"The Presentation Layer: Brand and UI. While I suspect trying to control the presentation layer long-term is futile (as users will eventually bring their own “interface agents” to interact with your data), for now, it remains a differentiator."
Isn't the point of the UI to remove the mental load from the user of having to deeply understand the business case?
I'm much more efficient with CLIs than UI's, but that's only because I deeply understand operating systems. Learning new CLI's is hard because it's learning a whole new set of commands and concepts. The point of a static UI is that I'm able to learn incrementally.
I like the idea of dynamic UI's but only for things that I know very well, and I'm technically minded. The idea of having to work with new systems with a "customized UI" seems counter-productive. I want a static UI someone else has experience with. This idea of "the UI adapts the data to you" always seemed weird to me because like, I'm not nearly smart enough for that. I hate when Apple changes the color of an iPhone button, imagine every UI I work with being different every time I try to use it.
Also, doesn't this mean I'm constantly a beginner and never get good at my tools? Reliable UI's allow me to turn my brain off. Does everyone else in the world secretly think that they wish things changed more, and I'm the only one too dumb to want this?
> Isn't the point of the UI to remove the mental load from the user of having to deeply understand the business case? ... This idea of "the UI adapts the data to you" always seemed weird to me ... Also, doesn't this mean I'm constantly a beginner and never get good at my tools?
By using your own “interface agents” you choose and control the interface. This doesn't mean it constantly changes or requires you to deeply understand what all the inputs should be. As you can choose what the interface looks like and whether inputs (if any) should be explicit (exact guidance for how to solve a problem) or implicit goals (you just state a problem you have).
What I'm proposing here is actually far more consistent than software today -- you manage/pick your own interface for all the various data/actions that SaaS tools are providing.
So really the question is: what is the ideal interface for you to interact with all the tools and data you have? If that's not a constantly changing dynamic prompt-based interface (which isn't for me and most people, although might be a bad stage 3 impl for some apps) then that's not what it will look like.
> we haven’t figured out the consistent native interface yet.
This line is doing a lot of heavy lifting. Curious to hear you say more about this. Are you looking towards Ink & Switch, or Cursor, or from creatives like Bret Victor...? Are you willing to try novel interfaces, or is there more weight on the "consistent" aspect?
I see it as a gradual shift towards familiarity and consistency (less so than some magic new UI/UX discovery). Could be Claude Cowork style or in a more extreme case like an embodied personal assistant.
Cursor offering to open a browser window inside the ide! That's the most forward thing I've seen yet. Do you think it's a waste to attempt a novel UI/UX? or is it simply that things don't happen like the iPhone anymore?
It's not a waste of time -- I'm just claiming that the UI that we use 5-10 years from now *might* actually not be that different from ones today. Although I'm not necessarily betting on this or claiming this is the case in the post.
The stages framework is clean but Stage 4 assumes the data layer stays clean and structured enough for agents to reliably ingest. Most enterprise data is a mess of inconsistent schemas and undocumented assumptions that took humans years to navigate. The jump from Stage 3 (AI as feature) to Stage 4 (AI as product) isn't just technical - its organizational, becuase companies that cant even document their own processes probably wont be able to scaffold them for agentic access either.
> The jump from Stage 3 (AI as feature) to Stage 4 (AI as product) isn't just technical - its organizational, becuase companies that cant even document their own processes probably wont be able to scaffold them for agentic access either.
Yes! *Today* this is one of the hardest parts of making stage 4 work and if these data isn't clean enough for an agent to ingest, it's pretty useless.
I don't see this as a long-term issue though as companies are market pressured to present data in an organized way (a stage 4 product is 10x+ more useful than a stage 3 one) and the core models themselves continue to get better at dealing with messy, underdocumented, unstructured data (e.g. how well Opus 4.5 now performs in non-agentic-optimized codebases vs past models).
It seemed like a troll comment? Did you actually have a point you wanted to convey?
Here was your original comment if it helps figure it out:
> Have fun with that. Im beginning to think my choice to not use these agentic tools is giving me a huge advantage. Have fun prompting and maintaining two dozen black boxes. I’d much rather spend whatever arbitrary amount of time building something that works forever and is scalable because I designed it well. Also I enjoy coding more than anything else in the world, I really dont understand you people.
"The Presentation Layer: Brand and UI. While I suspect trying to control the presentation layer long-term is futile (as users will eventually bring their own “interface agents” to interact with your data), for now, it remains a differentiator."
Isn't the point of the UI to remove the mental load from the user of having to deeply understand the business case?
