation, rather than bytecode JVM.
  - jwr 10mo ago
    I used CL for many years before I switched to Clojure. And by "used" I mean really used it, not just for applications, but I also dove into CL implementations, for example I added return-from-frame (also known as debug-return) to CMUCL. So, I kind of know what I'm talking about :-) And I don't miss anything from CL: I honestly can't find a single reason to switch back to CL.

TeMPOraL 10mo ago

The problem IMO is simpler. You have a product, which sits between your users and what your users want. That product has an UI for users to operate. Many (most, I imagine) users would prefer to hire an assistant to operate that UI for them, since UI is not the actual value your service provides. Now, s/assistant/AI agent/ and you can see that your product turns into a tool call. So the simpler problem is that your product now becomes merely a tool call for AI agents. That's what users want. Many SaaS companies won't like that, because it removes their advertising channel and commoditizes their product. It's the same reason why API access to SaaS is usually restricted or not available for the users except biggest customers. LLMs defeat that by turning the entire human experience into an API, without explicit coding.

MangoToupe 10mo ago
> Many (most, I imagine) users would prefer to hire an assistant to operate that UI for them, since UI is not the actual value your service provides That's ridiculous. A good ui will improve on assistant in every way. Do assistants have some use? Sure—querying.
- ben_w 10mo ago
  > A good ui will improve on assistant in every way. True. "Good" UI seems to be in short supply these days, even from trillion dollar corporations. But even with that, it is still not "ridiculous" for many to prefer to "hire an assistant to operate that UI for them". A lot of the complexity in UI is the balance between keeping common tasks highly visible without hiding the occasional-use stuff, allowing users to explore and learn more about what can be done without overwhelming them. If I want a spaceship in Blender and don't care which one you get — right now the spaceship models that any GenAI would give you are "pick your poison" between Diffusion models' weirdness and the 3D equivalent of the pelican-on-a-bike weirdness — the easiest UI is to say (or type) "give me a spaceship", not doing all the steps by hand. If you have some unformatted time series data and want to use it to forecast the next quarter, you could manually enter it into a spreadsheet, or you could say/type "here's a JPG of some time series data, use it to forecast the next quarter". Again, just to be clear, I agree with everyone saying current AI is only mediocre in performance, it does make mistakes and shouldn't be relied upon yet. But the error rates are going down, the task horizons they don't suck at are going up. I expect the money to run out before they get good enough to take on all SaaS, but at the same time they're already good enough to be interesting.

lonelyasacloud 10mo ago

Coding agents like Claude are just one line of AI making inroads. There are lot of nearly tasks that can be almost, but not quite, implemented effectivly with existing tools like Excel and Word. As they seek a return on their investments, are MS likely to target those nearly cases with AI in their Access, Excel, Word etc product lines?

dismalpedigree 10mo ago

Not specific to PartsBox, but we use Inventree (open source similar to PartsBox) and self host it. Over the past few months we noticed certain pain points in our workflow. Rather than looking for a new tool, we used Claude Code to write some backend services and some frontend modifications. Took 2 days of tinkering. Has easily saved that much time since we implemented it. While rolling the whole solution with an AI agent is not practical, taking a open source starting point and using AI to overcome specific workflow pain points as well as add features allows me to have a lower cost, specifically tailored solution to our needs.

jwr 10mo ago
I would include this in the "spreadsheet" metaphor. I do not know your use case, so please don't take this as addressed to you specifically, but I found that there is a learning/complexity problem: many people do not realize there is much more to inventory and production management than it seems. It might seem easy to AI-code something, only to find out later that things could have been done much better. This is actually a serious problem for me: my SaaS has a lot of very complex functionality under the hood, but it is not easily visible, and importantly it isn't necessarily appreciated when making a buying decision. Lot control is a good example: most people think it is only needed for coding batches of expiring products. In reality, it's an essential feature that pretty much everyone needs, because it lets you treat some inventory of the same part (e.g. a reel) differently from other inventory of this part (e.g. cut tape) and track those separately. AI-coding will help people get the features they know they need, but it won't guide them to the features they don't know they could use.
- emanresuymsti 10mo ago
  "AI-coding will help people get the features they know they need, but it won't guide them to the features they don't know they could use." Amen to this.
LunaSea 10mo ago
Won't this breakdown if you need to pull new changes from the original project?

eikenberry 10mo ago

> Coding is the last part, and the easiest one. The real value is in understanding the world (the processes involved) and modeling it in a way that cuts a good compromise between ease of use and complexity. Coding and modeling are interleaved. Prototyping is basically thinking through the models you are considering. If you split the two, you'll end up with a bad model, bad software or both.