On the Fly AI (field report)
I do monthly Snowflake retainer work for a client — weekly meeting cadence, steady progress, normal stuff.
This week they brought in a Python dev who was having trouble connecting to an API for a data source. I’m Python-capable, but it’s not my main lane.
Good news: this client is very supportive of using AI.
So I shared my screen and used Warp.dev (started life as a new-fangled terminal, and is now basically a Claude Code competitor living in your shell).
In about 30 minutes, we worked through the connection issue, got the code running, and committed everything to git for the Python dev to take it from there.
Key learnings
1) I currently prefer command-line AI agent tools over the IDE assistants.
Not 100% sure why yet. But CLI agents feel more direct. I’m not coding, I’m directing the coding. I do use an IDE to view the created code.
2) Feed the AI the docs.
In this case the API docs required login, so the agent couldn’t just fetch a URL and magically know what I was looking at.
I copy/pasted the relevant page into a local file and pointed the agent at it. Worked immediately.
3) Watching the agent loop is the whole point.
Write code → run code → see failure → adjust → run again.
That feedback loop is where the value lives. It’s not “one prompt and done.” It’s iterative debugging at machine speed. Keep in mind, the warp ai agent did the looping itself. I had very little need to interact.
4) Token burn is real.
I burned through a lot of Warp.dev tokens in that half hour.
I’m keeping the $20/mo plan in reserve as my “on-call sysadmin” and not treating it like unlimited agentic coding time.
What I like about Warp.dev is simple: it can do anything I can do in a terminal.
It earns the $20/mo every time I can say:
“My Windows/Mac/Linux box is behaving like XYZ… fix it.”
⸻
I’ve also been playing with OpenCode.ai connected to my $20/mo ChatGPT Plus subscription. So far, that combo looks really strong.
I continue to play and explore and learn. But I also simply use AI all the time now in numerous ways. It’s definitely a productivity AND quality enhancer - for me the way I use it.
