22 November 2022

The Future Of A.I. Businesses With Steph Smith

1Frameworks

13Insights

Frameworks in this episode

Entrepreneurship6 steps

AI Business Moat Test

Test an AI product for durable differentiation before investing or scaling

Read →

Insights & moments

The myth-busts, hot takes, explainers, and tools worth keeping.

Myth Buster· 1

Myth Buster10:00

AI Writing Handles Search Content Better Than Original Points of View

The hosts distinguish formulaic, search-driven writing from content built around a distinctive argument or story. AI can produce much of a conventional instructional article, but its generic output is less useful when the creator's value comes from original judgment and a strong point of view.

Search-driven content follows patterns AI can reproduce
Distinctive thought leadership depends on original judgment
A usable draft is not necessarily insightful content
The right tool depends on the content's purpose

“I think the use case is search driven content and you don't create that, you create point of view content.”
Kieran Flanagan · 10:00

“It actually probably will give you 80% of what you need.”
Kieran Flanagan · 10:30

#content#seo#writing#thought-leadership

Hot Take· 2

Hot Take02:00

Prompt Engineering May Become an Invisible Software Layer

Steph argues that manually composing prompts may be a temporary stage of AI adoption. Applications can infer the required prompt from existing inputs, such as turning a podcast transcript directly into a sharing image without asking the user to translate the task into special instructions.

Manual prompting may not remain a core human skill
Applications can derive prompts from workflow context
Automation can reduce AI interaction to uploading an asset and clicking a button

“I think in, not very long, the computer's just gonna do the prompting for you.”
Steph Smith · 02:00

“the only human requirement there is to just upload it, click a button.”
Steph Smith · 02:30

#prompting#automation#workflows

Hot Take27:00

Most People Will Use AI Without Calling It AI

The group predicts that mainstream adoption will often arrive through familiar products rather than standalone AI destinations. Search, transcription, editing, writing, and design tools will quietly add machine-learning features, while users focus on the improved outcome rather than the underlying technology.

AI already operates inside familiar products
Users value outcomes more than technical labels
Existing platforms can make adoption nearly invisible
Integrated features may spread farther than standalone tools

“a lot of this is integrated into existing products already”
Steph Smith · 27:30

“people using the tool won't be like, oh, this is like a new thing.”
Steph Smith · 27:30

#mainstream-adoption#platforms#product-design#integration

Explainer· 3

Explainer01:00

AI as the Electric Bicycle of the Mind

Steph Smith extends Steve Jobs's bicycle metaphor to explain AI as an intellectual accelerator. Humans still choose the destination, but AI can help them reach it faster and with less effort by drawing on knowledge encoded across the internet.

AI amplifies rather than merely stores human knowledge
The user still supplies intent and direction
The central benefit is faster, lower-effort intellectual work

“I think of AI as like the electric bicycle of the mind.”
Steph Smith · 01:30

“But you can get there so much more quickly and so much more effortlessly.”
Steph Smith · 01:30

#ai#productivity#mental-models

Explainer16:30

Prompt Libraries Turn Private Creative Work Into Shared Infrastructure

The discussion reframes prompts as visible instructions that teams and communities can inspect, reuse, and improve. Unlike creative processes that previously remained in individual minds, prompt libraries expose intermediate work and allow people to build on one another's discoveries.

Prompts reveal instructions behind generated outputs
Teams can maintain proprietary prompt libraries
Public examples shorten experimentation cycles
Shared prompts make AI collaboration more cumulative

“there's gonna be like team level collaboration.”
Steph Smith · 17:00

“our ability to like learn and build upon each other's work is orders of magnitudes higher than it's ever been.”
Kipp Bodnar · 18:30

#prompts#collaboration#knowledge-sharing#teams

Explainer15:00

The Two Stages of AI: Copilot First, Pilot Later

Kieran divides AI development into a safer copilot stage and a more autonomous pilot stage. In the first, humans retain strategy and control while AI assists execution; in the second, AI becomes the primary operator and the human's role diminishes.

Current AI mainly assists human-directed work
Humans still dictate strategy and make decisions
Future systems may take primary operational control
The transition changes responsibility as well as productivity

“AI is the co-pilot, it's helping you do things, but literally you're still kind of in control.”
Kieran Flanagan · 15:00

“at some point it flips to like, AI is the pilot and you're the co-pilot”
Kieran Flanagan · 15:00

#copilot#automation#future-of-work#autonomy

Story· 2

Story11:00

The Performance Review Almost Written Entirely by Jasper

Kipp recounts generating Kieran's performance review with Jasper and nearly submitting it. The result was plausible but shallow, illustrating how AI can create acceptable managerial boilerplate without delivering the specific depth a consequential review deserves.

