geopsist

I dont think you can get into flow easily - not talking from the dev pov but rather from growth. I find my self multi tasking like crazy and having to click yes yes yes in claude prompts :/ One way is to relax those permissions and have less interruptions. the other is to try and listen good music throughout.

I've been a founder since before any of this existed, same flavor of small team then and now. What I see now is that anyone can produce more but less due to their skills from scratch (slop?). A big company sees the headlines and a CEO reads that as "fewer people", or force use of LLMs and people seeing it as a preparation for AI processes taking over (and rightfully so). But frankly probably for smaller teams it reads "same people, wider surface". Imo bad CEOs existed before AI and will exist in the future...

coming from a different pov, I was (am) systems engineer who never became a proper coder, then founder, now mostly product/growrth. So AI and agents dint erode a mastery i had, they gave me one i never had. What you describe though makes sense for the skills I did have and I see a dependency as well. Things I used to work out myself, even slowly, I now reach for the agent first by reflex. The speed effect is indeed there. I get anxiety many times that I don't understand subject matters or I learn them slowly. Haven't found a good way other than keeping some of the writing or critical judgment calls to me. On a happy note i try not to push LLMs at all to my kids as I start reading such dependencies happening there.

considering my journey (systems engineer / not coder -> founder -> product growth) if you are in the "technical side of things" you probably do need a notion of coding. Maybe you don't need to be an expert in Node for instance but at least have some nuances when Claude Code gives you code snippets to run. A good example is having Codex tell me how to repair my car and me not having idea of what is it talking about (I don't, cars are a black box to me). So I believe technical affinity, and flexibility in grasping technical subjects and moving around them is the way to go. At least this is what i say to my kids but I am not Pythia the ancient Greek oracle in Delphi :P

Non-engineer here, doing growth at a small company, using these tools on side projects as well. The thing i find expensive is how to trust the output. I am not a coder so i cant rate the code to know if it's right. I end up running a second model to check the first one's output or testing. That takes time as well :)

imo helped in a matter of things professionally, as i managed to increase my output by reducing blockers (seeing this not from an eng pov but from a growth/marketing perspective). Have definitely increase anxiety that i am left behind and i don't know the best and latest.

windows user, not a dev, engineer doing growth. I use Claude, codex, Gemini in my projects. The failure modes I hit far more are 2: an operational one, ie shell confusion. Claude executes commands that do not work or fail in cmd / PowerShell, permissions issues etc. In code (and marketing / growth stuff) i find it many times that i will ask from claude to verify the code it proposes with codex and then come back with the results. Depending on the task i may ask for consensus from the three models - thats fun!

I'm on the GTM side at Trent, not engineering. I've been running our agents on my own side projects while the team builds this (a euroleague fantasy league app, and an affilaite site). Happy to talk about what that's like from a non-security-engineer angle.

You can always configure it per ^^^. And anyone can search by that flag as well so eitherway it should be a good practice.

Great post and insights!