Prototyping With AI
Wrapped up listening, observing, communicating, chatting, and speaking at the Learning Technologies '26 conference in London. There was one topic that dominated the Expo floor and almost every session:
Human Intelligence Vs. Artificial Intelligence
Who's winning? Is this a competition? What's hype and what's realistic today? Where's learning going? Are we making a difference? What's changing? What should be changing? Are we behind? Are humans interested in measuring impact or measuring the illusion of impact? Can we still connect as humans in the age of Artificial Intelligence (AI)? Here's my one-word takeaway from the experience: dialogue.
I've written two screenplays. One of them was bad. But in between, for years, I was working on my craft, creating good dialogues.
Dialogue is a conversation between two or more people, or the written exchange between characters in literature, plays, and films. It acts as a tool for characterization, revealing personalities and advancing the plot, and can also refer to a serious, cooperative exchange of ideas aimed at mutual understanding.
So, imagine, for a second, that we are characters in a movie. We all have a backstory, a belief system, a history of failures and successes, biases (known or unknown), etc. Some characters have human intelligence in our story, while others have artificial. We have a limited view of the world, past, present, or future. Dialogue takes place in scenes to drive the plot. Every scene matters in a movie. As they advance the plot, they reveal personalities and help characters grow.
Reflection in author's sunglasses
What Is A Dialogue Not?
Speeches, downloads, mansplaining, lectures, content, information dumps, dashboards, Sharepoint sites...
Scene 1: International Speakers' Dinner
Before the conference, some of the speakers and chairs of the conference got together for an informal dinner. What did we eat? I don't remember the food. But I remember the characters and the dialogue we had. Dialogue assumes the common goal of mutual understanding! Mutual understanding doesn't mean complete agreement. You can completely disagree with someone and yet have a dialogue with them. But this only happens when there's at least some level of mutual trust, respect, and openness. A dialogue includes listening. Active and open listening. Not waiting for your turn to speak. Waiting to respond.
We touched on psychological safety, playfulness, food, travel, and, of course, some learning-related topics. There were no slides, no job aids, and no clicking next. Building connections through dialogue will remain crucial in the age of AI.
Imagine two situations:
- Your manager sends you a beautifully crafted note about your accomplishment in a project. Brief, concise, emotional, with perfect grammar. Except, it's clearly written by AI.
- Your manager sends a note about the same accomplishment. It's not perfect, but it took some time and effort between two important meetings. It may even have a typo.
Most people would automatically say they prefer human-authentic messages and comms. But do we? There are AI influencers with brand authenticity driving online traffic, chatbots rated more empathetic than human doctors, or customer service AI agents replacing the long on-hold waiting because of "unusually high call volume."
I don't have the answer, but I suspect that where the interaction is transactional, practical, and you don't care about the long-term relationship, AI will dominate the dialogue.
Scene 2: Reality Vs. Hype
The current landscape of AI feels like the Land of Oz. On hand, the illusion of magic is dominating LinkedIn: experts in every corner with frameworks galore. Every single decent learning technology vendor now offers AI-driven features, from content creation to simulations. While L&D is still working on prompt engineering, some leaders have moved on to context-engineering, while the rest of the world is building chief of staffs for themselves with OpenClaw.
Where is the result?
DX has looked at AI and engineering results in a longitudinal study:
Many leaders feel their organizations are falling behind in the race to unlock AI-driven engineering velocity. Vendor marketing and social media set expectations at 3x or even 10x improvements. When leaders see more modest results, they assume something is wrong.
To provide that picture, DX analyzed engineering velocity from November 2024 to February 2026 across a sample from 400+ companies where AI adoption rose sharply. We found a 10-15% increase in PR throughput a real gain, but well below what most leaders expect.
The paper then dives into details of why the expectations of performance gain through AI have not been met so far [1].
What About L&D?
There is plenty of research now focusing on the impact of AI on L&D. Research findings coming from RedThread Research, Egle Vinauskaite, Markus Bernhardt, and others, provide some guidance on what's happening to L&D (and beyond), and how to take charge of the future.
Speaking of taking charge: My session was very specific to quick prototyping with AI tools. L&D always had a problem with quick, iterative design to show working models. It used to require technology expertise and often IT help. Today, AI can accelerate the process and enable learning professionals to experiment, iterate, and learn quickly through prototypes. I described this as a journey where you need a destination that is worth going to (business problem or opportunity), a vehicle (an AI tool that fits your need through cost, speed, and control), and a map on how to get there (not a static map in the old sense, more like GPS directions with just how to start the journey).
