A Blank Canvas

Step into any painters studio and you'll find a canvas waiting on its easel, a patch of drop cloth spread beneath—brushes and colors nearby, but no instructions about getting started or rules that dictate what comes next.

The tools are there, but they only step forward when called and never impose on the artist a type of technique or workflow they should follow.

When you're forced to be simple, you're forced to face the real problem. When you can't deliver ornament, you have to deliver substance.

The most thoughtful productivity tools, the apps we want to use everyday, remain useful over long periods of time because they enable us in the same way the canvas and easel enables painters. They step back, support experimentation, and trust you to shape your work—whether writing, synthesizing scattered research, threading together study materials or managing tasks, just like the readiness of the canvas and easel, the products I want to use the most have no barriers to getting started and they don't force a choice between simplicity and power.

As large language models become more capable, we're starting to see glimpses of what truly adaptive software might look like—and the old reasons to choose between simplicity and power are quickly disappearing. This shift made me realize the question isn't whether AI can change how we work, but whether our tools can become more like that painter's studio and create space for AI so that it does, or, will they just add more complexity onto already crowded interfaces.

Less on UI and more on AI.

What happens to all the traditional buttons, drop-downs and forms when the inference catches up with the anything in, anything out nature of large language models? Once it enables contextual interfaces that appear on demand, as fast as a static page loads? I won't attempt to predict the end of UI as we know it, realistically, most of us aren't ready to leave visual interfaces behind. Still, the tools of today I want to use most are already putting more emphasis on being adaptable, focusing less on UI and more on AI.

But adaptability isn't automatic, it requires intentional design. The best of these tools share a common approach: building a harness for AI that enables the same canvas and easel relationship—AI integrations that step back, support experimentation and trust you to shape your work. Understanding that most people don't want to be locked into a products built-in AI model, but rather the ability to use whatever AI model they want. This preference runs deeper than just picking between providers like Anthropic, OpenAI, or Google—most of us are looking for AI integrations that adapt to our context and workflow, not the other way around.

What I found is that harness, the interface built to bring model intelligence to the end user, can dictate the experience more than the models themselves. Build a better harness, and choosing a particular model becomes less important than how the product enables each model to be used. Yes each model has its strengths, but if the use case is defined rigidly by the product rather than left to be shaped by the user, that product isn't offering a blank canvas. It's not stepping back and trusting users to define their own work—ensuring tools are there, but they only step forward when called and never impose on us a type of workflow or way the AI should be used.

10 am on a Tuesday morning

The most thoughtful tools should pass what Charlie Deets calls the "Tuesday morning test"—anyone should be able to switch to it at 10am on a busy workday without missing a beat. No tutorials, no configuration, no risk of being slower at their job. Just like walking into a well-appointed studio, everything should feel immediately familiar yet somehow better than what came before.

These tools inspire communities that become a part of the discovery process—users sharing use cases and surfacing capabilities that might otherwise remain hidden. At first glance you might call this poor discoverability, but that's missing the point. If you haven't found a feature, it's probably because you haven't needed it yet. Smart UI in the AI era, and likely always, lets capabilities surface organically, when your work demands them, rather than cluttering the interface upfront. This creates tools that feel simple to newcomers while offering depth to those who've grown into needing it—exactly the experience painters have always had with their canvas and easel. The brushes and colors stay nearby, ready when needed. The techniques are there to discover, but only when the work itself calls for them.

These patterns are clear across the product landscape for those who code, whether professionally or as enthusiasts, but not so much for those who write, research, synthesize ideas or manage them. Which brings us back to that fundamental question about AI integration... The tools exist, but they haven't yet found their way into that seamless, canvas-like experience that lets you focus on the work itself rather than managing the workflow. While some have built impressive systems combining multiple tools, they've had to become systems integrators—managing subscriptions, learning technical interfaces, coordinating between tools that don't naturally work together.

The canvas and easel haven't changed in centuries because they got something fundamental right about the relationship between tool and creator. They offer a foundation as solid as it is open, a space that expands quietly to meet whatever shape your work might take. The right harness around Large language models might allow software to offer the same—tools that feel as natural as reaching for a pen, as ready as a blank canvas, and as responsive to your workflow.

Proven with Grit