AI can generate credible workplace prose
Plausibility can conceal a lack of depth
High-stakes feedback still benefits from personal judgment
Human authorship matters when specificity affects another person

“I did a version of your performance review-”
Kipp Bodnar · 11:30

“if you're a mediocre manager, what you would say at a performance review”
Kipp Bodnar · 12:00

#management#writing#jasper#performance-reviews

Story22:00

Why GitHub Copilot Became an Assistant Instead of an Autonomous Coder

Steph explains that Copilot's creators initially explored a more autonomous system that would simply deliver code. Because a small error can make code unusable, they shifted toward suggestions that developers evaluate, preserving human control while still surfacing occasional novel solutions.

Code has a low tolerance for small errors
Autonomous output was less reliable than assisted completion
Suggestions let developers validate each contribution
Imperfect AI remains useful inside a human review loop

“since code is binary, if code is ever even just a little bit wrong, it won't work.”
Steph Smith · 22:30

“instead it just suggests things as you're saying Kieran.”
Steph Smith · 22:30

#github-copilot#coding#human-in-the-loop#software

Q&A· 1

Q&A29:30

Who Gets Paid When AI Learns an Artist's Style?

The episode closes with the example of living artist Greg Rutkowski becoming a highly frequent Stable Diffusion prompt reference. The hosts raise unresolved questions about training on living creators' work and predict that AI markets may eventually develop licensing and revenue-sharing systems similar to music streaming and YouTube.

Generated art can depend heavily on identifiable creators' styles
Living artists create sharper consent and compensation questions
Earlier digital markets evolved toward licensing and revenue sharing
AI businesses may eventually need comparable payment structures

“these AI's are building off of existing people's work.”
Steph Smith · 30:00

“Seems like that's an inevitable part of the future of AI.”
Kipp Bodnar · 31:00

#copyright#artists#licensing#training-data#ethics

Tool· 2

Tool15:30

The Boring AI Use Case Worth Real Money: Finding Documents

Kieran identifies internal document retrieval as an unglamorous but expensive organizational problem. An AI system that understands company knowledge and helps teams locate the right material could recover hours of worker time and provide multiplayer value across an organization.

Employees lose substantial time searching for documents
Internal search affects entire teams rather than one user
Boring workflow problems can support valuable products
Shared knowledge retrieval is a natural enterprise AI application

“one of the biggest problems in all companies is how to find documents.”
Kieran Flanagan · 15:30

“those boring use cases are actually worth a lot of money.”
Kieran Flanagan · 16:00

#enterprise-search#knowledge-management#documents#collaboration

Tool17:30

Reverse-Engineering Better Image Prompts With Lexica Art

Steph describes using Lexica Art to study successful Stable Diffusion prompts after repeatedly failing to generate a convincing DNA helix. Searching for strong examples revealed words and concepts that other creators had used, providing a practical way to debug prompts through precedent.

Search existing outputs related to the desired image
Inspect prompts behind the strongest examples
Identify concepts omitted from the failed prompt
Use community examples to accelerate experimentation

“you can see the specific prompts and you can search by prompt.”
Steph Smith · 17:30

“I looked at the best ones and say, oh, they integrated this”
Steph Smith · 18:00

#lexica-art#stable-diffusion#image-generation#prompting

Takeaway· 2

Takeaway23:00

AI Innovation Will Move Faster Than Human Habits

Steph contrasts rapid technical progress with the muted response of friends outside technology and Japan's continued use of floppy disks. The lesson is that capability does not create immediate behavioral change: entrenched habits, inertia, and limited interest can delay adoption for years.

Technical excitement is concentrated among early adopters
Mainstream users may not care how a capability works
Existing habits slow migration to better tools
Forecasts should separate innovation speed from adoption speed

“adoption is often much slower than the innovation.”
Steph Smith · 23:30

“human habits are so rigid and inertia is a strong force.”
Steph Smith · 24:00

#adoption#innovation#behavior-change#technology

Takeaway28:00

Prompt-to-Prompt Animation Foreshadows AI Filmmaking

Steph highlights a Replicate demo that animates a transition between two prompts. Its immediate use is less important than what it signals: the progression from text generation to images and then video could let creators implement animations or entire visual stories directly from ideas.

Generative media is progressing from text to images to video
Simple demos can reveal the direction of a technology
Future creators may translate storylines directly into animation
Implementation costs for visual ideas may fall dramatically

“it basically helps you animate from two prompts from one to the other.”
Steph Smith · 28:30

“text to video is coming.”
Steph Smith · 28:30

#text-to-video#animation#replicate#generative-media