StarZero All shows Try the podcast clipper →

Oct 22, 2023 · 37m · another-podcast

Bundling/Unbundling AI

Benedict Evans · 28m spoken Toni Cowan-Brown · 5m spoken
0:00 / 0:00

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this podcast discussion, Benedict Evans and Toni Caron-Brown explore how large language models are evolving beyond general-purpose chat boxes into specialized vertical applications. By examining historical software cycles and interface design constraints, they illustrate why AI must unbundle into task-specific workflows and invisible utilities.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 83.6% of the talking time here. How this is scored →

The hosts as informed peer 8.0 Guest teaching 1.2 Guest disagreement 1.2 The hosts pushing back 1.6
05100:0010:0020:0030:004:18–9:17 · The hosts as informed peer 8/10 The Interface Dilemma: Blank Canvases and Infinite Interns Evans lays out the foundational framing that LLMs present a blank canvas problem similar to early PC applications, using his 'infinite intern' and 'battleships' analogies. Caron-Brown acts collaboratively, affirming the concepts with her own examples of Notion templates.9:17–15:28 · The hosts as informed peer 8/10 Prompt Engineering, Command Lines, and Software Unbundling Cycles Evans deconstructs prompt engineering as essentially a regression to command-line interfaces and details the recurring historical software cycles of bundling and unbundling. Caron-Brown adds observations about user feature adoption, maintaining a supportive peer dynamic.15:29–24:13 · The hosts as informed peer 8/10 Discovering Use Cases: From PC History to Current AI Adoption Evans illustrates early technology discovery curves using historical examples of 1970s PC marketing, Japanese Excel word processing, and Deloitte survey data. Caron-Brown validates this by sharing her personal experience using specialized niche AI tools instead of generalized ChatGPT prompts.24:14–31:49 · The hosts as informed peer 8/10 Vertical Enterprise Solutions Versus Thin API Wrappers Evans breaks down enterprise AI adoption, contrasting deep vertical solutions for legal discovery with thin API wrappers. Caron-Brown draws on her background selling campaign software to reinforce how enterprise customers resist complexity in favor of familiar spreadsheets.31:49–37:44 · The hosts as informed peer 8/10 Invisible Machine Learning and the Limits of Conversational Interfaces Evans explores invisible machine learning features in smartphones and draws a parallel between conversational UI skepticism and the historical limitations of pen computing. Caron-Brown offers a brief nuance regarding specialized artistic use cases before agreeing on the broader thesis.4:18–9:17 · Guest teaching 1/10 The Interface Dilemma: Blank Canvases and Infinite Interns Evans lays out the foundational framing that LLMs present a blank canvas problem similar to early PC applications, using his 'infinite intern' and 'battleships' analogies. Caron-Brown acts collaboratively, affirming the concepts with her own examples of Notion templates.9:17–15:28 · Guest teaching 1/10 Prompt Engineering, Command Lines, and Software Unbundling Cycles Evans deconstructs prompt engineering as essentially a regression to command-line interfaces and details the recurring historical software cycles of bundling and unbundling. Caron-Brown adds observations about user feature adoption, maintaining a supportive peer dynamic.15:29–24:13 · Guest teaching 1/10 Discovering Use Cases: From PC History to Current AI Adoption Evans illustrates early technology discovery curves using historical examples of 1970s PC marketing, Japanese Excel word processing, and Deloitte survey data. Caron-Brown validates this by sharing her personal experience using specialized niche AI tools instead of generalized ChatGPT prompts.24:14–31:49 · Guest teaching 2/10 Vertical Enterprise Solutions Versus Thin API Wrappers Evans breaks down enterprise AI adoption, contrasting deep vertical solutions for legal discovery with thin API wrappers. Caron-Brown draws on her background selling campaign software to reinforce how enterprise customers resist complexity in favor of familiar spreadsheets.31:49–37:44 · Guest teaching 1/10 Invisible Machine Learning and the Limits of Conversational Interfaces Evans explores invisible machine learning features in smartphones and draws a parallel between conversational UI skepticism and the historical limitations of pen computing. Caron-Brown offers a brief nuance regarding specialized artistic use cases before agreeing on the broader thesis.4:18–9:17 · Guest disagreement 1/10 The Interface Dilemma: Blank Canvases and Infinite Interns Evans lays out the foundational framing that LLMs present a blank canvas problem similar to early PC applications, using his 'infinite intern' and 'battleships' analogies. Caron-Brown acts collaboratively, affirming the concepts with her own examples of Notion templates.9:17–15:28 · Guest disagreement 1/10 Prompt Engineering, Command Lines, and Software Unbundling Cycles Evans deconstructs prompt engineering as essentially a regression to command-line interfaces and details the recurring historical software cycles of bundling and unbundling. Caron-Brown adds observations about user feature adoption, maintaining a supportive peer dynamic.15:29–24:13 · Guest disagreement 1/10 Discovering Use Cases: From PC History to Current AI Adoption Evans illustrates early technology discovery curves using historical examples of 1970s PC marketing, Japanese Excel word processing, and Deloitte survey data. Caron-Brown validates this by sharing her personal experience using specialized niche AI tools instead of generalized ChatGPT prompts.24:14–31:49 · Guest disagreement 1/10 Vertical Enterprise Solutions Versus Thin API Wrappers Evans breaks down enterprise AI adoption, contrasting deep vertical solutions for legal discovery with thin API wrappers. Caron-Brown draws on her background selling campaign software to reinforce how enterprise customers resist complexity in favor of familiar spreadsheets.31:49–37:44 · Guest disagreement 2/10 Invisible Machine Learning and the Limits of Conversational Interfaces Evans explores invisible machine learning features in smartphones and draws a parallel between conversational UI skepticism and the historical limitations of pen computing. Caron-Brown offers a brief nuance regarding specialized artistic use cases before agreeing on the broader thesis.4:18–9:17 · The hosts pushing back 2/10 The Interface Dilemma: Blank Canvases and Infinite Interns Evans lays out the foundational framing that LLMs present a blank canvas problem similar to early PC applications, using his 'infinite intern' and 'battleships' analogies. Caron-Brown acts collaboratively, affirming the concepts with her own examples of Notion templates.9:17–15:28 · The hosts pushing back 2/10 Prompt Engineering, Command Lines, and Software Unbundling Cycles Evans deconstructs prompt engineering as essentially a regression to command-line interfaces and details the recurring historical software cycles of bundling and unbundling. Caron-Brown adds observations about user feature adoption, maintaining a supportive peer dynamic.15:29–24:13 · The hosts pushing back 1/10 Discovering Use Cases: From PC History to Current AI Adoption Evans illustrates early technology discovery curves using historical examples of 1970s PC marketing, Japanese Excel word processing, and Deloitte survey data. Caron-Brown validates this by sharing her personal experience using specialized niche AI tools instead of generalized ChatGPT prompts.24:14–31:49 · The hosts pushing back 1/10 Vertical Enterprise Solutions Versus Thin API Wrappers Evans breaks down enterprise AI adoption, contrasting deep vertical solutions for legal discovery with thin API wrappers. Caron-Brown draws on her background selling campaign software to reinforce how enterprise customers resist complexity in favor of familiar spreadsheets.31:49–37:44 · The hosts pushing back 2/10 Invisible Machine Learning and the Limits of Conversational Interfaces Evans explores invisible machine learning features in smartphones and draws a parallel between conversational UI skepticism and the historical limitations of pen computing. Caron-Brown offers a brief nuance regarding specialized artistic use cases before agreeing on the broader thesis.

