I have tried every single tool I can get my hands on in this space. From the Codeium VS Code extension, introduced a few years ago, which has now evolved into the powerful Windsurf IDE, as well as others like Blackbox, Refraction.dev (low-key super awesome still), v0, Replit, others I can't remember right now, and of course one of the most popular, Cursor.
They all were wonderful in their own ways, each with its own style, quirks, and various ways of tackling the automation and streamlining the prompt -> model output -> codebase flow. Every time a new model is released, I tend to go back and try them all again, to see how they improve upon previous iterations of the tools they each provide. This caused me to spend quite a bit in subscription fees, as I'm sure all of you have as well.
I think at this point I have gone back and forth between Cursor and Windsurf at least four times, but who's counting, right? I do my best to make evidence-based recommendations, as hype-free as possible, to my friends and colleagues on which tools will help them more efficiently develop and ship the software and hardware they are all feverishly building. At this point, they know if I tell them that switching to a new tool or an old one that has been updated, I am doing it out of love for the game, not to promote things. Well, I've started using a new tool, recommended to me by my friend Robert Scoble (@Scobleizer), and it's the best one yet. Factory AI (by The San Francisco AI Factory Inc.)
I don't need to explain how coding agents work to y'all, so I'm not going to do that. What I will tell you is that the team behind it seems to be a group of some of the most genuine people in the space with talent to produce what I consider to be the SOTA code agent framework to date (5/29/25). This could change overnight considering the pace of acceleration we are all barely keeping up with, but I would give it a try because I'm absolutely positive you won't be disappointed.
Here is my workflow now (use whatever tools you want, not here to argue over that stuff):
1. Use Windsurf to flesh out the structure and initial code, push to GitHub.
2. Decide what features I want to add, do some research using Grok/Sonnet-3.6/o3/o4-mini APIs (yes, I'm crazy).
3. Create a development plan and prompt using the Anthropic Workbench (my favorite prompt building method).
4. Send the plan and prompt to Factory AI's Code Droid (code agent).
After testing everything that's built, I continue through their different Droids (Reliability, Knowledge, and Product) until I have what I envisioned in the first place.
What I am very excited about and has opened up more free time for me, is the fact that the Factory Droids work until they are done, and their ability to ingest and understand VERY large codebases is unmatched. In my experience (so far), the need for human intervention has been almost non-existent; the agents just chug along and get the work done while I go for a walk to clear my head.
We live in the future, and I'm loving every second of it!
They all were wonderful in their own ways, each with its own style, quirks, and various ways of tackling the automation and streamlining the prompt -> model output -> codebase flow. Every time a new model is released, I tend to go back and try them all again, to see how they improve upon previous iterations of the tools they each provide. This caused me to spend quite a bit in subscription fees, as I'm sure all of you have as well.
I think at this point I have gone back and forth between Cursor and Windsurf at least four times, but who's counting, right? I do my best to make evidence-based recommendations, as hype-free as possible, to my friends and colleagues on which tools will help them more efficiently develop and ship the software and hardware they are all feverishly building. At this point, they know if I tell them that switching to a new tool or an old one that has been updated, I am doing it out of love for the game, not to promote things. Well, I've started using a new tool, recommended to me by my friend Robert Scoble (@Scobleizer), and it's the best one yet. Factory AI (by The San Francisco AI Factory Inc.)
I don't need to explain how coding agents work to y'all, so I'm not going to do that. What I will tell you is that the team behind it seems to be a group of some of the most genuine people in the space with talent to produce what I consider to be the SOTA code agent framework to date (5/29/25). This could change overnight considering the pace of acceleration we are all barely keeping up with, but I would give it a try because I'm absolutely positive you won't be disappointed. Here is my workflow now (use whatever tools you want, not here to argue over that stuff):
1. Use Windsurf to flesh out the structure and initial code, push to GitHub. 2. Decide what features I want to add, do some research using Grok/Sonnet-3.6/o3/o4-mini APIs (yes, I'm crazy). 3. Create a development plan and prompt using the Anthropic Workbench (my favorite prompt building method). 4. Send the plan and prompt to Factory AI's Code Droid (code agent).
After testing everything that's built, I continue through their different Droids (Reliability, Knowledge, and Product) until I have what I envisioned in the first place.
What I am very excited about and has opened up more free time for me, is the fact that the Factory Droids work until they are done, and their ability to ingest and understand VERY large codebases is unmatched. In my experience (so far), the need for human intervention has been almost non-existent; the agents just chug along and get the work done while I go for a walk to clear my head.
We live in the future, and I'm loving every second of it!
Go try it out at Factory dot ai