The landscape today has changed dramatically. We all know AI is getting really good at churning out basic code and MVPs. But I still believe there's real value in human creativity and understanding the deeper nuances of building great products.
Here's what I'm thinking: 1. Keep it small and focused (just 3-4 of us) - engineers and designers who genuinely care about building great stuff 2. Take on meaningful client work while building our own products 3. Get involved in open source and build useful tools for the community 4. Keep working on our own products (starting with my form builder)
What I'm inspired by: 1. 37signals: Started as a web design agency, created Basecamp, and eventually built Hey.com 2. überdosis: Built products for clients and worked on an open source text editor called TipTap (tiptap.dev) which is now their primary focus
What I want to avoid: 1. Becoming yet another "digital transformation" agency (you know the type) 2. Racing to the bottom against AI-generated MVPs 3. The typical agency trap of scaling up with juniors and PMs 4. Saying yes to every project that comes along
Some things I keep wondering about: 1. How would you position an agency today with AI in the picture? 2. Anyone here managed to build their own products while running client work? 3. What kind of work should we focus on that AI won't easily replace? 4. Has anyone grown their agency through open source work?
Really want to hear from folks who: 1. Have built small but solid agencies 2. Are actually using AI in their dev workflow (not just talking about it) 3. Have juggled client work with building products 4. Have grown through open source
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.
I think you would have to design your code system on the idea of simplicity and good building blocks. Then iterate on top of that to customize what the customer wants.
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.
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