Geoffrey Litt
@geoffreylitt.com
Malleable software @inkandswitch.com
prev PhD @csail.mit.edu
geoffreylitt.com
🇯🇵🇺🇸
prev PhD @csail.mit.edu
geoffreylitt.com
🇯🇵🇺🇸
you can also write any spreadsheet formula you want and use the result as a filter 😎
and, you can filter on *outputs* and filter "backwards" showing what inputs are compatible w/ that output!
eg: in this video we see what we can afford while staying under budget:
5/
and, you can filter on *outputs* and filter "backwards" showing what inputs are compatible w/ that output!
eg: in this video we see what we can afford while staying under budget:
5/
April 22, 2025 at 9:00 PM
you can also write any spreadsheet formula you want and use the result as a filter 😎
and, you can filter on *outputs* and filter "backwards" showing what inputs are compatible w/ that output!
eg: in this video we see what we can afford while staying under budget:
5/
and, you can filter on *outputs* and filter "backwards" showing what inputs are compatible w/ that output!
eg: in this video we see what we can afford while staying under budget:
5/
this starts very simple: just check a box to focus down on a subset of the values in your sheet.
eg: "if i lease the cheaper car, what might my budget total look like"?
but, because we're in an open-ended spreadsheet, it can also get more interesting than that... 4/
eg: "if i lease the cheaper car, what might my budget total look like"?
but, because we're in an open-ended spreadsheet, it can also get more interesting than that... 4/
April 22, 2025 at 9:00 PM
this starts very simple: just check a box to focus down on a subset of the values in your sheet.
eg: "if i lease the cheaper car, what might my budget total look like"?
but, because we're in an open-ended spreadsheet, it can also get more interesting than that... 4/
eg: "if i lease the cheaper car, what might my budget total look like"?
but, because we're in an open-ended spreadsheet, it can also get more interesting than that... 4/
one way to explore a space is *filters*.
we've all seen filters on shopping websites: you can use checkboxes, range sliders, etc to narrow down a list.
we can borrow from that familiar paradigm and bring it into our ambsheet. 3/
we've all seen filters on shopping websites: you can use checkboxes, range sliders, etc to narrow down a list.
we can borrow from that familiar paradigm and bring it into our ambsheet. 3/
April 22, 2025 at 9:00 PM
one way to explore a space is *filters*.
we've all seen filters on shopping websites: you can use checkboxes, range sliders, etc to narrow down a list.
we can borrow from that familiar paradigm and bring it into our ambsheet. 3/
we've all seen filters on shopping websites: you can use checkboxes, range sliders, etc to narrow down a list.
we can borrow from that familiar paradigm and bring it into our ambsheet. 3/
what if you could model a problem in a spreadsheet, and then instantly start applying powerful *filters* to explore your space of options?
that's the idea @alexwarth.bsky.social and I have been exploring recently at @inkandswitch.com ... 1/
that's the idea @alexwarth.bsky.social and I have been exploring recently at @inkandswitch.com ... 1/
April 22, 2025 at 9:00 PM
what if you could model a problem in a spreadsheet, and then instantly start applying powerful *filters* to explore your space of options?
that's the idea @alexwarth.bsky.social and I have been exploring recently at @inkandswitch.com ... 1/
that's the idea @alexwarth.bsky.social and I have been exploring recently at @inkandswitch.com ... 1/
The Deep Research Printer:
Say any question, get a custom printed report.
The calmness of paper. The meditative quality of *actually reading*, not spawning a million tabs.
Combined with the vast knowledge of the Web, answering our most niche curiosities.
Say any question, get a custom printed report.
The calmness of paper. The meditative quality of *actually reading*, not spawning a million tabs.
Combined with the vast knowledge of the Web, answering our most niche curiosities.
March 23, 2025 at 9:44 PM
The Deep Research Printer:
Say any question, get a custom printed report.
