Alright folks, four pieces of advice on agentic development to get past your resistance and start experimenting.
”AI can’t do this thing, I’m scared it’ll mess up”
Right, so it doesn’t have to do it. You do it. There are things the agent does and things you do. How do I decide what the agent does and what I do? These tools are built precisely to let you define that inside your own environment, whether it’s Claude Code, OpenCode, Codex, or whatever. They let you define where to step in, when, what the agent does, what you do. So don’t see it in black and white. It’s something you can decide and customize.
”I have to review everything it does and it’s a drag”
Also not black and white. You have to do a risk assessment. If you’re writing code, if you’re building software that carries a big responsibility, both on your part and on the part of what the software itself does, then it has to be reviewed line by line. You don’t give the agent permission to write to the cart from anywhere. Every edit the agent makes has to be looked at by you first.
For experiments, though, things you can completely vibe code because they’re throwaway or you’re just playing around and you want some creative satisfaction, in that case you can let it run fully on automatic. And this too is part of your environment’s configuration, which you can tune from project to project. It’s not black and white.
”It takes away the work”
This is a big load of nonsense. It doesn’t take away the work. If anything, it takes away tasks. Which tasks? It takes away the grunt work, the lower-level stuff, and you become a bit more of an architect, someone who does monitoring and writes the specs for what needs to be done, how it needs to be done, and how it needs to be tested. So again, the black-and-white framing gets you nowhere. I think in terms of tasks. Some tasks you do, some the agent does, and you get to decide who does what. It doesn’t take away your work. It changes it, and it changes it profoundly.
”I like programming, why should I hand it to an agent?”
I’ve always loved what I do. It took me a long time to earn these skills, and that love is absolutely noble and worthy of respect. But here too you’re seeing it in black and white. The agent doesn’t have to program everything in your place, as I already said. In particular, if you’re one of those people who genuinely loves the act of coding, one hundred percent, then decide which parts you write by hand because you enjoy it, and God bless you, and which parts the agent handles because you don’t feel like doing them. For example: you like writing tests, so write only the tests. You like writing certain endpoints but not others, so write only those endpoints.
Look at all the shades in between, all the different ways to organize your work. This whole block of stuff is not agent versus no agent, vibe coding versus no vibe coding. This is a new methodology for creating software. And I don’t say “development” because development is something you do if you want to, and if you don’t want to, the agent does it. Either way there’s a continuous risk assessment of what you’re doing, and there’s the customization of this agentic environment.
