I’ve been shipping data pipelines and slinging code for a long time, so trust me when I say: tools like Cline, Copilot, Cursor aren’t magic, but they sure feel like rocket fuel. After a long time far from code (in managerial positions) I approached the keyboard again in quite a decent way. With the due respect, I resembles what I heard in this interview to the CEO on Spotify, Daniel Ek.
Here’s why I’m all in on using them—even if they sometimes hallucinate like a sleep-deprived intern. Here’s the thing: leveraging LLMs in development or model engineering feels like standing on the shoulders of giants. You get the accumulated knowledge of thousands of Stack Overflow answers, countless blog posts, and terabytes of open-source code, all at your fingertips. But you still call the shots. You still vet every line. You still steer the ship. However there are quite a bunch of haters out there, but not him.
Remember
LLM is your co-pilot, not the captain. And let me tell you, once you’ve flown this way, you won’t go back to paddling a canoe when you could be piloting a jet.
1. They’re not perfect—welcome to accelerated iteration
Yes, LLMs make mistakes. They’ll invent APIs, miss edge cases, and sprinkle in half-baked ideas. But so what? In my world, perfection is the enemy of progress. A half-working CRUD endpoint generated in seconds beats hours of boilerplate any day. I can prompt an LLM to scaffold my ingestion job, tweak the prompts for data validation, and have a prototype streaming pipeline running before my coffee’s cold. Even if I spend ten minutes fixing its slip-ups, I still save hours of manual typing and context-googling.
2. Your team’s productivity doesn’t hinge on who’s on call
Remember the days when a random question—“How do I unpack a nested JSON in Spark?”—meant pinging three senior engineers, waiting for their vacuum-packed schedules to free up, then chasing down credentials to reproduce their setup? LLMs answer those menial queries instantly. Suddenly, my juniors aren’t stuck waiting for my approval to write basic ETL transforms. They spin up tests, ask the model for sample code, and iterate on their own. I get to focus on the gnarly data-model debates, not on whether a comma belongs inside or outside the JSON path.
3. Small teams, big impact
Early in a project, you don’t need a full platoon of data engineers and ML specialists. Give me two people and an LLM, and we’ll bootstrap a crawler, ingestion pipeline, and a dashboard in hours. No need for endless architecture reviews on JIRA tickets. I’ve seen solo devs ship MVPs on Fly.io with just a terminal, a Dockerfile, and a well-trained prompt. The model handles the scaffolding; we handle the domain logic. It’s like having an extra pair of hands that never sleeps, never complains, and never asks for a raise.
Poor man's tip
Treat your coding agent assistant as your dearest team colleague
4. They do the boring sh.t
Let’s be real: most code is boring. Writing linters, crafting CI configs, refactoring ancient helper functions—yeah, that’s soul-sucking. I tell my LLM agent, “Hey, refactor all these tests to use pytest fixtures,” and it churns out a PR I can review over espresso. I’m not shirking real work—I’m outsourcing the grunt work so I can spend my brain cycles on feature design and performance tuning. If you think that’s cheating, you’ve never stared at a 500-line SQL window wondering why your LEFT JOIN returns nulls.
That whole “vibe coding” thing? … no way
YouTube gurus selling “vibe coding with LLMs” make it sound like anyone can sit back, sip oat-milk lattes, and watch code appear by telepathy. Newsflash: you need skill. You need experience to write prompts that yield useful output, to understand when the model is bluffing, to know how to integrate generated snippets into your existing codebase without breaking everything. Without that expertise, you’ll spend more time undoing the model’s mistakes than writing code yourself. LLMs augment your skill—they don’t replace it.
So yeah, LLMs aren’t perfect. They’ll hallucinate, they’ll spit out garbage, and they’ll never replace the intuition you’ve built over years of debugging midnight outages. But they will accelerate your cycles, turbocharge your team’s productivity, let you run leaner crews, and banish the most tedious chores to automation. In the end, that’s how real pros get shit done—by standing tall on the shoulders of those giant language models, and crushing the work that matters most.