speaking balance: gold is the hosts, purple is the guest (3 minute bins)

0:00 · the hosts 92.2% · guest 7.8%0:00 · the hosts 92.2% · guest 7.8%3:00 · the hosts 74% · guest 26%3:00 · the hosts 74% · guest 26%6:00 · the hosts 85.6% · guest 14.4%6:00 · the hosts 85.6% · guest 14.4%9:00 · the hosts 95% · guest 5%9:00 · the hosts 95% · guest 5%12:00 · the hosts 79.8% · guest 20.2%12:00 · the hosts 79.8% · guest 20.2%15:00 · the hosts 75.5% · guest 24.5%15:00 · the hosts 75.5% · guest 24.5%18:00 · the hosts 94.3% · guest 5.7%18:00 · the hosts 94.3% · guest 5.7%21:00 · the hosts 62.3% · guest 37.7%21:00 · the hosts 62.3% · guest 37.7%24:00 · the hosts 88.1% · guest 11.9%24:00 · the hosts 88.1% · guest 11.9%27:00 · the hosts 81.5% · guest 18.5%27:00 · the hosts 81.5% · guest 18.5%30:00 · the hosts 88.7% · guest 11.3%30:00 · the hosts 88.7% · guest 11.3%33:00 · the hosts 83.4% · guest 16.6%33:00 · the hosts 83.4% · guest 16.6%36:00 · the hosts 88.5% · guest 11.5%36:00 · the hosts 88.5% · guest 11.5%
Sharpest disagreement ▶ 36:39 Pushback on pen utility

