Powering Past the Prototype
FM (Friday Morning) Reflection #45
One of the things I’ve written about here is how AI makes it possible for creators to become tech builders.
Now this idea is poised to become mainstream. If you paid attention to the ads in Super Bowl LX’s matchup between the Seattle Seahawks and New England Patriots (less than three full weeks ago but seems like ancient history), you may have noticed the plays taking place in a high-stakes battle off the field.
I’m talking about the rivalry between AI companies working on the next hot-new-exciting-different thing to capture more of your time, attention, and wallet share. And since consumer attitudes influence corporate technology trends, they’re worth paying attention to.
Anthropic took a sharp swipe at OpenAI’s move toward ad-testing, while newcomers like Base44 and Genspark leaned heavily into the “get things done” narrative. It mirrors a shift I anticipated months ago: the evolution of AI from a novelty to a foundational tool for builders.
The ads depicted instant success while your own experience may vary — after all, sometimes we try something out to find out it is not, in fact, a good idea. But now that we can prototype and “fail fast” quickly, it’s easier to find out.
Time and time again, I’ve sat down at my computer with an unproven idea and wondered, “Will this work? Let’s try it.” Many have, to my surprise and delight, and several of those proofs of concept led to large-scale product and tech initiatives.
Answering “what if” with a working prototype allows you to “show not tell” and is a powerful tool for rallying support around an idea.
Over the past months I’ve immersed myself in the experience of taking these AI app-building tools out for a test drive — building new enterprise-quality digital products and services with AI assistants — and here’s what I’ve learned.
Step One: Start!
Google’s AI Studio is a consumer-oriented tool that you can use to get a first prototype of an app built in as little as 10 minutes. It will definitely be a rough cut, but it’s often a good jumping-off point, and might even be good enough for your own personal use or to share with a small group.
If you’re a serious technology leader or builder, don’t let the “consumer-oriented” label turn you off. AI Studio will let you produce a single-page application (SPA) that you can easily port to an enterprise development environment later. If you want to start prototyping directly within Google Cloud, try out Vertex Studio. It is evolving to provide similar functionality albeit in a wonkier interface and with the overhead that comes with a cloud project (not for beginners).
If you’re more proficient with coding and comfortable working in more of a command-line interface (CLI) or IDE environment, Codex from OpenAI, the Google Gemini CLI, or Claude Code might be more up your alley.
Step Two: Decision Time
Once you have your prototype and you’re loving it, it’s time to get real about who you’re building for: is it for you or for someone else (e.g., a consumer audience or the enterprise)?
If you’re building for yourself, then you can keep at it. Drive that bus to the ends of the earth if you want to.
But if you’re creating with the intent to deploy to a consumer marketplace or to the enterprise, you need to make some choices. Each of them deserves a full post, so I’ll summarize some of the key points you’ll need to consider:
Design the Workflow: Is there a particular way that people will want to (or need to) approach the work they do in your app? Is there a natural order to the process or will they use your app in an ad-hoc manner or as a point solution?
The simplest way to answer this question is to pull out a piece of paper and start to draw out how the process should work using simple boxes and arrows. If you can draw it simply, then chances are good that you can design and implement it.Clarify Your Architecture: Are you adhering to a design philosophy that separates concerns? Ensure your business logic, data storage, and user interface are distinct layers.
As you design your workflow (above), think about how to separate out jobs to be done by the app. This is different than the work the user will do — think about it like a kitchen at a restaurant: you put in your order, but behind the scenes, the work is split out between different workstations and people who are skilled at specific tasks.
Identify those groups of tasks and consider dividing the work across different services, workers, or functions.Map the Integrations: AI shouldn’t be an island. How will it communicate with your existing systems of record (CRM, ERP, etc.)? Define clear, secure interfaces for these handshakes to ensure your new tool plays nicely with your legacy tech.
Define Data Persistence: If a user creates something valuable, can they save it? Decide early how you will store user state and data. Is it ephemeral (gone when the window closes) or persistent? If it’s persistent, where does it live, how is it backed up, and who owns it?
Evaluate Your Security Posture: Your prototype app isn’t likely to have any security built-in. Do you need to handle authentication and authorization? On the back end, if AI is creating content or making decisions, how will you monitor the AI’s activities, and what is your “circuit breaker” in place to stop it when it need redirection?
Plan for Scale: A prototype works perfectly for one user. What breaks when a thousand users hit the app or API at once? What will the costs for processing be? This includes understanding the how latency tolerance, token costs, and rate limits all come into play before you put your app into production.
Step Three: Go or No Go
After you work through all the steps above and make decisions on how to move forward, you’ll be able to separate the talkers from the doers when it comes to who knows how to drive innovation with AI.
What is missing in today’s tools are easy-to-implement plug-in frameworks that can safely and reliably handle the boring but essential functions: security, messaging, storage, commerce, and keeping users and their data safe.
Failure to consider how a prototype can be scaled up and into the enterprise environment is one of the main reasons that I believe we hear so much about AI initiatives that go off the rails.
When I read about failures to scale AI applications, I suspect this is a primary culprit for why things often get stuck, in addition to the known-known challenges we’ve been talking about for some time: probabilistic outputs, prompt injections, and jailbreaking.
This gap brought me back to a lesson I learned early in my career: before you can direct the show, you have to learn how every bit of it is put together. It takes practice, sensitivity, and a high degree of situational awareness to seamlessly orchestrate a production from beginning to end.
Conclusion
Learning to lead through the transition from AI as a novelty to a foundational tool for building mirrors the transition that we each go through in elevating our skills from contributor to manager to leader. It requires a willingness to learn every single part of the process, and to put together an experience that provides value while respecting the users of the products and services that we create.
So as we navigate the “Super Bowl” era of AI competition, let’s remember that the winners won’t be those with the flashiest ads. They will be the builders who understand how to bridge the gap between idea, design, implementation, and delivery for our customers and consumers.
Related Reading
Photo Bonus
Speaking of knowing your audience — I took this photo for me, and upon reflection, realized that you might like it too. If you’re a long-time reader, you may remember an image that I shared in Being ready for the magic moment when everything clicks.
This time, I was marveling at the sight of the moon rising over the horizon and decided to grab my camera. The clouds moving in front of it created a spooky Halloween-y feel, and when I lifted the camera to my eye, the shot was photobombed unexpectedly.
At first, I wasn’t sure if it was a bug or something on my lens, but quickly realized it was a passing plane, and that made the shot.