The calmness of paper. The meditative quality of *actually reading*, not spawning a million tabs.
Combined with the vast knowledge of the Web, answering our most niche curiosities.
Say any question, get a custom printed report.
The calmness of paper. The meditative quality of *actually reading*, not spawning a million tabs.
Combined with the vast knowledge of the Web, answering our most niche curiosities.
In school why do teachers read your work?
“In school… you’re attaching a $100 bill to your paper, handing it to your instructor, and saying hey will you read this…”
Seriously just go watch the talk it’s so good
“In school… you’re attaching a $100 bill to your paper, handing it to your instructor, and saying hey will you read this…”
Seriously just go watch the talk it’s so good
March 20, 2025 at 10:11 PM
In school why do teachers read your work?
“In school… you’re attaching a $100 bill to your paper, handing it to your instructor, and saying hey will you read this…”
Seriously just go watch the talk it’s so good
“In school… you’re attaching a $100 bill to your paper, handing it to your instructor, and saying hey will you read this…”
Seriously just go watch the talk it’s so good
“In the real world - not in school - the purpose of your text is to change what readers think about the world”
So simple, so good
So simple, so good
March 20, 2025 at 9:59 PM
“In the real world - not in school - the purpose of your text is to change what readers think about the world”
So simple, so good
So simple, so good
# avoid the nightmare bicycle
March 3, 2025 at 10:31 PM
# avoid the nightmare bicycle
15/ Anyway, that's the lens I'm finding helpful these days. tldr: in the wild west, speed wins, and AI is fast. So we can thoughtfully apply AI to make better prototypes.
If you're using similar techniques would be curious how you think about it.
If you're using similar techniques would be curious how you think about it.
February 12, 2025 at 3:28 PM
15/ Anyway, that's the lens I'm finding helpful these days. tldr: in the wild west, speed wins, and AI is fast. So we can thoughtfully apply AI to make better prototypes.
If you're using similar techniques would be curious how you think about it.
If you're using similar techniques would be curious how you think about it.
14/ As the models improve, it's even feeling like the combo is possible sometimes: move faster than ever before, at higher quality.
I'm cautious with the super high speeds though: preserving creative agency does require a lot of iteration and pondering the essential problem.
I'm cautious with the super high speeds though: preserving creative agency does require a lot of iteration and pondering the essential problem.
February 12, 2025 at 3:28 PM
14/ As the models improve, it's even feeling like the combo is possible sometimes: move faster than ever before, at higher quality.
I'm cautious with the super high speeds though: preserving creative agency does require a lot of iteration and pondering the essential problem.
I'm cautious with the super high speeds though: preserving creative agency does require a lot of iteration and pondering the essential problem.
12/ OK, but what about the UI itself? am I abdicating too much responsibility to the AI?
Perhaps. I do worry about this a lot... But I actually think the opposite may be true? I can be more attentive to UI detail when I'm not furiously slinging around code.
Perhaps. I do worry about this a lot... But I actually think the opposite may be true? I can be more attentive to UI detail when I'm not furiously slinging around code.
February 12, 2025 at 3:28 PM
12/ OK, but what about the UI itself? am I abdicating too much responsibility to the AI?
Perhaps. I do worry about this a lot... But I actually think the opposite may be true? I can be more attentive to UI detail when I'm not furiously slinging around code.
Perhaps. I do worry about this a lot... But I actually think the opposite may be true? I can be more attentive to UI detail when I'm not furiously slinging around code.
11/ This is not one sentence prompts. It's long detailed prompts, iteration, code review. In the weeds!
I can make a proper plan for how something should work... and then the AI executes it in minutes! So we're still moving fast enough for prototyping.
I can make a proper plan for how something should work... and then the AI executes it in minutes! So we're still moving fast enough for prototyping.
February 12, 2025 at 3:28 PM
11/ This is not one sentence prompts. It's long detailed prompts, iteration, code review. In the weeds!