I'm much more efficient with CLIs than UI's, but that's only because I deeply understand operating systems. Learning new CLI's is hard because it's learning a whole new set of commands and concepts. The point of a static UI is that I'm able to learn incrementally.
I like the idea of dynamic UI's but only for things that I know very well, and I'm technically minded. The idea of having to work with new systems with a "customized UI" seems counter-productive. I want a static UI someone else has experience with. This idea of "the UI adapts the data to you" always seemed weird to me because like, I'm not nearly smart enough for that. I hate when Apple changes the color of an iPhone button, imagine every UI I work with being different every time I try to use it.
Also, doesn't this mean I'm constantly a beginner and never get good at my tools? Reliable UI's allow me to turn my brain off. Does everyone else in the world secretly think that they wish things changed more, and I'm the only one too dumb to want this?
Hi Mik!
> Isn't the point of the UI to remove the mental load from the user of having to deeply understand the business case? ... This idea of "the UI adapts the data to you" always seemed weird to me ... Also, doesn't this mean I'm constantly a beginner and never get good at my tools?
By using your own “interface agents” you choose and control the interface. This doesn't mean it constantly changes or requires you to deeply understand what all the inputs should be. As you can choose what the interface looks like and whether inputs (if any) should be explicit (exact guidance for how to solve a problem) or implicit goals (you just state a problem you have).
What I'm proposing here is actually far more consistent than software today -- you manage/pick your own interface for all the various data/actions that SaaS tools are providing.
So really the question is: what is the ideal interface for you to interact with all the tools and data you have? If that's not a constantly changing dynamic prompt-based interface (which isn't for me and most people, although might be a bad stage 3 impl for some apps) then that's not what it will look like.
> we haven’t figured out the consistent native interface yet.
This line is doing a lot of heavy lifting. Curious to hear you say more about this. Are you looking towards Ink & Switch, or Cursor, or from creatives like Bret Victor...? Are you willing to try novel interfaces, or is there more weight on the "consistent" aspect?
I see it as a gradual shift towards familiarity and consistency (less so than some magic new UI/UX discovery). Could be Claude Cowork style or in a more extreme case like an embodied personal assistant.
Tell me where I'm off; I'm surprised you are not referring to a novel HCI;
So more akin to like Cmd+Z being "undo"?
Or like pull-to-refresh?
Or (where it all started) tab-completion?
... not because someone designed it cleverly, but because it becomes ubiquitous?
Those feel a bit narrow (a UI/UX component vs an interface) but sort of yes.
I think the winner of "familiarity and consistency" will prob be a mix of cleverness and market share.
Cursor offering to open a browser window inside the ide! That's the most forward thing I've seen yet. Do you think it's a waste to attempt a novel UI/UX? or is it simply that things don't happen like the iPhone anymore?
It's not a waste of time -- I'm just claiming that the UI that we use 5-10 years from now *might* actually not be that different from ones today. Although I'm not necessarily betting on this or claiming this is the case in the post.
The stages framework is clean but Stage 4 assumes the data layer stays clean and structured enough for agents to reliably ingest. Most enterprise data is a mess of inconsistent schemas and undocumented assumptions that took humans years to navigate. The jump from Stage 3 (AI as feature) to Stage 4 (AI as product) isn't just technical - its organizational, becuase companies that cant even document their own processes probably wont be able to scaffold them for agentic access either.
> The jump from Stage 3 (AI as feature) to Stage 4 (AI as product) isn't just technical - its organizational, becuase companies that cant even document their own processes probably wont be able to scaffold them for agentic access either.
Yes! *Today* this is one of the hardest parts of making stage 4 work and if these data isn't clean enough for an agent to ingest, it's pretty useless.
I don't see this as a long-term issue though as companies are market pressured to present data in an organized way (a stage 4 product is 10x+ more useful than a stage 3 one) and the core models themselves continue to get better at dealing with messy, underdocumented, unstructured data (e.g. how well Opus 4.5 now performs in non-agentic-optimized codebases vs past models).
Wild, do you really delete comments that are critical of your approaches? Weird. You LLM shillers are the fakest of the fake.
It seemed like a troll comment? Did you actually have a point you wanted to convey?
Here was your original comment if it helps figure it out:
> Have fun with that. Im beginning to think my choice to not use these agentic tools is giving me a huge advantage. Have fun prompting and maintaining two dozen black boxes. I’d much rather spend whatever arbitrary amount of time building something that works forever and is scalable because I designed it well. Also I enjoy coding more than anything else in the world, I really dont understand you people.