Great article, but I have to ask a background question that will likely make me sound lazy: what are you managing that you need that kind of fidelity and productivity bandwidth? Even though the agent is working constantly, that still seems like an incredible amount to manage with any system. Again, not criticizing, and this article is fantastic, but it's just what I kept asking myself while reading. Best, Gary
It's a very good question, actually, and many people are asking a very similar one. Why do we even need that kind of power? That's a very good one.
For me, I am doing a lot of things in parallel, and the AI agent just enabled me to touch many things in an execution era very, very fast. What this is, I mean, when I was starting to build my first iteration of an AI agent, it was basics, right? Like summarize some email or research some things or find some specific data I want, very, very low things, I would say. Then I was doing more and more. For example, some things I started to automate with the agent, like maintenance of my server. I started really low with very separate and isolated things, and then I started to do that to check my security; then I expanded this to take care of my little shop; then I started more and more in e-commerce, because I'm by day also an e-commerce manager, so it makes sense for me to automate many, many things. That's only, you know, a tip of the iceberg. I can talk about that daily, but the thing is that I was building that over time, and when I started I also was like, "Okay, so what to do?" Because I am also a huge fan of Zapier and Make and that kind of tools, I rather had some kind of idea what I can do with this, and yeah.
I wonder if you use test frameworks when you develop, whether you follow TDD (including integration and e2e testing from the start), whether you follow DDD, and whether you follow SDD. I found that if you follow DDD at the core, it makes things very predictable. So the combination of these three works like magic when it comes to software development- bringing good old best practices into the world of hyperactive development with AI. Then whatever one builds becomes and stays stable.
Honestly, minimal formal TDD on the agent code. It rewrites too often to justify heavy test coverage at that stage. What did help was keeping clear domain separation between components (messaging, scheduling, execution) → which is DDD in spirit if not strictly in practice.
The irony is the software I replaced was over-engineered with too much abstraction. The replacement is far simpler and more stable precisely because it has less architecture. Maybe the lesson is that good practices work best when the problem is stable. When the problem is still figuring out what it is, you need room to throw things away fast.
A really interesting deep dive and quality article! Thank you for that. I’m in the process of starting my journey here and your post brings both the calmness and clarity that resonates with me (contrary to the many hectic, superficial “noise-articles” that are popping up everywhere).
I wonder and am curious about how AI & agentic working has impacted
your day-to-day life over the last years? I scrolled all the way back to the April 2023 post about RemoteRise and it seems like quite a journey in 3 years :)
Great article, but I have to ask a background question that will likely make me sound lazy: what are you managing that you need that kind of fidelity and productivity bandwidth? Even though the agent is working constantly, that still seems like an incredible amount to manage with any system. Again, not criticizing, and this article is fantastic, but it's just what I kept asking myself while reading. Best, Gary
It's a very good question, actually, and many people are asking a very similar one. Why do we even need that kind of power? That's a very good one.
For me, I am doing a lot of things in parallel, and the AI agent just enabled me to touch many things in an execution era very, very fast. What this is, I mean, when I was starting to build my first iteration of an AI agent, it was basics, right? Like summarize some email or research some things or find some specific data I want, very, very low things, I would say. Then I was doing more and more. For example, some things I started to automate with the agent, like maintenance of my server. I started really low with very separate and isolated things, and then I started to do that to check my security; then I expanded this to take care of my little shop; then I started more and more in e-commerce, because I'm by day also an e-commerce manager, so it makes sense for me to automate many, many things. That's only, you know, a tip of the iceberg. I can talk about that daily, but the thing is that I was building that over time, and when I started I also was like, "Okay, so what to do?" Because I am also a huge fan of Zapier and Make and that kind of tools, I rather had some kind of idea what I can do with this, and yeah.
Thank you for your answer Pawel, it makes sense and it sounds like you found a good solution.
awesome read man! really interesting to see the updates. the hard urls and crons… i’ve definitely been there!
awesome update :)
Thanks! Just sharing my progress here, happy see people resonate with it!
hell yeah!
I wonder if you use test frameworks when you develop, whether you follow TDD (including integration and e2e testing from the start), whether you follow DDD, and whether you follow SDD. I found that if you follow DDD at the core, it makes things very predictable. So the combination of these three works like magic when it comes to software development- bringing good old best practices into the world of hyperactive development with AI. Then whatever one builds becomes and stays stable.
Honestly, minimal formal TDD on the agent code. It rewrites too often to justify heavy test coverage at that stage. What did help was keeping clear domain separation between components (messaging, scheduling, execution) → which is DDD in spirit if not strictly in practice.
The irony is the software I replaced was over-engineered with too much abstraction. The replacement is far simpler and more stable precisely because it has less architecture. Maybe the lesson is that good practices work best when the problem is stable. When the problem is still figuring out what it is, you need room to throw things away fast.
A really interesting deep dive and quality article! Thank you for that. I’m in the process of starting my journey here and your post brings both the calmness and clarity that resonates with me (contrary to the many hectic, superficial “noise-articles” that are popping up everywhere).
I wonder and am curious about how AI & agentic working has impacted
your day-to-day life over the last years? I scrolled all the way back to the April 2023 post about RemoteRise and it seems like quite a journey in 3 years :)
WOAH, RemoteRise Project was wild and very early back then :D
As for life - I also wrote about productivity paradox some weeks ago - more productive, also more tiered :D
The timing is a bit ironic. This weekend I build a custom kanban board to integrate client projects.
Hahah, similar minds!