Dave Killeen: [00:00:00] What if you could build an entire application that lives inside a prompt in under a minute? Think about it for a second. We are so used to thinking of prompts as questions that we ask, but what if the prompt is an application? Here's what I'm gonna show you today. I'm gonna build an AI career coach, a prompt based application with four distinct modes triggered by commands.

It asks me targeted questions instead of me guessing what context to provide it. And it produces different outputs depending on what command we give it. It remembers everything across all the chats that we've had with it. This isn't a conversation. It's a tool. It's an application architecture built entirely in natural language.

And here's the kicker. I'm gonna build this entire Multimode application in under a minute by describing what I want, in plain English to Claude Console from episode four, who we met. Then you give it a vague prompt and have Claude generate this massive war and peace prompt. An amazing, detailed prompt over pages.[00:01:00] 

And then I'll show you how I actually use this prompt every Friday in my weekly check-in, right through to having monthly reviews, end of year reviews and promotion assessments. This is bonkers, total bonkers. And this is what happens when you stop thinking of prompts as inputs and start thinking of them as software that you can just create with words.

So we're folk new to this podcast. Here is just a few seconds for me to properly introduce myself. My name is Dave Colleen, and I'm the Field CPO at Pendo here in EMEA. And a very warm welcome to the Vibe PM podcast episode five. And for those joining, this is the podcast where I do deep dives on AI tools, tricks, and shenanigans so you don't have to. Super short, practical demos all under 10 minutes to help us all dance more effectively with AI. And for returning folk, a warm welcome back. At today's episode, we're building on everything from the previous episode, so speech to text from episode one. If you haven't checked it, please do. Claude Projects from episode three, Claude Console from episode four, and today we're bringing [00:02:00] it all together to build on something you'll actually use every single week. And while I'm calling out Claude for this demonstration, these principles apply equally to ChatGPT, with custom GPTs, or Google Gemini with gems.

The concept of prompt based application works across all the frontier LLMs, and I'm just showing you Claude, because it's what I use. So let's get to it. Okay, so I've hopped on over to cloud projects. This is from episode three. Do have a look at it if you haven't already, and this is where I'm gonna load in my context, my persistent knowledge that the AI will then dance with, as in when I have a chat through this project. Here you see I've got five key documents for the purpose of this demo. I've got a career ladder, so all the competencies expected of me, of my current role and what's needed to get promoted to the next level. I've got this insights profile, which is a strengths and weaknesses profile that I get here at Pendo.

And I've got my career goals, and then I've got my job description, and then I've got my recent annual review. So every time then I dance with this particular chat. In this project, the AI will then leverage all of these documents in [00:03:00] here. Now I'm gonna hop over to Claude console because what we haven't done yet is actually given it the prompt that we wanna generate.

So I'm now in Claude Console and I'm just gonna do the following with voice that from episode one, if you haven't checked that out, do check it out. You'll see it's quite powerful. I need the career coach assistant. That helps me with professional development. Here's what it needs to do. I want to be able to do a weekly brain dump about my work, just rambling thoughts about projects, challenges, wins, frustrations. Before generating any output, it should ask me targeted, clarifying questions to pull out the most relevant context. It needs four different modes triggered by commands. Number one, generate report. This creates a professional weekly report from a manager with sections for projects, achievements, challenges, and support needed.

Two, prepare monthly reflection. This analyzes patterns across my weekly check-ins and gives me insights on trends and focus areas. Number three, prepare best self review. This generates a comprehensive yearly reflection for annual reviews. Number [00:04:00] four, promotion assessment. This compares my demonstrated competencies against the career ladder in project knowledge, cross references my insights, profile strengths and weaknesses. That's also in the project knowledge and assesses readiness for promotion based on my career goals and my current job description, which is also in the project knowledge. It should all use a professional but supportive tone and give specific, actionable insights and always just format this output clearly with bullet points and sections.

All that goes in. I'll go to generate that and we'll see what it comes back with.

Perfect.

Again, you can just see how good Claude code is and it's just so cheap. Like I said before, in previous episodes, about $5 gets you a whole year's worth of credits to do this dance back and forth. This is so cheap to work with, but obviously incredibly, incredibly effective. So if I just continue to take that in and pop on over into the workbench in [00:05:00] Claude console, and if I just copy and paste that and then take that into my prompt over here and put that in here, then I've got everything I need and we're good then to go. And I did invite a single line of prompt engineering for that. Now, before I show you what we do next in practice, quick disclaimer for me, the weekly check-in that we're about to do is entirely thankfully for dishes.

