Five Product Lessons I Learned from Failing at Developing an AI Agile Coach
I'm sharing 5 critical mistakes I did developing a product so you don't have to make the same mistakes! Be kind and let me know what you think about these lessons I learned
⭐️⭐️⭐️⭐️ This post is sponsored by The Global Agile Summit, where you can find the most advanced ideas on Software, Product Development and Agile! Join The Global Agile Summit, the most forward looking Agile event of 2025
Building a product that nobody knew how to use is an ego-crushing, yet profoundly educational, experience.
I say this not just from theory but from the trenches, where I watched my AI Agile Coach go from an ambitious idea to a product that left users confused, uninterested, or asking it to summarize books instead of helping be better coaches.
This post is about what I learned along the way, so you don’t have to make the same mistakes. If you’re working on AI-driven products or any innovation that solves a problem people don’t yet recognize, this is for you.
1. You Can’t Teach People a Problem They Don’t Think They Have
I was so sure my AI Agile Coach was solving a real issue: the lack of on-demand coaching support in Agile teams. Turns out, while I understood the problem in depth, my potential users… did not.
No matter how much I tried, I couldn’t teach them the problem they were facing if they didn’t already recognize it themselves. People don’t go looking for solutions to problems they don’t believe they have. They seek solutions to pains they feel, not pains they should theoretically feel.
Lesson: I knew the problem "better than they did", a classic mistake. How do we avoid this problem? See below for some tips.
2. People Want AI to Do What They Already Do—Just Faster
Once I started showing the product, a pattern emerged. People didn’t want an AI coach that engaged in deep Agile coaching conversations. They wanted it to be a glorified Google search:
“Tell me the best 5 books on product management.”
“Summarize these 5 books on Scrum.”
Great questions, sure, but totally missing the point of a coaching tool. The expectation wasn’t “help me explore ideas,” it was “give me quick answers.” And my AI didn’t answer these questions better than a search engine, making it effectively useless for them.
Lesson: AI adoption starts where people already are, not where you want them to be. Want to know what kind of questions are best to ask from an AI coach? Read below for some tips.
3. A Product Is Not Just a Product—It’s the Whole Ecosystem
We initially thought we could just sell the AI Agile Coach as a standalone tool. Wrong. The reality? It required a lot of training and change management for organizations to get real value from it. Individuals were not ready to adopt this tool directly, it needed to be a supported process in an organization.
So, we adjusted our go-to-market approach. Instead of targeting end-users directly, we bundled the product with training and change management services, shifting our target customers to mid-sized software organizations.
This change made the sales conversations easier but significantly increased the effort required to implement and support the product.
Lesson: If your product requires new behaviors, be prepared to offer training and hand-holding—or rethink your entire GTM strategy. We came to this conclusion from running product-market fit experiments. Want to know how we setup these experiments? Read below for some tips
4. The Value Isn’t in the Software. It’s in the Relationships.
The value isn't in the software or the prompt (in fact, you can get our full prompt if you comment below, I'll send that prompt to you via DM), it's in the relationship we build with customers.
People were excited about the AI Agile Coach when they saw it. But what actually led to sales? The relationships we were able to build in those early conversations.
People got interested in having a conversation when we demoed the product, but it was the relationships we were able to build from those conversations that led to the best opportunities
Lesson: Technology doesn’t sell itself. Relationships do. Want to know how we structured our conversations with potential users? Read below for some tips on this.
5. AI Adoption Will Be Slow Because We Don’t Know How to Have Conversations, Even if We Have a Goal in Mind
A common assumption about AI is that it will magically make us more productive. But here’s the kicker: most people don’t even know how to articulate their problems effectively in a conversation.
I saw this firsthand. Coaching is about structured dialogue—problem exploration, context analysis, experimentation, and follow-up. But people just wanted quick, one-shot answers. They weren’t engaging in the process; they were hoping for shortcuts.
Lesson: AI won’t replace real problem-solving. If anything, it reveals how much we still need to work on the quality of our conversations. And when it comes to humans, that's the only thing that will make an impact: better conversations. I'll explain what, and why below, and include some tips for better conversations which help you reach consensual solutions, and implement those solutions faster.
Tips to Avoid the Pitfalls I Faced
Tip 1: Describe the Problem in Your Customer’s Own Words
If you find yourself saying, “They just don’t get it,” stop. That’s your cue that you don’t get it.
A common problem I face when coaching product teams is the "we know better" syndrome. It starts off innocently enough, with the idea that "because we've researched this problem, and several customers for so long, we must know more than every individual customer." Not so fast Fittipaldi!
In practice though, this is rarely the case. Hear me out.
For every single customer or user out there, their problem is unique to them. Sure, you can abstract and categorize problems so that they seem generic and similar across many customers, but that's an assumption, not reality.
You see, from each person's perspective, the problems they face are theirs, not generic problems.
What this means for product teams is that we need to be super focused on each customer interaction to understand their problem exactly in their own words. And we need to be able to describe the solution we propose, to them, in their own words.