But if we let AI drive this process, and we just passively participate, it's going to be an expensive journey to learn how fast we can go to places we never meant to be.
The reality is that AI is not a technology that L&D should "adopt." At least, that's not the only angle. And it's definitely not the starting point. It's tempting to show the efficiency gain by using AI to automate content creation, for example. My challenge for all L&D leaders is to move on from the faster content creation and measure effectiveness. And that doesn't start with AI. It starts with understanding how we work today, and how we should work tomorrow:
- How do things get done today? What's the workflow?
- Who makes what decisions?
- Who's responsible for what output?
- How do you define quality for a specific output? How do you check quality?
- What's the outcome expectation?
I know asking questions can feel like it's slowing you down, but it will help accelerate you on the journey while reducing the dead ends you'd be running into.
Scene 3: Why To Prototype, What To Prototype?
A common mistake is to treat a prototype as a cheap version of the real thing. These prototypes often get stuck in the prototype stage because they're not scaling and aren't actually answering any questions (other than "can we build it?").
A prototype is for learning. Learning something quickly and iteratively. The prototype should be focused on the most critical part of the experience you're simulating. If it is your first AI chatbot to assist employees, you don't need to build out a full-blown application to learn that what it produces is not relevant for your audience. Play-testing with real business problems and real users is key.
What if you learn that your idea doesn't work? Well, you saved resources and time to build something that will. I've seen so many application "adoption issues" within the corporate world because the team didn't prototype the core experience. The "if build it, they'll come" is not a strategy.
What To Prototype?
First, start with a business problem or opportunity that is worth solving for. Efficiency is an easy target, but it can backfire. Once, I created an automation that took text and created a PowerPoint deck from the content in minutes. I thought I saved hundreds of HeH (human equivalent hours). Sort of. It helped us drag on with building an ineffective voice-over presentation faster. Again, make sure there's a business case for the future, not only for the present stage.
Second, start with the end in mind: who your audience is and how they'll access the solution. The prototype doesn't have to be perfect, but for scalability, you need to keep your ultimate delivery in mind while making a prototype version of it.
Who's the target audience?
- Yourself
It can be a practical application that helps with proficiency or quality check. For example, if you're responsible for checking assessment question quality, it is a great target for a skilled AI agent. If you're not building an AI agent yet, but you want to improve the User Experience in the eLearning courses you create, that can also be a practical target. - Your peers
What if you could solve for bottlenecks in your team's current workflow? What if you could build something that augments that process or even replaces some of the elements? For example, if you're using xAPI, you can create a statement builder for your team that follows your standards and produces drop-in-ready code. If you're still fiddling with SCORM, you can build the same. - Your organization
What if you could solve for cross-functional workflow bottlenecks? What if a utility tool could help others do their job easier, faster, or find relevant information quicker? What if you could get rid of old, stale training courses and replace them with an interactive assistant for real-time support? - Employees ("learners")
What if you could embed a dialogue inside a learning experience? Or a simulation that is tailored to the role, location, and previous skills level? Sometimes, you just need to be "innovative" in the sense of being resourceful: you already have an LMS that authenticates users and stores data (via SCORM cmi statements), so you could deploy a utility tool that is relevant, customized, and practical, with a deep-link launch. Of course, a dedicated web server with single sign-on would be better, but in the meantime, you could prototype the tool.
Speaking of access: I suggested in my session that, regardless of how small the first prototype would be, everyone should start with planning. Specifically, planning the whole solution (not just the prototype) in a product requirement document (PRD). All LLMs know exactly what a PRD is, and they can build the foundations for you. You can then expand this document as one of the project artifacts.
Whatever AI tool you're using (I'm alternating between Windsurf, Claude Code/Coworker, and Github Copilot), this fundamental PRD will help make decisions and set the tight scope of the prototype with the ultimate solution in mind. All of these above are related to one thing: dialogue. Meaningful, iterative conversations between humans and AI.
Now, Go And Build Something!
P.S. If you wonder what the picture represents (beyond reflections in sunglasses), you'll need to investigate the Banksy sculpture in the background. Originally, it is supposed to be about blind patriotism, with a person blinded by the flag stepping into a free fall. For me, it brings similarities to AI. Take charge, learn, and experiment. Don't just blindly follow influencers.
Image Credits:
- The photo within the body of the article was supplied by the author.
References:
[1] AI and engineering velocity: A longitudinal analysis