Caron-Brown lightly challenges Evans' dismissal of stylus interfaces by arguing that they remain necessary for specialized creative professionals like artists and illustrators.

 Hardest push from the hosts ▶ 36:49 Refusal of conversational interface framing

Evans firmly insists that analog skeuomorphism fails when applied to workflows like email and calendar management, holding ground against chat as a universal interface.

 Biggest teaching moment ▶ 27:43 Enterprise software sales reality

Caron-Brown educates Evans on the practical resistance enterprise software salespeople encounter when trying to persuade clients to move away from simple spreadsheets.

 The host holds their own ▶ 18:20 Early personal computer adoption history

Evans demonstrates deep historical tech expertise by explaining how personal computers lacked clear mass-market utility until the arrival of the web.
the scores for every segment, with the reasoning behind each
ChapterTopicThe hosts as informed peerGuest teachingGuest disagreementThe hosts pushing backWhy
The Interface Dilemma: Blank Canvases and Infinite Interns 8112 Evans lays out the foundational framing that LLMs present a blank canvas problem similar to early PC applications, using his 'infinite intern' and 'battleships' analogies. Caron-Brown acts collaboratively, affirming the concepts with her own examples of Notion templates.
Prompt Engineering, Command Lines, and Software Unbundling Cycles 8112 Evans deconstructs prompt engineering as essentially a regression to command-line interfaces and details the recurring historical software cycles of bundling and unbundling. Caron-Brown adds observations about user feature adoption, maintaining a supportive peer dynamic.
Discovering Use Cases: From PC History to Current AI Adoption 8111 Evans illustrates early technology discovery curves using historical examples of 1970s PC marketing, Japanese Excel word processing, and Deloitte survey data. Caron-Brown validates this by sharing her personal experience using specialized niche AI tools instead of generalized ChatGPT prompts.
Vertical Enterprise Solutions Versus Thin API Wrappers 8211 Evans breaks down enterprise AI adoption, contrasting deep vertical solutions for legal discovery with thin API wrappers. Caron-Brown draws on her background selling campaign software to reinforce how enterprise customers resist complexity in favor of familiar spreadsheets.
Invisible Machine Learning and the Limits of Conversational Interfaces 8122 Evans explores invisible machine learning features in smartphones and draws a parallel between conversational UI skepticism and the historical limitations of pen computing. Caron-Brown offers a brief nuance regarding specialized artistic use cases before agreeing on the broader thesis.

Statements from this episode (5)

Insight
Evans: Machine learning gives you 'infinite interns'
“The way I always used to talk about machine learning is it gives you infinite interns. You can get an intern to do something. Anything you can get an intern to do, machine learning can do for you know.”
Benedict Evans Oct 22, 2023 ▶ 6:52
Insight
Evans: AI prompts are just a return to the command line
“And so this point is a command line again. And this is Bill Gates's point that GUIs replace command lines. Well, you get, you've just given me a command line now. That's what a prompt is.”
Benedict Evans Oct 22, 2023 ▶ 10:19
Assertion Supported
Evans: Most Excel spreadsheets are just tables without formulas
“Old observation that most spreadsheets don't have any formula. Most people use Excel to make tables, not to make, not for financial calculations.”
Benedict Evans Oct 22, 2023 ▶ 17:54
Prediction Not checkable as stated
Evans: AI wrappers will evolve into specialized vertical software
“What will happen over time is that this will, that those will become part of much richer, much more vertical, much more specialized pieces of software.”
Benedict Evans Oct 22, 2023 ▶ 29:28
Insight
Evans: Chatbots are the wrong interface for many use cases
“I mean, there's a bunch of use cases where it doesn't matter how good the model is, a chatbot is just not the right interface.”
Benedict Evans Oct 22, 2023 ▶ 35:33
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 100 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more