Amjad Masad - The Future of Software Creation - [Invest Like the Best, EP.310]

Amjad Masad - The Future of Software Creation - [Invest Like the Best, EP.310]

At a Glance

Replit founder and CEO Amjad Masad discusses the future of software creation, with the core thesis being that current human-computer interaction methods are inefficient. Replit aims to lower the barrier to programming through a browser-based IDE, targeting the "next billion software creators." The most impactful judgment in the episode: Masad believes that in the future, Google Docs, JavaScript files, and ChatGPT-like interfaces will "merge into one"—"in the limit, Google Docs, JavaScript files, and ChatGPT-like interfaces are all the same thing" — implying that the boundaries between programming, writing, and knowledge retrieval will completely blur.

---

I. Steve Jobs’ “Black Pill”: The Computer’s Fall from Creative Tool to Consumption Device

Amjad Masad argues that there is an underexplored dark side to Steve Jobs’ legacy—he made computers so easy to use that people’s relationship with devices has become purely consumptive.

• Historical contrast: In the 1970s–80s, PC users faced a command-line prompt (DOS, Commodore 64) upon booting up, and every professional was “writing programs.” Today, even professionals are constrained by software designers—when hitting a limitation, they can only “complain on Twitter or switch to another app.”
• Concrete example: Want to program an iPhone? First, buy an iPhone for $1,000, then a MacBook for $2,000–$3,000, pay a $100 developer license fee, and finally wait for Apple’s approval to release the software. Masad calls this “Steve Jobs’ black pill”—“our relationship with computers has been fundamentally changed by his influence, and now most of us are consumers of computers.”
• The white pill side: New-generation tools like Notion, Airtable, and Figma are creating “end-user programmable” software, but Masad believes this is not enough—“you need a full programming experience to truly harness the full power of a computer.”

---

2. REPL and Replit: Reducing Programming Friction Through a "Primitives" Mindset

Masad believes the ideal programming environment should be as simple as a notepad, and Replit achieves this through a "primitives" mindset.

• REPL Concept: REPL stands for Read, Eval, Print, Loop — "a single line of code can bootstrap a complete programming environment." Replit was originally born from this minimalist philosophy, "as simple as a notepad."
• Primitives Mindset: Masad emphasizes that computers are fundamentally built from a single NAND gate, and Replit follows the same philosophy — its core primitives include:
• The REPL itself: Accepts user input, evaluates it, and returns results
• Universal package manager: Automatically identifies and installs any library required by the code
• Auto-hosting: Detects when a user is running a web server and deploys it automatically
• The AI Moment Explosion: ChatGPT excels at generating code, but "there is only one place on the internet where you can paste code and run it directly — and that is Replit." Masad notes that after ChatGPT gained popularity, Replit's revenue and usage grew in tandem, "all the primitives we built were ready for this AI moment."

---

3. AI’s Impact on Programming: From "Compiler Thinking" to "Statistical Learning"

Masad has been contemplating AI’s impact on code since the 2010s, believing that machine learning will fundamentally transform software development—just as it has reshaped board games and autonomous driving.

• Historical Context:
• 2017: OpenAI’s "sentiment neuron" paper—unsupervised learning to predict the next word unexpectedly mastered sentiment analysis
• 2019: GPT-2 (1 billion parameters)—"Anyone with imagination who saw GPT-2 would realize the world had fundamentally changed"
• 2020: GPT-3—"in-context learning" (i.e., prompt engineering) emerged, enabling "programming" models via natural language without retraining
• Replit’s Response: Upon seeing GPT-3, Masad immediately concluded, "This is what we need," and began investing in the AI product Ghostwriter (which generates code based on GPT-3-like models), which is now growing extremely fast.
• GPT-4 Expectations: Masad believes GPT-4 will bring a qualitative leap, but the path to commercialization remains unclear—"Costs need to drop by a factor of 1,000 to achieve the economics of a search engine."

---

4. Skill Reshaping in the AI Era: The Return of Creative Talent

Masad argues that the traditional "step-by-step" model of education and work is fading, and creativity and independent thinking will become the most scarce abilities.

• End of the Traditional Model: "In the post-war world, you went to school, did what you were told, learned multiplication tables and wrote essays, followed the steps, and you could have a good life—that world is gone."
• Return of the "Idea People": Masad notes that the tech world has long mocked "people who only come up with ideas," but now—"If you're an idea person, you can use Stable Diffusion and DALL-E to make posters, GPT-3 to write articles, and Replit to build software. Idea people are making a strong comeback."
• Google Interview Transformation: Masad predicts that "within two years, no one will ask traditional Google-style algorithm interview questions"—interviews will return to problem-solving ability, rather than "jumping through hoops."
• Widening Wealth Inequality: Masad acknowledges that these technologies "will accelerate existing disparities," but believes that "the fairer the opportunities and the more honest the market, the more natural the wealth inequality becomes."

---

5. Assessment of Tech Giants: Google's "Gym Moment"

Masad offers differentiated assessments of major tech giants, arguing that Google is not doomed to fail, while Microsoft may be the biggest winner.

