Feeling Behind on AI? Two Prompts Every Product Leader Should Run
Aug 09, 2026Viktoria Otstavnaya isn't new to AI; she was putting AI into production in 2019. She'd seen machine learning applications in the wild. So she didn't anticipate how quickly AI would revolutionize software after the launch of ChatGPT in November 2022.
"When the whole AI hype started, I did not even think that this will be something huge. Knowing what happened in 2019 and 2020, I thought another couple of companies will [add some machine learning features], but this is where it will stop."
Here is the part I can't stop thinking about. That woman, the one who was there first, who built the thing, who has spent a career modernizing telecom, fintech, healthcare and utility systems, sat down a few years later and felt like she'd missed it.
"I don't like that I'm feeling behind the whole AI train, and I need to jump on it or take another train. The first one already departed."
If the person who was early still feels late, then the feeling isn't a measurement of your skill. It's a symptom of how you're consuming the field.
Viktoria noticed the same thing in nearly everyone she talked to:
"There is a lot of fear-based motivation that is happening. I have not yet heard anybody telling a positive emotion behind it. It's something new, and it's natural for people to start from uncomfortable feelings. The most important is that you go there."
This is the hardest, but most valuable advice for senior product leaders today. Embrace the discomfort. Go there anyway.
She tested the "you don't need to code" claim on herself
Viktoria started as an Android developer before moving into product, so when she heard that people with no coding background were building working software, she didn't believe it. She had every reason to read that as marketing.
Her capstone project in the AI Career Boost Blueprint was a Google Drive optimizer, and she chose it almost apologetically. She couldn't find an idea she loved, so she went with the practical one: her Drive was a mess, it goes back to being a mess every three months, and that was a real problem worth solving. The idea was never the point.
"That was really cool, proving to my skeptical self: yes, people don't really have to have any coding knowledge. Though I had it, I literally could have entered it without having any coding knowledge whatsoever."
The genuinely useful part is was realizing what stalled her, because it wasn't the AI.
"The main pain point was not even about how the whole system works. It's about how to connect through Google to set up authentication. Why is it not working in test environment? I literally was only testing in prod, because it was not possible to set it up."
Anyone who has shipped software just laughed along. She got the intelligent part working and then lost her time to OAuth configuration and the fact that she couldn't stand up a test environment, so she tested in production like everyone's least favorite engineer. That is what real building feels like. It's the lived experience that gives you confidence collaborating with engineers, and makes you their favorite product leader to work with.
Viktoria posts her best AI tips as she finds them — follow her on LinkedIn. She generously shared advice on the top two prompts every product leader should be running:
Prompt one: audit how you're actually using AI
This is crazy high-leverage, and takes two minutes. Open the AI tool you use most for work and ask it:
Based on everything you know about me and all chats and artifacts I've made with you, please tell me: (1) how I'm scoring on the quality of things I'm using you for, (2) summarize what I'm using you for, (3) where I can improve on how I use you.
She ran it on her own Claude account and scored 7.5 out of 10. Above average, and specific about why it wasn't a 9. Some of her prompts were doing five unrelated jobs at once, which meant every answer came back shallow. She was pasting Confluence and Jira links rather than the content itself, so the model couldn't actually read what she was pointing at. She was closing prompts with vague throwaways like "emails refinement."
None of that is a knowledge gap. It's a habit gap, and you can only see it if you look. Two other tips Claude gave Viktoria, that benefit everyone:
- Say what shape you want the answer in, every time. "Rank these 1 to 5." "Give me the top two." She did this occasionally and was told to do it always, because naming the format up front saves the round trip where you ask again for the same thing in a usable form.
- Ask it to push back. Adding "tell me if any of my assumptions are wrong" or "what am I missing?" to an analytical prompt. She'd done it once, and it was flagged as her best habit.
Read that second one again, because it is the player-coach instinct exactly. The instruction that improves the output most is the one that invites disagreement.
Prompt two: ask for a distribution, not an answer
There's a paper out of Northeastern and Stanford called Verbalized Sampling that names something you've probably felt without having a word for it.
Training a model on human preferences teaches it to prefer the most typical answer. Annotators reward familiar-sounding text, so the model learns to collapse toward the safe middle of everything it knows. Researchers call it mode collapse. You experience it as asking for ideas and getting the ideas everybody gets.
The fix is almost insultingly small. Instead of "write me a tagline," ask for five, with their probabilities. Instead of "give me a positioning angle," ask for five with their probabilities.
Asking for a distribution rather than a single answer releases the pressure to be typical. The researchers measured roughly double the diversity in creative tasks, using the same model, at the same temperature, with no special access and no retraining. Just a different ask.
Which is the same lesson as Viktoria's, arrived at from a different direction. Tell the model the shape you want, and ask for the range instead of the answer.
The player-coach move
There's one more thing from our conversation that I want product leaders to hear, because it's the sharpest distinction in it.
Most product managers working with an AI vendor treat the model as a sealed box. They hand over requirements, wait, and accept whatever comes back. Viktoria doesn't.
"Even if you do not own the actual custom model, you understand what the model needs to perform better. What is mandatory, what is optional, talking to them about which rules they have and what they recommend, gives you clues to what they're using, maybe how they're prompting."
She reads the required fields and the optional ones. She reads which third-party libraries the vendor asks her to deploy. She asks the vendor's own engineers what their other clients configure and what produces the best results. Same vendor, same tool, better outcomes, because she knows enough about what's under the hood to ask a real question.
That's the player-coach move in one example. You don't have to build it yourself. You do have to know enough to tell whether what came back is any good.
Where you actually stand
Both of those prompts tell you about your tools. Neither tells you about you.
That's the harder audit, and it's the one I'm running live at the next masterclass. We build, in the room, an audit of your own skills and experience against what employers are asking for and what companies need right now. Not a checklist I hand you. Yours, on your background.
If you're done feeling behind but you look at the pile and think there's far too much to catch up on, that session shrinks the pile. You leave knowing what to learn, what to practice, and what to be able to show.
Viktoria didn't out-read the feeling. She found out what she was actually missing, then went and got it.
Listen to the full conversation: Ep 64, Afraid you're falling behind in AI? Watch this first, with Viktoria Otstavnaya — YouTube · Apple Podcasts · Spotify
(For more, drop AI Career Boost a follow on YouTube!)
