Consistently my favorite substack read. One issue with custom software is the myriad of bugs and forgotten use cases. I wonder if we get an explosion of stable, secure open source instead. Maybe AI mostly just glues existing packages together.
Bugs, missing features, etc. will certainly be a factor. Basically, for this to work, it needs to be easy to fix bugs (and add missing features) as they arise.
More generally, there are a lot of open questions about how things will play out – open vs. closed-source, bespoke vs. large-audience, etc. I'm sure the reality will be some messy mix.
This may well be true but it also suggests that attributing any productivity improvement to AI coding assistants is going to be very hard to measure as it is so diffuse.
This is great -- but as I think about the domain I mostly work in -- web sites for e-commerce -- it's unclear if the analogy holds. A company that's trying to sell things still has just one web site, showing its products. (Or, in the case of a marketplace, its sellers' products.) What would it even mean to be bespoke, simple software in that context?
The current trends in e-commerce seem to be "agentic" -- e.g., https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/nrf-2026/ . Instead of searching a web site to find shoes, you interact with an AI agent that more-or-less simulates a sales person. "I'm looking for sneakers similar to the ones I'm wearing now, but wider and a little bit flashier", or whatever.
But that's not custom software, in the sense that this article postulates. So what would be?
As I see it, there are two sets of users here: sellers and consumers. Each might want their own custom experience. A seller might have unique preferences for how to manage inventory, set prices, ship orders, anticipate stocking needs, etc. A user might have idiosyncratic ways of choosing products ("I want a sneaker that's mentioned positively on the AllTrails forum"), or they might want to build a custom app that allows their children to place orders but only under certain tight limits on cost and type of product, etc.
Consistently my favorite substack read. One issue with custom software is the myriad of bugs and forgotten use cases. I wonder if we get an explosion of stable, secure open source instead. Maybe AI mostly just glues existing packages together.
Bugs, missing features, etc. will certainly be a factor. Basically, for this to work, it needs to be easy to fix bugs (and add missing features) as they arise.
More generally, there are a lot of open questions about how things will play out – open vs. closed-source, bespoke vs. large-audience, etc. I'm sure the reality will be some messy mix.
This may well be true but it also suggests that attributing any productivity improvement to AI coding assistants is going to be very hard to measure as it is so diffuse.
This is great -- but as I think about the domain I mostly work in -- web sites for e-commerce -- it's unclear if the analogy holds. A company that's trying to sell things still has just one web site, showing its products. (Or, in the case of a marketplace, its sellers' products.) What would it even mean to be bespoke, simple software in that context?
The current trends in e-commerce seem to be "agentic" -- e.g., https://blog.google/innovation-and-ai/infrastructure-and-cloud/google-cloud/nrf-2026/ . Instead of searching a web site to find shoes, you interact with an AI agent that more-or-less simulates a sales person. "I'm looking for sneakers similar to the ones I'm wearing now, but wider and a little bit flashier", or whatever.
But that's not custom software, in the sense that this article postulates. So what would be?
Fascinating questions!
As I see it, there are two sets of users here: sellers and consumers. Each might want their own custom experience. A seller might have unique preferences for how to manage inventory, set prices, ship orders, anticipate stocking needs, etc. A user might have idiosyncratic ways of choosing products ("I want a sneaker that's mentioned positively on the AllTrails forum"), or they might want to build a custom app that allows their children to place orders but only under certain tight limits on cost and type of product, etc.