I end up asking the same question when experimenting with tools like Cursor. When it can one-shot a small feature, it works like magic. When it struggles, and the context gets poisoned and I have to roll back commits and retry part of the way through something, it hits a point where it was probably easier for me to just write it. Or maybe template it and have it finish it. Or vice versa. I guess the point being that best practices have yet to truly be established, but totally hands-off uses have not worked well for me so far.

Why commit halfway through implementing something with Cursor? Can you not wait until it’s created a feature or task that has been validated and tests written for it?

Why wait until everything is finalized before committing? Git is distributed/local, so while one philosophy is to interact with it as little as possible, the other one is to commit early and commit often, and easily be able to rollback to a previous (working) state, with the caveat that you clean-up history before firing off a PR.

Why not create a branch and rollback only what needs to be rolled back? Branches are O(1) with git, right?

OP was insinuating that rolling back commits is a pain point.

Well, same statement applies. Rolling back commits is also O(1) and just as easy. And if you branch to start with it's not even a "rollback" through the commit history, it's just a branch switch. Feel like OP has never used git before or something.

Which seems like a tooling issue, imo. In Aider, it's just /undo.

Easily 5-10x or even more in certain special cases (when it'd take me a lot of upfront effort to get context on some problem domain). And it can do all the "P2"s that I'd realistically never get to. There was a day where I landed 7 small-to-medium-size pull requests before lunch.

There are also cases where it fails to do what I wanted, and then I just stop trying after a few iterations. But I've learned what to expect it to do well in and I am mostly calibrated now.

The biggest difference is that I can have agents working on 3-4 parallel tasks at any given point.

This has been my experience too. Certain tickets that would’ve taken me hours (and in one case, days), I’ve been able to finish in minutes.

Other tasks take maybe the same amount of time.

But just autocomplete saves micro-effort all day long.

For me, it’s not that the actual coding is faster. It’s that you can do other things at the same time.

If I’m writing an integration, I can be researching the docs while the agent is coding something up. Worst case, I throw all of the agents work away while now having done research. Best case, it gets a good enough implementation that I can run with.

Totally. I feel like it’s akin to jamming with someone. We both go down our own paths for a bit, then I have a next step for it, and I can review what it last came up with and iterate while it does more of its own thing. Rinse, repeat. This is more fun and less energy consuming than “do it all yourself”, which certainly means a lot.

This way works for me. Any time I tried to treat it as a colleague that I can just assign tasks to, it’s failed miserably.

> Worst case, I throw all of the agents work away while now having done research

The worst case is you take the agent's work without really understanding it, continue doing it indefinitely and at some point get a buggy repo you have no idea how to handle - at the exact same moment some critical issue pops up and your agent has no clue how to help you anymore.

I don't think GP said they couldn't do their job, but you instantly jumped to incompetence. That seems little uncharitable to me.

Have no idea what GP can or cannot do and wasn't talking about that. I'm saying what the worst case that can happen when people work with agents, and it can happen to anyone who isn't carefully verifying and testing the agent's work.

At the current capabilities of most LLMs + my personal tolerance for slop, the most productive workflow seems to be: spin up multiple agents in the background to work on small scope, straightforward tasks while I work on something bigger that requires more exploration, requirements gathering, or just plain more complex/broad changes to the code. Review the output of the agents or unstick them when there is downtime.

IMO just keeping an IDE window open and babysitting an agent while it works is less productive than just writing the code mostly yourself with AI assistance in the form of autocomplete and maybe highly targeted oneshots using manual context provided "Edit" mode or inline prompting.

My company is dragging their feet on AI governance and let the OpenAI key I was using expire, and what I noticed was that my output of small QoL PRs and bugfixes dropped drastically because my attention remains focused on higher impact work.