It's not a reflection of my day to day or week to week here at Pendo. So. What I'm now gonna do is I'm gonna go into Claude on mobile app using our voice and I'm gonna go, Hey Claude Dave here. Happy Friday. Uh, alright, Claude. So this week was intense. We finally shipped the new analytics dashboard and that felt good, but honestly, I'm frustrated with how the stakeholder alignment went.

Engineering is super stretched thin. And Paul, I'm wondering if I'm delegating effectively. I'm also having very tough conversations with sales about timeline expectations. And then I'll just dump that in, right? And I can just keep dumping in other notes. And then all of that is just raw, unstructured thoughts.

No planning, just [00:06:00] me dumping and runting. And here's the magic. The career coach just doesn't regurgitate what I said. It will then ask intelligent follow-up questions as you're about to see next. Okay, so what you can see in the screen here is it's got my voice note from mobile on the web app here. So the whole thing is obviously in sync.

What you do on mobile is shown on web, by the way. It's a great way to work with your LLM if ever you're trying to give feedback to what it produces. Take out your mobile app. It has the same thing you're seeing on the web app, and then you can just use your voice on the mobile app to again give it feedback.

So what we're seeing here now is it's come back with a whole bunch of questions on what I dumped in to get more clarifying detail so that we can generate the report for me. So I'm just gonna go through voice and answer all those questions. I won't go through all of them in the interest of your time.

But first up, the dashboard role. I was leading the product decisions and working with engineering on prioritization. It felt really good because we have been working towards this for three quarters and customers have been asking for it. On the second one, stakeholder alignment. The answer there is that product and engineering were.

Aligned on [00:07:00] scope, but sales had different expectations. The key stakeholders were the sales and leadership team from our regional VPs who were making promises to customers. And so in the interest of your time, just assume you've gone through all of that back and forth, and then 10 minutes past and next what you'll see is what it produces.

And this is fascinating. So just imagine what you're seeing here on the screen is me having given a voice note on all of these five different sections that it's asking me for further detail on. So that's what all of that is, and now it comes back. And because the prompt is designed in such a strong way.

It's given me options on what to do next, right? So option A, would you like a weekly report for your line manager? Option B, would you like some personal processing and action planning? So how can we help identify what's in your control versus not? Can we create for you a concrete plan for next week, particularly around that delegation issue?

Option C. Would you like to go through a strategic framing exercise referencing the fact that my line manager in my annual review has talked about my role being focused on one thing and [00:08:00] maybe not going into doing other things. And so really is trying to keep me on track here so it's so, so good. Right?

And then I'll just go look Option A, please. I'd just like a weekly report and then here you go. Like a cookery program one we made earlier. It's just great. Right? And this just takes 10 minutes of a dance. And of course, like I said to you earlier. All of this builds up, all of this compounds over time, and this is great fodder, great content then for the AI to also dance with as you move forward on a weekly basis and start doing your monthly reflections and so on and so on and so on.

So this is, this is it. And remember that's just one of four modes. There's a monthly reflections mode in here. There's a yearly reviews mode, there's a promotions assessment mode. All triggered by simple commands that build on you. Just having a regular chat with the app on mobile with your voice once a week, and because you're in that same chat with all that rich context, with that mega prompt, it's getting smarter the more you use it.

It's remembering everything and it's leveraging all that knowledge. And so you're not just using [00:09:00] AI here, you're building an application. With a prompt that works for you. And here's the best part. Claude Console makes this bonkers, cheap, and accessible for the price of a cup of coffee. You get about a year's worth to the best prompt generation that's out there.

$5 will see you at least I would say 12 months through. There are other tools out there that lets you do this. But ultimately I do believe Claude Console is far better than any of them out there because it's from Anthropic, the leaders, when it comes to prompt engineering. So if you have any workflow, think it through and here's how you would then reproduce it.

Open up Claude console, and describe at a very high level what you want. Claude console, as you'll see, will transform it from something very simple into a super sophisticated prompt. What you put in, what you get out night and day. Number two, load your context. So create that Claude project or a custom GPT or a gem from Gemini and upload all that context, those documents that you wanted to dance with along with the prompt.

And number three, use that [00:10:00] same chat over and over. Keep going back into it every time you can, but consistently. And that value compounds over time as it learns from you. And so there you have it. And here's the big idea. Stop building prompts. Start building prompt based applications. Pick one problem you face every week.

Use Claude console to design a flip script application for it. Thank you again for joining me on episode five of The Vibe PM and if you found this valuable, please share it with someone who needs to hear it. And please spread the vibes and I'll see you next time. Thank you. Take care.