You mentioned 37 Signals so I think its appropriate to mention the scaffolding in Rails has been there for a long time. If you can carefully articulate the data models and data types, anyone can build a base MVP structure with `rails generate scaffold`. I've asked AI to generate code and honestly, the quality from the Rails scaffold is more consistent (for now). You're better off asking AI to help you craft the scaffold commands these days than having it generate the code itself.
I would suggest getting started on the client side of things with your 3-4 people first. Clients will have problems that need solving and in solving those you may find yourself building something over and over and that would be a good candidate for an open source tool. The technical side of a digital business isn't the hard part for developers.
This may sound cynical but there is probably a need for an agency type that can take "no code" and/or AI generated apps/solutions and take them beyond what no-code or AI platforms can do. Like small companies that have some AI generated stuff and don't have staff to take that to the next level. Charge them a monthly retainer for X hours of maintenance per month so they don't have to deal with it. Quote them for projects to take their AI generated app/solution to the next level or to make changes, etc. Again, you may find patterns once you have a few clients.
> 3. What kind of work should we focus on that AI won't easily replace?
I have enough confidence to answer just this question. You should definitely leverage AI where it's good at, as a tool for improving the quality of your work. But what I noticed, since the advent of LLMs for coding, is that quality of software hasn't improved, maybe output(? but can't be sure). There's been a lot of talk in the industry about decreasing quality of software in the recent years, way before than LLMs. So my obvious answer to that is to focus on products where a great UX makes the whole difference :) Maybe a bit generic but I believe there's a great value in that. A while ago I heard Casey Muratori(search the name if you don't know who he's) claiming that a product that would focus on performance alone could be a successful product, even without coming up with anything new, just copy something and make it better. I believe that nowadays that's true and there are a lot of opportunities out there given how bad some software is. E.g. Zed editor
If you need to talk more and/or need some dev workforce feel free to contact me :) amirmani at fastmail dot com
You've laid out the dream for a lot of developers around the world.
Agency life is probably 80% sales and 20% delivery. The agency pattern you've mentioned is common because it consistently works and everyone (myself included) fawns over the basecamp model but they're about the only ones who've made it work.
AI-generated SaaS tools just remind me of white-label SaaS products that were the hot thing however many years ago. They do about 50% of what the customer wants and a PITA to customize.
Having said that if I started again I'd still build one or more AI-centric B2B products with a focus on closed AI systems and look for agency work (ad-hoc development) as an offshoot.
You'd be able to build up a client base with your B2B products and have tangible examples to demonstrate what you're capable of while playing on the fact that the client's data is secure and never shared with third parties.
Great insights so far. I'd also love to dig deeper into the early days: What did your first 3-6 months look like? Especially curious about strategies beyond referrals - how did you directly find and land your first clients before having a strong portfolio?
You mentioned 37 Signals so I think its appropriate to mention the scaffolding in Rails has been there for a long time. If you can carefully articulate the data models and data types, anyone can build a base MVP structure with `rails generate scaffold`. I've asked AI to generate code and honestly, the quality from the Rails scaffold is more consistent (for now). You're better off asking AI to help you craft the scaffold commands these days than having it generate the code itself.
I would suggest getting started on the client side of things with your 3-4 people first. Clients will have problems that need solving and in solving those you may find yourself building something over and over and that would be a good candidate for an open source tool. The technical side of a digital business isn't the hard part for developers.
This may sound cynical but there is probably a need for an agency type that can take "no code" and/or AI generated apps/solutions and take them beyond what no-code or AI platforms can do. Like small companies that have some AI generated stuff and don't have staff to take that to the next level. Charge them a monthly retainer for X hours of maintenance per month so they don't have to deal with it. Quote them for projects to take their AI generated app/solution to the next level or to make changes, etc. Again, you may find patterns once you have a few clients.
> 3. What kind of work should we focus on that AI won't easily replace?
I have enough confidence to answer just this question. You should definitely leverage AI where it's good at, as a tool for improving the quality of your work. But what I noticed, since the advent of LLMs for coding, is that quality of software hasn't improved, maybe output(? but can't be sure). There's been a lot of talk in the industry about decreasing quality of software in the recent years, way before than LLMs. So my obvious answer to that is to focus on products where a great UX makes the whole difference :) Maybe a bit generic but I believe there's a great value in that. A while ago I heard Casey Muratori(search the name if you don't know who he's) claiming that a product that would focus on performance alone could be a successful product, even without coming up with anything new, just copy something and make it better. I believe that nowadays that's true and there are a lot of opportunities out there given how bad some software is. E.g. Zed editor
If you need to talk more and/or need some dev workforce feel free to contact me :) amirmani at fastmail dot com
You've laid out the dream for a lot of developers around the world.
Agency life is probably 80% sales and 20% delivery. The agency pattern you've mentioned is common because it consistently works and everyone (myself included) fawns over the basecamp model but they're about the only ones who've made it work.
AI-generated SaaS tools just remind me of white-label SaaS products that were the hot thing however many years ago. They do about 50% of what the customer wants and a PITA to customize.
Having said that if I started again I'd still build one or more AI-centric B2B products with a focus on closed AI systems and look for agency work (ad-hoc development) as an offshoot.
You'd be able to build up a client base with your B2B products and have tangible examples to demonstrate what you're capable of while playing on the fact that the client's data is secure and never shared with third parties.
Great insights so far. I'd also love to dig deeper into the early days: What did your first 3-6 months look like? Especially curious about strategies beyond referrals - how did you directly find and land your first clients before having a strong portfolio?
Errm. Don't start an "agency". It's literally outsourcing, and cost is most often the focus for most companies.