I can make a proper plan for how something should work... and then the AI executes it in minutes! So we're still moving fast enough for prototyping.
I can make a proper plan for how something should work... and then the AI executes it in minutes! So we're still moving fast enough for prototyping.
10/ There's a second move I find more subtle and interesting:
While maintaining my previous speed of prototyping, I can build at much higher quality.
While maintaining my previous speed of prototyping, I can build at much higher quality.
February 12, 2025 at 3:28 PM
10/ There's a second move I find more subtle and interesting:
While maintaining my previous speed of prototyping, I can build at much higher quality.
While maintaining my previous speed of prototyping, I can build at much higher quality.
9/ There's an obvious here: MOVE FAST 🏎️. One sentence prompt, make an app, boom you're done!
This makes for fun demos and can be useful for random throwaway tools...
But not useful at all when I'm trying to prototype a good UI that hasn't been made before.
This makes for fun demos and can be useful for random throwaway tools...
But not useful at all when I'm trying to prototype a good UI that hasn't been made before.
February 12, 2025 at 3:28 PM
9/ There's an obvious here: MOVE FAST 🏎️. One sentence prompt, make an app, boom you're done!
This makes for fun demos and can be useful for random throwaway tools...
But not useful at all when I'm trying to prototype a good UI that hasn't been made before.
This makes for fun demos and can be useful for random throwaway tools...
But not useful at all when I'm trying to prototype a good UI that hasn't been made before.
8/ From my personal experience, AI *significantly* pushes out the speed-quality frontier for rapidly prototyping web UIs in React.
(I mainly use Cursor Composer w/ 3.5-sonnet, plus a smattering of other tools)
The question is: what do we do with this power?
(I mainly use Cursor Composer w/ 3.5-sonnet, plus a smattering of other tools)
The question is: what do we do with this power?
February 12, 2025 at 3:28 PM
8/ From my personal experience, AI *significantly* pushes out the speed-quality frontier for rapidly prototyping web UIs in React.
(I mainly use Cursor Composer w/ 3.5-sonnet, plus a smattering of other tools)
The question is: what do we do with this power?
(I mainly use Cursor Composer w/ 3.5-sonnet, plus a smattering of other tools)
The question is: what do we do with this power?
7/ Let's first address the 💼 Land of proper engineering.
Maybe AI helps you move faster at constant quality? Maybe your codebase turns to slop and things are terrible?
Honestly I have no idea! Not the topic here. Let's move on to the wild west where I typically operate.
Maybe AI helps you move faster at constant quality? Maybe your codebase turns to slop and things are terrible?
Honestly I have no idea! Not the topic here. Let's move on to the wild west where I typically operate.
February 12, 2025 at 3:28 PM
7/ Let's first address the 💼 Land of proper engineering.
Maybe AI helps you move faster at constant quality? Maybe your codebase turns to slop and things are terrible?
Honestly I have no idea! Not the topic here. Let's move on to the wild west where I typically operate.
Maybe AI helps you move faster at constant quality? Maybe your codebase turns to slop and things are terrible?
Honestly I have no idea! Not the topic here. Let's move on to the wild west where I typically operate.
6/ ok, now we're ready to discuss AI.
The green line is my extremely scientific estimate of how AI support moves the boundary today.
The green line is my extremely scientific estimate of how AI support moves the boundary today.
February 12, 2025 at 3:28 PM
6/ ok, now we're ready to discuss AI.
The green line is my extremely scientific estimate of how AI support moves the boundary today.
The green line is my extremely scientific estimate of how AI support moves the boundary today.
5/ The thing is...code quality still matters out here!
A well-built prototype often provides better signal by holding up to real use, and makes it easier to try out new ideas.
I think a lot about surfing the speed/quality tradeoff here and trying to push the boundary.
A well-built prototype often provides better signal by holding up to real use, and makes it easier to try out new ideas.
