After several months of using AI for me, personally and at work, some general conclusions emerge.
First, I still meet seniors that simply say "AI hallucinate", and avoid research AI more deeply. I belive that ALL seniors have to go through one or two quality AI trainings to get on speed and understanding NOW!!! Even if they hate or afraid the thing. Maybe this is an example where they should be forced to do it. Old times are gone.
Second, "I'm back" - after 50 - I'm back, and feeling strong again, writing code. It is as easy, effortless, rewarding for me now as when I was 20. I believe even one-man products are possible again these times. I'm sure, as I always was that "the less people working on any product, the better". I have been dreaming about times when multitude meetings, for the sake of spending time and not delivering much, will not be needed anymore and more time for deep work will come. It's possible again.
Third, I can rise quality higher that was possible in the past - e.g. add extra validations, provide proactive diagnostics or fitness functions, quality dashboards checking everything around. IMHO ensuring quality while using AI is not to focus on code itself but on multitude completely new possibilities. There was not much time for it in the past. Now within a few hours I can built a tool saving weeks of work. Also many tools will appear, grow, mature and disappear. Some may bring brilliant changes to the way we work e.g. see Spec-Driven-Development: https://martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html - still very weak but promising. Maybe...
Finally. How it applies to software agencies? I believe they should deliberately seek, experiment, choose, evaluate patterns (both in software and processes). This would allow to deliver faster while rising quality in the same time. Paradox. No more approach "more projects, more people". But I'm afraid that managers of such agencies may not be capable of doing so. I recommend trying to code again to all that were good engineers in the past and later on become managers and spent decades without coding. If one loved coding she/he might be very, very surprised what is possible now.
Starting from the end of your share, I agree that the best way for software agency management to get their head around AI is to get back to coding now. That is why I got back to coding and did the AI_devs course (heavy on coding). It is a new paradigm that is hard to grasp if you won’t try it. It requires a new mental model to operate it.
The leverage AI gives a software developer, or should I say product creator now, is tremendous. You can be a one-man army and create a product. And I plan on doing so! ;-) I’m writing a book on Augmented Mind now that creates a foundational model for my ideas, which later I want to try packaging as a product.
Continue inspiring other software developers to get on that boat fast. I don’t agree and believe in AGI narrative and hype. But I see the leverage AI gives you if you master it, not only in software development. And I believe in the slogan that is circulating the internet lately: Don’t be afraid of AI taking your job, be afraid of other people proficient in using AI to take over your job.
Spec-driven development is something I truly believe in but tbh I haven’t tried much yet. It has its comeback. Funnily enough, partly as a solution to the limited context of LLMs :-) But the side effect is you are precisely specifying from scratch what needs to be created. Check out Shotgun from a Polish team if you haven’t yet (https://app.shotgun.sh).
Another idea I want to explore is to change the coding experience from generating code to talking architecture. Do you remember my training on designing software architecture using 4 perspectives (https://c4model.com)? Imagine you talk to an agent about architecture, not code. You move around those 4 perspectives. You can even have specialized agents (security, scaling, MLOps, etc.) as your thinking partners to bounce ideas and co-design an ideal model of your system. Also, imagine you use voice for that instead of writing. Agents take care of the “boring” code generation underneath. In a way, Google releasing yesterday their Antigravity editor (worth checking out!) made a similar change to the coding experience. The code happens in the background but you talk to an agent. You still can look under the hood and edit something manually if you want. What do you think about that C4 coding agent idea?
I’m so happy you found your way back to your passion - coding. I always knew you are an engineer at heart ;-) I’m finding my way back now to my true passion: researching, finding clarity, sharing with others what I learned and how they can navigate new advancements in tech. Kind of what I did in your R&D team back in the day.
Btw. I love you are following my work and I love your comments. Sometimes it takes me some time to think what I want to respond.
After several months of using AI for me, personally and at work, some general conclusions emerge.
First, I still meet seniors that simply say "AI hallucinate", and avoid research AI more deeply. I belive that ALL seniors have to go through one or two quality AI trainings to get on speed and understanding NOW!!! Even if they hate or afraid the thing. Maybe this is an example where they should be forced to do it. Old times are gone.
Second, "I'm back" - after 50 - I'm back, and feeling strong again, writing code. It is as easy, effortless, rewarding for me now as when I was 20. I believe even one-man products are possible again these times. I'm sure, as I always was that "the less people working on any product, the better". I have been dreaming about times when multitude meetings, for the sake of spending time and not delivering much, will not be needed anymore and more time for deep work will come. It's possible again.
Third, I can rise quality higher that was possible in the past - e.g. add extra validations, provide proactive diagnostics or fitness functions, quality dashboards checking everything around. IMHO ensuring quality while using AI is not to focus on code itself but on multitude completely new possibilities. There was not much time for it in the past. Now within a few hours I can built a tool saving weeks of work. Also many tools will appear, grow, mature and disappear. Some may bring brilliant changes to the way we work e.g. see Spec-Driven-Development: https://martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html - still very weak but promising. Maybe...
Finally. How it applies to software agencies? I believe they should deliberately seek, experiment, choose, evaluate patterns (both in software and processes). This would allow to deliver faster while rising quality in the same time. Paradox. No more approach "more projects, more people". But I'm afraid that managers of such agencies may not be capable of doing so. I recommend trying to code again to all that were good engineers in the past and later on become managers and spent decades without coding. If one loved coding she/he might be very, very surprised what is possible now.
Starting from the end of your share, I agree that the best way for software agency management to get their head around AI is to get back to coding now. That is why I got back to coding and did the AI_devs course (heavy on coding). It is a new paradigm that is hard to grasp if you won’t try it. It requires a new mental model to operate it.
The leverage AI gives a software developer, or should I say product creator now, is tremendous. You can be a one-man army and create a product. And I plan on doing so! ;-) I’m writing a book on Augmented Mind now that creates a foundational model for my ideas, which later I want to try packaging as a product.
Continue inspiring other software developers to get on that boat fast. I don’t agree and believe in AGI narrative and hype. But I see the leverage AI gives you if you master it, not only in software development. And I believe in the slogan that is circulating the internet lately: Don’t be afraid of AI taking your job, be afraid of other people proficient in using AI to take over your job.
Spec-driven development is something I truly believe in but tbh I haven’t tried much yet. It has its comeback. Funnily enough, partly as a solution to the limited context of LLMs :-) But the side effect is you are precisely specifying from scratch what needs to be created. Check out Shotgun from a Polish team if you haven’t yet (https://app.shotgun.sh).
Another idea I want to explore is to change the coding experience from generating code to talking architecture. Do you remember my training on designing software architecture using 4 perspectives (https://c4model.com)? Imagine you talk to an agent about architecture, not code. You move around those 4 perspectives. You can even have specialized agents (security, scaling, MLOps, etc.) as your thinking partners to bounce ideas and co-design an ideal model of your system. Also, imagine you use voice for that instead of writing. Agents take care of the “boring” code generation underneath. In a way, Google releasing yesterday their Antigravity editor (worth checking out!) made a similar change to the coding experience. The code happens in the background but you talk to an agent. You still can look under the hood and edit something manually if you want. What do you think about that C4 coding agent idea?
I’m so happy you found your way back to your passion - coding. I always knew you are an engineer at heart ;-) I’m finding my way back now to my true passion: researching, finding clarity, sharing with others what I learned and how they can navigate new advancements in tech. Kind of what I did in your R&D team back in the day.
Btw. I love you are following my work and I love your comments. Sometimes it takes me some time to think what I want to respond.