coming back from paternity, i knew i needed to change my approach. prior, i was like 3/4 agentic — i wasn’t artisnally writing code and most code that i was reviewing was reviewing verification steps — where i’d still want to deeply understand the implementation details. also, i didn’t think the model mattered as much, so i’d lag behind a bit. i was on opus 4.6 for a long time, then stuck with 4.8 when 5.1 released and was still on 4.8 when i left for paternity. i took some space from software engineering while on paternity and i went all existential.

upon my return, i knew i needed to approach it differently, because i wasn’t enjoying work. this was the first time i wasn’t excited about work or software engineering. i was in this place of “what i enjoy has been taken away”. and while i do think part of that is true, i think most of it can be fixed by a perspective change. so i decided i’d go full agentic. and now with a baby in hand, i figured, let’s try doing it with dictation — having a conversation with a computer. and surprising my first week back was super enjoyable! i caught back up in only a couple hours — the model pulled full slack, email and doc context while i was out and helped me pick up the next highest priority task.

i will admit i believe the ability to and success of going agentic is hugely company, team and project dependent. the problem i faced before was that i really wanted to get deep in the “learning weeds” of this project by stepping through and hand-writing all of the code and hand-debugging everything. but that proves to be really hard when the rest of the team is full agentic. therefore, i believe it’s crucial that teams — and probably even organizations within a company — agree on an ai adoption strategy. i believe it’s crucial to have this conversation instead of assuming people are using it a certain way. since i joined the team in Jan 2026, i was struggling to keep up with my team and it was giving me severe imposter syndrome to the point where it was debilitating. i was moving so slow relative to the rest of the team that it felt like i wasnt adding the necessary value or doing so fast enough. they could “swoop” in and do anything that was taking me days and do it in minutes or hours.

where i was wrong was in my thinking that “full agentic” meant the agent was taking learning away from me. turns out there are actually a ton more learning opportunities at this new “architect layer”. and i think this is precisely why the more senior folks to me on my team were doing so well with agentic. it’s true, this isn’t first principle learning at this layer. you are now two layers removed — i.e., code > language. but most of us professionaly engineers weren’t in the 90th percentile programmers in our language of choice anyway. and now we all are with agents and this frees us up to become better at this “new” (not really) layer of software engineering.

i believe this new layer of software engineering will require us to spend most time focused on verification. dont worry about the code if you trust the verification. as humans, the agents are much better at understanding the code anyway, now let’s focus on ensuring we’re encoding our desires and expectations. and this is where that “architect-level thinking” comes in to play. i’ve worked with phenomenal principal+ engineers who — even pre-ai — were good at asking questions. and i think this is a skill that i’ve been weaker at at work and need to work on. i think this is precisely why folks who aren’t yet principal or higher are fearful because most of their time is spent working on parts of software engineering that the agents are good at. and it’s on these folks — including myself — to work on becoming better at this higher (architect) -level thinking.

and with dictation, im working on my speaking abilities. being able to respond with a deeply technical question on the spot rather than taking it “offline” and spending hours or days investigating.

again, i think your ai adoption is very dependent on your company (domain), organization and team (again domain) and there will be places in a company where working at those layers below the abstraction boundary is necessary. but i do believe that most of us in software engineering can move to this new layer of abstraction. and this is where i don’t think the choice is entirely yours. especially if you work for a for-profit company, in a domain or team where it makes sense to go agentic. we are paid to do a job, effectively and efficiently and i think if you believe you are a better programmer than the current agents, you are being delusional. and i think it’s more important than ever that you separate work and non-work hours. put your time in at work where you have less choice and use your free time to do whatever you want to do, including artisnal programming — i sure do because i love typing and moving slow. life is about process, not results. but you dont always get to choose where you apply your opinions, especially when you are getting paid.

on this topic, i found this post inspiring and hopeful that maybe our fields move past historical and current systems and rather than fields being “solved” or “over”, we work to re-define the fields with the new tools that we have available.