In my case, the AI Agile Coach was supposed to solve a problem I classified as "we don't always have access to colleagues, but the AI is always available, does not judge us, and has helpful ideas we can turn into coaching experiments". The problem I was solving was: partnership in coaching assignments (not feeling alone), and lack of creativity because of fear of being judged.
However, the problem my customers were trying to solve could be better described as "I need a solution for this problem, without having to debate, or describe the problem in detail". As I write these phrases it becomes, once again, obvious to me why AI will not be able to help us on the day to day coaching assignments we have: we want to use AI as a shortcut, not as a colleague, or as a coach that helps us with the coaching work.
The key tip to avoid this problem is: describe your user/customer's problem in their own words. Not in yours.
Tip 2: Helping Users Ask Better Questions
Many people are flocking to AI services like Claude, Gemini, etc. to get some of their work done faster.
This is understandable. But when it comes to learning how to solve a problem (intellectual labor), "faster" does not help us. I mean think about it, we often let ideas simmer in our minds for days, or even weeks, before we take action. When we interact with an AI chat service, we often expect (and our early users did) an immediate answer. And that's fine for some cases, but not for coaching assignments.
We explicitly designed the AI Agile Coach around the coaching process (problem - context - experiments - follow-up), but it became clear very quickly that what people wanted was a question - immediate answer process. In other words, they wanted to describe a problem with minimal text (no context, lack of detailed information, no agreement on problem definition), and then get an immediate and single answer.
This is understandable, but not appropriate for a coaching problem.
As coaches, we learn very quickly that "pushing" a solution will often backfire, and that we need to engage in a process of helping the client find a solution to try, and then do a follow-up later on to see how it worked, and make adjustments.
But what I witnessed was that coaches using an AI Agile Coach were not able to engage with the coaching process that the AI was trying to help them implement.
This led me to understand that we needed to help people learn how to use an AI Agile Coach through training, and that we needed to start with helping users ask better questions.
If you are ready to practice the coaching process with the help of AI, here is a good prompt you can use:
[Define Coaching Experiments] Coaching experiments are practical activities, reflection exercises, or small behavioral changes that I can try between sessions to gain insights or make progress on my problem. They should be specific, measurable, achievable, relevant, and time-bound.
[Define the Coaching approach that works for you] Use a solution-focused, strengths-based coaching approach that emphasizes my capabilities and potential solutions rather than dwelling on problems. Ask powerful questions that expand my thinking rather than providing direct advice.
[Action prompt] Act as a professional coach using a solution-focused, strengths-based approach. Help me reflect on a problem I'm facing by following this structure:
1. Begin by actively listening to my problem and mirroring it back in your own words to confirm your understanding. Ask a series of open-ended clarifying questions to deeply understand my situation, including the context, emotions, underlying beliefs, and goals involved
2. Once I confirm your understanding is accurate, ask 2-3 powerful open-ended questions that challenge my assumptions, explore new perspectives, or help me connect to my values.
3. Help me identify strengths and resources I already have that could help with this situation.
4. Only after completing steps 1-3, offer me 3 options for coaching experiments. Each experiment should be clearly explained with its purpose and expected impact.
5. Help me select and refine one experiment that I'm willing to commit to.
Always maintain a balance between being supportive and appropriately challenging.
Focus on helping me discover my own solutions rather than being prescriptiveThe goal of this prompt is to effectively practice the coaching process as we interact with the AI Agile Coach.
In other words, the AI Agile Coach is not just helping us address the coaching challenge, it is helping us to practice our own coaching skills.
Here's the catch: this is not a shortcut, is a structured, goal oriented process so that we progress in our assignment (help our clients), and keep learning how to be a better coach.
Tip 3: How to Setup Product-Market Fit Experiments
The first question many ask when developing a new product is: "can we find customers that benefit from this solution". However, a much better question to ask before is: "What problems do my potential customers already know they have?"
And this should drive our product-market fit early activities: validating that we understand the problem our customers face.
Unfortunately, I skipped this process, and started working on the solution first. That led to developing something that people didn't know how to use, even if they liked the "idea of the solution".
In other words, I went straight to customer-creation (from Customer Development process), instead of spending some time in the Customer Discovery, and Customer Validation steps. If you want to know more about these different steps, be sure to read "Four Steps To Epiphany" by Steve Blank.
Let me shortly explain why skipping Customer Discovery and Customer Discovery was a problem for my business:
By focusing on building a solution, I lost the opportunity to learn how the customers see themselves, and define their problem.
By testing the solution thorugh demos, instead of doing some Customer Discovery, I failed to grasp what part of the "problem space" that the customer was already aware of.
So, how can we avoid these problems? Great question. We avoid these problems by setting up product-market fit experiments, but starting with Customer Discovery, not focusing immediately on the solution as we imagined it.
Here's a few ideas of what I could have done differently to discover and validate customers before going into testing the solution directly:
Problem-Focused Interviews (which I had to do later): Instead of jumping straight to explaining my solution, I could have conducted structured interviews focused solely on understanding customers' existing workflows and pain points.
