The raise of the "Super PM"
How AI would help us to speed up delivery and "flatten"? org structures.
For those who prefer the audio format, this audio has been created using NotebookLM.
Introduction
I've always been a fan of collaboration, which you can see reflected in the experiences I typically share in my articles.
or
or
Even when conducting interviews, one of the things I value most is the connection and degree of collaboration you have with your team members. As I highlighted in my previous article, such collaboration can yield exponential results.
Furthermore, when hiring managers ask about my biggest pet peeves in interviews, I always respond the same way: I don't like to pretend that I'm a "super PM" because I'm not. I'm fully aware of the benefits of having people focus on specific parts of the product and creating a culture of feedback where we can work together to discover, validate, test, and deliver products.
Why am I writing about this?
In my current role, I'm facing a unique situation that might become the new normal: both the product analyst and product designer are gone.
I was initially freaking out, thinking about the worst-case scenarios — cost of delay, time to value, and subconsciously feeling pressure as a new PM to demonstrate my value. On the other side, I have two superpowers that overshadow my fear: curiosity and a willingness to build.
The question I asked myself was: Can we do great work even without product analytics or design partners? Could AI compensate for the missing team members? Let's give it a try.
How did I start?
There are three scenarios or use cases where I've used AI, along with the tools and their economic considerations:
1. Analytics Assistant
Although I studied SQL in university and spent eight years as a programmer, I'm not particularly skilled at it. So, I built a custom GPT where I loaded all our database structures and our use cases, then started generating queries for daily dashboards and deeper analysis.
An interesting discovery while using AI is how different models excel at various tasks. When I started, I would grab the output from ChatGPT and ask Claude to compare approaches. I quickly realized that Claude was much better at generating code.
Today, I'm proud that some of our A/B testing in production and the KPI dashboards I review daily have contributions from queries I've built.
Sparring Partner
This is the most general and probably most widespread use case. When facing an analysis, I ensure I provide as much context as possible to the AI agent. From understanding factors to analyze when rebuilding our ranking algorithm to generating ideas for the upcoming quarter, the goal is to build a more solid and robust strategy.
In the AI world, context is king. I predominantly use Claude projects to create the initial context, continuously adding documents about the company and specific initiatives. Some may be skeptical about information potentially being used to train models, but with Claude, that's not a concern — they don't use chats for model training.
Dedicated Developer / Designer
This has been the most disruptive use case for me, given my previous interactions were mostly with human team members. I want to be clear: I'm not replacing conversations with engineers or designers. Instead, I'm using AI to get started at a much faster pace, allowing more refined conversations with human experts.
I began by addressing how users could narrow down results more intuitively. I analyzed usage data (using my analyst assistant), conducted benchmarking, and identified main barriers to filter usage (using my sparring partner).
After brainstorming ideas, I used Vercel to quickly generate four prototypes matching our website's style. Previously, this would have taken two days of work in Figma or Miro. Now, I had four concepts to discuss with my designer on the same day.
A similar approach worked for a specific feature in our ranking algorithm. I needed a Python notebook to read a CSV file and calculate a score using a complex formula. In a typical scenario, I would have involved an engineer, explained the context, verified results, and integrated calculations. Using Cursor AI (you could have used Replit also), I iterated the formula for different scenarios in less than three hours. Now, engineers have a base code they can quickly integrate and evaluate.
How much is it going to cost me?
Assuming you have your preferred options, here are the yearly subscription costs (without applied discounts):
Vercel: $20/month → $240/year
ChatGPT Plus: Covered by my company → $240/year
Claude Pro: €21/month → €250/year
Cursor AI Pro: $20/month → $240/year
Replit Core: $15/month → $180/year
Considering you might use a maximum of three tools (with some tools improving at specific tasks), I'd recommend a maximum of two. You'd be spending around $500 a year. For me, it's an investment I'd make without hesitation.
Conclusion
I was incredibly inspired by talks from Tal Raviv and
discussing how they've built remarkable things with even busier schedules. A huge shoutout to them!I believe AI is bringing three transformative changes to work:
The Era of Builders: Entry barriers are lower than ever. Responses like "I can squeeze some time next week" or "Let me investigate and get back to you" are evolving into "I'll try it myself, get something going, and return with a precise solution."
Rediscovering Work's Fun: For me, the ability to engage with different disciplines has a multiplying effect. I'm learning things I wouldn't have if someone else were doing the tasks, and I can move much faster while using experts' time more wisely.
Significant Improvement for Small Investment: Context is crucial. Investing time upfront to build a solid context pays off. The initial effort is quickly outweighed by subsequent benefits.
The Changing Paradigm of Professional Development
This might be a bold statement, but the professional development landscape is shifting. When my manager asked about courses for 2025, I suggested investing in a robust AI work toolkit rather than traditional training.
This doesn't mean abandoning specialized courses entirely — I'm still interested in specific topics like personalization or search. But how long until such knowledge becomes universally accessible? In fact, I've already used AI to create a comprehensive ElasticSearch course with modules and exercises.
Here are two resources to start with that are a must:
Thanks for sharing Fede, really interesting article, and also thanks for the links