Hey IT folks! I'm Alfiya, and I'm growing in IT project management.
— How many devs are on your team? — a colleague from another department asks me.
— Zero, — I reply with a grin, chewing on my sandwich.
In the age of LLMs, this is reality. A project manager can totally get by with just something like Cursor or Claude Code, managing only the LLM's work. But how do you make it actually more efficient than a whole dev team? That’s what I’ve written all these words about below...
ABOUT ME (SKIP IF YOU'RE NOT INTERESTED)
Once upon a time, I ran my own project portfolio in child neuropsychology. I did it on intuition and, I think, quite successfully. Over that time, I helped over 100 kids with developmental challenges, created my own correctional program, published a teaching guide, organized both offline and online training for specialists, set up production and sales of educational equipment — and a bunch of other stuff.
Then some big personal changes happened, and suddenly my motivation for that work just vanished — like someone flipped a switch. So I started looking toward IT...
WATERFALL OR AGILE?
In January 2026, I joined an IT company as a project management trainee, and now I'm managing a sandbox project. Not long ago, our CEO launched a new department where anyone could use LLMs to generate ideas for new services and then build and launch them using the same AI. The idea is that an LLM drastically cuts the time and resources needed for an IT project, shrinking the dev team down to just a manager + LLM. So in the near future, we won't have huge bloated teams; development speed will multiply, and ideas will go from concept to reality in days. And I totally agree! It really does seem like soon one person — say, a product or project manager with Cursor or Claude Code, maybe with a couple of devs on standby — will build services that can compete with Big Tech.
Honestly, before this department existed, I didn't know much about neural networks and didn't use them often — just as a fancy search engine. But I love learning and diving into uncharted waters. So I held my breath and jumped headfirst into the ocean of LLMs. It was tough at first, as always when you're new to something. But after a month of working side‑by‑side with Claude Code, AI has become a part of my daily life. Any question, problem, or idea now gets run through it: need to draft some management document? Evaluate an idea for monetization? What to do if your tennis coach whacks you in the stomach with a ball? Or which vacuum filter to buy? I solve it all with AI, and I can feel deep down how much time, energy, and resources I'm saving!
But that's not what this article is about. While building yet another service with my dear Claudie (Claude Code), I realized something: before, I'd write a short prompt, get some rough product, then endlessly tweak it — or even restart from scratch because I hadn't thought things through. Now I use a different strategy. First, I write a detailed, careful prompt to generate all the project documentation: specs, functional requirements, descriptions, test cases, and so on. I think through every little detail from the start, and only then do I let Claudie build the product based on that documentation. The goal is to get the most polished, mature product from the LLM on the first try, minimising rework and fixes. Yeah, perfect on the first iteration still doesn't happen (yet), but the more detailed and precise your documentation, the better the output — which means less time and effort later.
But now, if we have to write all that detailed documentation upfront and only then start coding, what does that mean? Is IT project management with LLMs no longer Agile but Waterfall? The whole point of Agile was that when you're in a situation of uncertainty and tight deadlines, you just need to start somewhere, show an MVP to the customer as soon as possible, and worry about docs and contracts later. But when your main team member is an LLM that can build your idea in hours, you don't need to rush. You have loads of time for planning. So it turns out it's much more efficient to work with Waterfall elements: you create super detailed docs, specs, functional requirements — again with the help of an LLM — and only then launch the development. In that case, documentation becomes the primary focus. If it's strategically correct, complete, and accurate, the LLM will crank out the project in a few hours, and you can quickly show the customer an MVP. That's faster than any traditional team, so saving time in project management is no longer the top priority.
Of course, calling it pure Waterfall would be an oversimplification. We still iterate — but at the level of prompts and documentation, not code. The key difference is that we no longer rush implementation because it now takes hours instead of weeks. So the classic Agile principles (quick start, constant feedback) give way to "slow design, fast build." This isn't Waterfall in the old sense, but rather "documentation‑driven development."
Now the focus has to shift to increasing quality and reducing project cost. And that's exactly where a well‑crafted prompt helps. It becomes the cornerstone of AI‑driven IT development. Do the math yourself. Every extra iteration with an LLM burns tokens — which means real money from the project budget. Plus my own time as a PM, which could've gone into strategy instead of endless reworks. A good prompt cuts the number of iterations from five down to one or two — saving on API costs and work hours, directly hitting the unit economics of the project. And if we release the MVP a week earlier, that's extra market capture and the kind of "fast money" that business customers love. So now my main KPI isn't the number of tasks closed, but the cost of one successful iteration.
Right now, I'm aiming to write a prompt that gets me the best possible answer from AI in a single iteration. Then I can say I got maximum quality in minimum time, spending the minimum tokens (project cost). And isn't that the ultimate goal of any PM?
WRAPPING UP
So, in the age of AI, the PM transforms from a task dispatcher into a prompt architect. The main skill is no longer team management, but turning uncertainty into structured documentation that becomes the single source of truth for the model. Agile isn't dying — it's mutating into a "prompt‑oriented approach," where iterations shift from code to prompts. Maybe that's the next stage in the evolution of project management. What do you think?
I'm looking for a job as an IT project manager and I'm open to any form of collaboration — as long as it's exciting and makes a difference.
Let me know if you want any tweaks — shorter, more formal, or even more casual. Good luck with the publication!
Top comments (0)