Day-in-the-Life Observation Sessions: I could have arranged to observe potential customers as they work through their day, paying special attention to the moments of friction, workarounds, and frustrations they experience.
Tip 4: Structure Your Customer Conversations for Validation
Many years ago, I was exposed to the idea of customer validation. In short, customer validation is about testing your business hypotheses through direct interaction with potential customers before fully developing your product or service.
Before you build anything, ask: What problems do my potential customers already know they have?
I skipped this step and jumped straight to product development, which led to a misalignment between what I built and what people needed. Luckily I could got back a step and ask the right questions from more customers!
In order to validate my assumptions about the customer, their goals, their expectations, and the alternative solutions they had already tried, we structured our conversations around 3 key questions:
What questions would they ask from an AI Agile Coach
How would they perceive the interaction with the AI Agile Coach
After interacting with the AI Agile Coach, what would they would do next? (use the alternative solution as a grounding factor)
When preparing for these Customer Validation conversations, here are some simple rules to follow:
Don't ask customers what they "want" from your service. They don't know. It's our job as product developers, to know what customers want. Instead, focus on asking them how they solve their problem now, perhaps using other alternative solutions. In our case, people said they would call on a friend or colleauge to debate the coaching assignment they had
Ask the customer to voice their thoughts as they interact with your product, and look for emotional responses (surprise, anger, frustrations, smiles). Emotional responses are "where value is hiding", as I usually put it when coaching product teams.
When you ask the customer a question try to trigger memory, not imagination. For example in that third set of questions about what custoemrs would do next after talking to the AI Agile Coach, we would ground their answers by asking: "when you had this type of conversations with a colleauge or a friend, what were the usual actions you took". When we appeal to customers imagination, we get wishlists that they may or may not act on. But when we ask them to recall the past, we get a list of things they actually did and decisions they actually made.
A great book to read on this is "The Mom Test" by Rob Fitzpatrick, where the author advocates for asking questions about customers' actual behaviors, problems, and past experiences rather than their opinions about your solution.
Tip 5: Conversations Are the Key to Adoption
We learn very quickly in school that there's a "single right solution" for everything. This, it turns out, is a big obstacle to having meaningful and impactful conversations at work and at home.
Especially in a coaching assignment, the goal should always be to help the client's move towards a context where they can excel. The client could be a single person, a team, or even an organization. And that makes the art of conversation even harder.
Many of us have, through practice learned to have a conversation with a single person, or a limited group of people that all share the same concrete medium (video call, chat, physical space), but not many of us have learned how conversations happen over time (instead of a single moment), or with a large group of people that share much more fuzzy medium (an organization or a team), instead of a concrete medium.
In short, we work as if:
We believe there's a "right solution" for any challenge or problem
We believe conversations are limited in time and medium
But when we are helping a person, a team, or an organization improve over time, we must learn that:
There are several possible solutions that help move us forward towards a goal, but we need to agree on the goal first, instead of agreeing on the problem first
Conversations last through time, instead of being that single interaction in a meeting, call or 1-on-1
So, we must prepare to keep conversations alive and meaningful over time, and we need to be able to recall what happened, where we were, and what we explored the last time we interacted with someone or a group of people. Luckily an AI coach is a great help for that.
Here's a process you can apply right now, to have better conversations, and keep the memory of those conversations alive over time:
Prepare for important interactions with the help of an AI Coach (see the prompt above)
Once the interaction ended, add the notes from that interaction to the conversation with the AI coach, and ask for feedback
As you and the people involved work through the coaching experiments (remember, we learn what works through action), come back and reflect on the findings with the AI coach
Use the AI coach to prepare your next interaction with that person or group, specifically recalling what were the previous agreements, and what you need to learn about in the next interaction (results of a coaching experiment? Decisions that were pending? etc.)
Final Thoughts: AI Won’t Replace Coaching—But It Can Enhance It
The biggest lesson I learned? Coaching is way harder to formalize than I thought.
Would I recommend people to use an Agile AI Coach? Absolutely! But first, I'd recommend everyone to get a human coach! Because if you don’t know how to benefit from working with a real coach, you won’t get much out of an AI one either.
Building this product was humbling. But if sharing these lessons helps even one founder avoid the same mistakes, it was worth it.
Now, over to you—have you faced similar challenges building products? Let’s chat in the comments.
Exploring the future of software development
At the Global Agile Summit, we’re bringing together real-world Agile success stories—including practitioners who have successfully organized their work with business impact as the main goal.
We have several CEO's, Product Leaders and more sharing their journey on stage. High stakes success stories that frame how Agile is having huge business impact today.
🔗 Register here for the Global Agile Summit
The Agile mindset has the key to get back to our original mission: to deliver value as early and often as possible.
Our software industry is too important for us to stay stuck in the “let’s deliver more products and features” doom loop. We need focus on building products that have a positive impact on the business, and the wider society. The Global Agile Summit brings you those stories, from people you can talk to!