I think a lot about surfing the speed/quality tradeoff here and trying to push the boundary.
February 12, 2025 at 3:28 PM
5/ The thing is...code quality still matters out here!
A well-built prototype often provides better signal by holding up to real use, and makes it easier to try out new ideas.
I think a lot about surfing the speed/quality tradeoff here and trying to push the boundary.
A well-built prototype often provides better signal by holding up to real use, and makes it easier to try out new ideas.
I think a lot about surfing the speed/quality tradeoff here and trying to push the boundary.
4/ But personally... I'm usually out in the 🤠 Wild west of UI prototyping! (I research tools for thought and programming interfaces.)
Here there's a min *speed*: try many ideas, discard most. Which implies a quality *ceiling* - no time to engineer perfectly.
Here there's a min *speed*: try many ideas, discard most. Which implies a quality *ceiling* - no time to engineer perfectly.
February 12, 2025 at 3:28 PM
4/ But personally... I'm usually out in the 🤠 Wild west of UI prototyping! (I research tools for thought and programming interfaces.)
Here there's a min *speed*: try many ideas, discard most. Which implies a quality *ceiling* - no time to engineer perfectly.
Here there's a min *speed*: try many ideas, discard most. Which implies a quality *ceiling* - no time to engineer perfectly.
3/ if you're in the 💼 Land of proper engineering (production engineering), you're gonna impose some minimum quality bar, which in turn imposes a speed limit.
February 12, 2025 at 3:28 PM
3/ if you're in the 💼 Land of proper engineering (production engineering), you're gonna impose some minimum quality bar, which in turn imposes a speed limit.
2/ first, some preliminaries before we talk AI:
for anyone building software, there's some tradeoff between speed and code quality. (obviously you can work to push the frontier, but the tradeoff exists)
for anyone building software, there's some tradeoff between speed and code quality. (obviously you can work to push the frontier, but the tradeoff exists)
February 12, 2025 at 3:28 PM
2/ first, some preliminaries before we talk AI:
for anyone building software, there's some tradeoff between speed and code quality. (obviously you can work to push the frontier, but the tradeoff exists)
for anyone building software, there's some tradeoff between speed and code quality. (obviously you can work to push the frontier, but the tradeoff exists)
1/ here's a visual model for how AI support makes me better at UI prototyping
with the surprising conclusion that AI may *improve* code quality in some specific cases!
lets walk thru it step by step...
with the surprising conclusion that AI may *improve* code quality in some specific cases!
lets walk thru it step by step...
February 12, 2025 at 3:28 PM
1/ here's a visual model for how AI support makes me better at UI prototyping
with the surprising conclusion that AI may *improve* code quality in some specific cases!
lets walk thru it step by step...
with the surprising conclusion that AI may *improve* code quality in some specific cases!
lets walk thru it step by step...
still learning how to use reasoning models, but i tried to give a good set of constraints on the answer here, and also made sure to agree on a plan before coding. (helped me catch a possible bug ahead of time!)
February 11, 2025 at 6:14 PM
still learning how to use reasoning models, but i tried to give a good set of constraints on the answer here, and also made sure to agree on a plan before coding. (helped me catch a possible bug ahead of time!)
when you're tired after a work day, it really helps to have tasks broken down like this:
February 11, 2025 at 3:06 AM
when you're tired after a work day, it really helps to have tasks broken down like this:
The latest Cursor Composer + Sonnet is accelerating me to a ridiculous degree these days—I'm feeling ~2-3x faster at executing major code changes.
Makes trying new ideas feel so much lighter and worth doing.
Makes trying new ideas feel so much lighter and worth doing.
January 24, 2025 at 2:09 AM
The latest Cursor Composer + Sonnet is accelerating me to a ridiculous degree these days—I'm feeling ~2-3x faster at executing major code changes.
Makes trying new ideas feel so much lighter and worth doing.
Makes trying new ideas feel so much lighter and worth doing.