Great quote: “The great horizontal killer applications are actually just fancy data structures”.
I would add: it’s not just data structures, but information representations that expose those data structures in an intuitive way.
Intuitive how?
In that the graphical representation and interactive controls provide a transparent interface to the underlying data structure, where:
- its graphical representation easily communicates that data structure’s current state, and
- its interactive controls allow for a wide range of operations for manipulating the data structure’s state
This idea of Excel not being widely used as a calculation tool, but a way to lay out lists and tables, seems to suggest that most people see software first as a medium, then a tool, secondarily.
A medium lets you shape information into specific representations.
Different representations help you externalize information in a different ways that facilitate certain styles of thinking:
A Kanban board lets you visualize tasks in a left to right fashion that represents the status of a task, so you can visually grok progress. The columns let you see how “full” your workload is, how congested your pipelines are.
A diagram lets you see relationships between parts of a system, or between concepts, more easily. Without which, you would struggle to think non-linearly.
A table lets you think of your data in two dimensions.
Yes, some of that data could be quantitative in nature, and so this medium lends itself well to calculation, aggregation, and charting.
But before you can manipulate data, you first need to be able to think about that data.
That’s why spreadsheets still feel more natural to use than SQL reporting tools: those tools conspicuously lack a medium for which data can be represented as useful information first and foremost.
I would add: it’s not just data structures, but information representations that expose those data structures in an intuitive way.
Intuitive how?
In that the graphical representation and interactive controls provide a transparent interface to the underlying data structure, where:
- its graphical representation easily communicates that data structure’s current state, and
- its interactive controls allow for a wide range of operations for manipulating the data structure’s state
This idea of Excel not being widely used as a calculation tool, but a way to lay out lists and tables, seems to suggest that most people see software first as a medium, then a tool, secondarily.
A medium lets you shape information into specific representations.
Different representations help you externalize information in a different ways that facilitate certain styles of thinking:
A Kanban board lets you visualize tasks in a left to right fashion that represents the status of a task, so you can visually grok progress. The columns let you see how “full” your workload is, how congested your pipelines are.
A diagram lets you see relationships between parts of a system, or between concepts, more easily. Without which, you would struggle to think non-linearly.
A table lets you think of your data in two dimensions.
Yes, some of that data could be quantitative in nature, and so this medium lends itself well to calculation, aggregation, and charting.
But before you can manipulate data, you first need to be able to think about that data.
That’s why spreadsheets still feel more natural to use than SQL reporting tools: those tools conspicuously lack a medium for which data can be represented as useful information first and foremost.
What constitutes a medium for information, as opposed to a more prosaic data manipulation tool?
I think of it like this: a tool needs to operate on a medium.
A medium is a graphical representation of some data, which means said data needs to be able to be visualized from the get go.
The tool, then, comes later, to directly manipulate that visualization in a way that shapes that visualization into a desired state.
Early productivity software that succeeded got this basic principle: desktop publishing, DAWs, photo editing tools, and non-linear video editing tools.
A pencil operates on pieces of paper.
An eyedropper on color.
A knife on a video reel.
A design principle: When designing productivity software, conceptualize it as a medium first and foremost, and a tool secondarily.
I think of it like this: a tool needs to operate on a medium.
A medium is a graphical representation of some data, which means said data needs to be able to be visualized from the get go.
The tool, then, comes later, to directly manipulate that visualization in a way that shapes that visualization into a desired state.
Early productivity software that succeeded got this basic principle: desktop publishing, DAWs, photo editing tools, and non-linear video editing tools.
A pencil operates on pieces of paper.
An eyedropper on color.
A knife on a video reel.
A design principle: When designing productivity software, conceptualize it as a medium first and foremost, and a tool secondarily.
Coming back to the original thread by @simonw
The original claim: ChatGPT is a power user tool, just like Excel
I’d argue that ChatGPT is not like Excel, and quite the opposite:
Excel expresses a rich medium for information manipulation, the spreadsheet.
ChatGPT expresses no such thing, and doesn’t yet rise to the level of a rich information medium, and is still at the level of a mere tool, albeit a powerful one.
It’s a tool that @OpenAI hopes will be able to be applied to all sorts of other media living outside ChatGPT, such as existing desktop apps.
Its browsing, code interpreter, and Dall-E tools are testament to this upside-down design philosophy: the tools go to work first, only thereafter you see some output.
With the exception of code interpreter, interactions with ChatGPT are largely stateless and transactional. Manipulations are indirect.
Whereas Excel is stateful, to such a great degree that it’s functionally reactive: all interactions result in near-instant feedback.
This leaves third-party developers to include ChatGPT as a tool inside of their apps. ChatGPT plays host to that app’s native medium.
I think that makes a lot of sense for OpenAI.
It’s also good thing for everybody else who isn’t OpenAI.
If you’re building software that’s a rich information medium and not just a tool, there’s no need to worry that OpenAI’s future model capabilities will outpace your own, because OpenAI’s future model capabilities are merely an improvement in tools that your software’s native medium will play host to.
The original claim: ChatGPT is a power user tool, just like Excel
I’d argue that ChatGPT is not like Excel, and quite the opposite:
Excel expresses a rich medium for information manipulation, the spreadsheet.
ChatGPT expresses no such thing, and doesn’t yet rise to the level of a rich information medium, and is still at the level of a mere tool, albeit a powerful one.
It’s a tool that @OpenAI hopes will be able to be applied to all sorts of other media living outside ChatGPT, such as existing desktop apps.
Its browsing, code interpreter, and Dall-E tools are testament to this upside-down design philosophy: the tools go to work first, only thereafter you see some output.
With the exception of code interpreter, interactions with ChatGPT are largely stateless and transactional. Manipulations are indirect.
Whereas Excel is stateful, to such a great degree that it’s functionally reactive: all interactions result in near-instant feedback.
This leaves third-party developers to include ChatGPT as a tool inside of their apps. ChatGPT plays host to that app’s native medium.
I think that makes a lot of sense for OpenAI.
It’s also good thing for everybody else who isn’t OpenAI.
If you’re building software that’s a rich information medium and not just a tool, there’s no need to worry that OpenAI’s future model capabilities will outpace your own, because OpenAI’s future model capabilities are merely an improvement in tools that your software’s native medium will play host to.
https://x.com/simonw/status/1799455534506283191
And meanwhile, Apple goes: well, image gen is just one of many drawing tools in Apple Notes
https://twitter.com/sdw/status/1800235186619310566
Math Notes from this year’s WWDC was another great example of a rich interactive medium.
There’s something visceral about having something both directly manipulable with your hands, and also digitally computable.
There’s something visceral about having something both directly manipulable with your hands, and also digitally computable.
https://twitter.com/astralwave/status/1800253486485168514
• • •
Missing some Tweet in this thread? You can try to
force a refresh