Company	Masad's Assessment	Key Argument
Google	Not doomed to fail, has a real opportunity	"Google has been training in the gym every day for the past 15 years, and now is their moment" — Transformer technology was invented at Google (2017), and Google Search has used BERT since 2019; however, costs need to drop by 1,000x for large-scale deployment
Microsoft	Biggest winner	Close ties with OpenAI, Copilot already leads the market, Power Platform will be fully infused with AI, "could be the first company to fully productize AI across its entire suite"
Meta/Facebook	Uncertain	The metaverse is an "interesting bet," but AI's impact is uncertain; potential opportunity: AI virtual influencers could disrupt the social media model
AWS	Neutral	"Will be fine" — many people just use AWS for GPUs, with no particularly large AI bet
NVIDIA	Positive	Training more models requires more GPUs

---

6. Three Visions of AGI and Masad’s Middle Ground

Masad outlines three mainstream views on Artificial General Intelligence (AGI) and offers his own assessment.

• The Pessimists (Eliezer Yudkowsky et al.): AGI will kill everyone—the "alignment problem" is unsolvable. Classic example: a paperclip entrepreneur instructs AGI to "maximize paperclip production," and the AGI turns the entire universe into a paperclip factory.
• The Optimists (Sam Altman/OpenAI): Alignment is possible, and ChatGPT itself can help align the next generation of models—"We will learn a lot before we have AGI."
• Masad’s Middle Ground: "I think raw computer intelligence can become quite high, but until we truly understand the nature of consciousness and human intelligence, we cannot bridge the gap to create human-like agents. We know almost nothing about the brain—expecting to create something that acts like a human without understanding what a human is, is like expecting to get abs without working out."

---

Mentioned Positions

Position	Guest Stance	Key Data
Replit	Bullish (founder's perspective)	Revenue and usage grew alongside ChatGPT's popularity; Bounties project paid $15,000 to creators within 3-4 weeks, with 3x week-over-week growth; mobile app gained 250,000 downloads in India and Brazil
OpenAI	Bullish (infrastructure layer)	GPT-2 (1 billion parameters, 2019), GPT-3 (2020, in-context learning), GPT-4 coming soon; Replit benefited from ChatGPT's popularity
NVIDIA	Bullish	Training more AI models requires more GPUs
Microsoft	Bullish	Copilot leads the market; close relationship with OpenAI; Power Platform to be fully infused with AI
Google	Neutral to positive	Invented Transformer technology (2017); BERT used in search (2019); but costs need to drop 1,000x for large-scale deployment
AWS	Neutral	"Will be fine," no particularly large AI bets
Meta/Facebook	Uncertain	Metaverse is an "interesting bet"; AI virtual influencers could disrupt social media
Notion, Airtable, Figma	Positive (industry trend)	Represent a new generation of tools enabling "end-user programmability"
Perplexity AI	Positive	Interesting AI search engine, "can ask questions, draw images, generate SQL, query Twitter"
Thumbtack (Jonathan Swanson)	Case reference	Founder completed a complex task via Replit Bounties for $20 in 48 hours; the developer was a 14-year-old

---

Judgments Worth Remembering

1. "At the limit, Google Docs, JavaScript files, and ChatGPT-like interfaces are all the same thing" (Amjad Masad) — The boundaries between programming, writing, and knowledge retrieval will completely blur, merging into a unified "knowledge creation" interface.

2. Steve Jobs' "black pill": He made computers so easy to use that people's relationship with devices became purely consumptive (Amjad Masad) — "You spend $1,000 on an iPhone, but to program you need to spend another $2,000–$3,000 on a MacBook, pay $100 for a developer license, and finally wait for Apple's approval."

3. Replit's "primitive" thinking: Just as a single NAND gate builds an entire computer, Replit builds its entire platform from three primitives — REPL, a universal package manager, and auto-hosting (Amjad Masad) — "ChatGPT is good at generating code, but there's only one place on the internet where you can paste code and run it directly — and that's Replit."

4. "Google has been training in the gym every day for the past 15 years; now it's their moment" (Amjad Masad) — Transformer technology was invented at Google (2017), BERT was used for search (2019), but costs needed to drop 1,000x for large-scale deployment; Google is not doomed to fail.

5. "In two years, no one will ask traditional Google-style algorithmic interview questions anymore" (Amjad Masad) — AI will eliminate "hoop-jumping" interviews and return to problem-solving ability; the entire interview prep industry needs fundamental change.

6. "Idea people are making a strong comeback" (Amjad Masad) — "The tech world has long mocked people who only have ideas, but now you can go to Stable Diffusion for posters, GPT-3 for writing, and Replit for software."

7. Masad's middle ground on AGI: "Expecting to create something that acts like a human without understanding what a human is, is like expecting to get abs without working out" (Amjad Masad) — He believes computers can have high raw intelligence, but without understanding consciousness and the nature of human intelligence, the gap cannot be crossed.

8. "The worst thing is to raise children to 'follow the rules'" (Amjad Masad) — "Teach them to think independently, adapt to change, and be different. The post-war model of following a set path to a good life is gone."

9. The lesson from Replit Bounties: A 14-year-old completed a complex AI fine-tuning interface in 48 hours for $20 (Amjad Masad) — "Now 13- and 14-year-olds can create value and earn money, which was impossible before."

10. "AI allows people to effectively command an army" (Amjad Masad) — The most productive people will become "insanely more productive" (e.g., Elon Musk-type individuals could achieve 1,000x efficiency gains), but "the potential black pill is that some companies are building AI systems that are more 'pilot' than 'co-pilot'."