Amjad Masad on Replit, Company Culture and AI-Assisted Software

Replit founder and CEO Amjad Masad joins Harry Stebbings in March 2023 to discuss company building and AI. Masad grew up in Jordan, where he describes himself as a rebellious child who often got into trouble and stood out in school.

Masad recalls selling toys with his brother by buying them cheaply near home and reselling them near their school. He later started a software business as a teenager. Those experiences and his interest in great entrepreneurs and historical figures gave him an early desire to build something exceptional.

He says he looks for signs of unusual drive when hiring, whether in business, athletics or another field. Asked what he is running from, he names mediocrity. He advises Stebbings to be authentic and take risks in personal relationships rather than appeal to everyone; these are his personal views, not a general measure of potential.

Masad describes great startups as small groups with an insight and a shared ambition to change the world. He uses OpenAI as a contemporary example of researchers pursuing language-model scaling before many others paid attention. He compares such groups with historical movements, while saying he would leave it to historians to judge the analogy.

Replit’s own thesis, Masad says, is that software is a superpower too narrowly held and should become accessible to many more people. He compares that ambition to the spread of literacy after the printing press. As Replit’s CEO, he has a direct commercial interest in this vision.

Outside specialist roles, Masad values people who can simply get things done: entrepreneurial business, legal and operations staff who solve problems without letting work fall between teams. He thinks hiring for that disposition matters more than designing processes to make an unwilling person work hard.

Stebbings suggests a generation has lost its appetite for hard work. Masad agrees and offers a broader cultural critique about passive consumption and meaning. This is their subjective view of work and culture, not evidence about an entire generation.

Replit gives candidates material explaining why they might not want to join, including hard work, uncertainty and the possibility that the company may not succeed. Masad says the aim is mutual selection: different people fit different companies, and candidates should understand the demands before accepting.

He recounts Mark Zuckerberg’s heuristic that a meaningful value has a reasonable opposite: ‘move fast and break things’ can be contrasted with a deliberate preference for moving slowly and protecting stability. A value nobody could reasonably disagree with may say little about a company’s actual choices.

One Replit value is ‘seek pain’. Masad means confronting uncomfortable facts, such as a product failing customers or lacking product-market fit, instead of allowing problems to grow. He says the company itself has delayed confronting issues in the past, making them worse later.

Personally, he tries to remind himself that worthwhile outcomes often require discomfort. At team level, he argues that delaying a painful decision can make the eventual consequences more severe. This is his management philosophy, not a claim that pain always produces improvement.

Stebbings raises the difficulty of an early employee no longer fitting an expanded role. Masad says loyalty to one person must be weighed against obligations to colleagues, customers and investors. He sometimes seeks a different role where the person can contribute and retain dignity, while acknowledging that this is difficult and that a replacement or additional leader may be needed.

His own hiring mistake has sometimes been excessive attachment to creative misfits. Early Replit staff included programmers with unconventional backgrounds, which he values, but he says he has occasionally kept a talented person whose behavior harmed teammates for too long. Creativity does not erase the need to work well with others.

Asked whether a misfit culture can scale as the company grows, Masad says he is working through the question and cannot yet give a definitive answer. He hopes unconventional people can continue contributing, but doubts a large organization can consist entirely of misfits.

Masad sketches a history of programming from machine code and assembly through C and scripting languages. Each layer, in his account, let developers work further from hardware and closer to human concepts. His large productivity multiples are illustrative estimates and he notes that others may contest them.

He credits reusable open-source packages with another leap: developers no longer have to write every common function themselves. He argues that AI may begin a new productivity curve after a period without a comparable jump. His historical comparison and forecast are his interpretation.

Masad cites an engineer’s reported AI-written code and says Replit Ghostwriter users had 30% to 50% of code written by AI in the figures he describes. He also cites a GitHub Copilot study reporting a 55% productivity improvement. These refer to different people, tools and measures, not a single industry-wide coding rate; Replit sells an AI coding product.

He sees two ways language models can affect software. One is an assistant embedded in the development environment, such as Copilot or Replit’s planned Ghostwriter chat. The other is using a model as a component of an application—for example, extracting structured information from a website—although reliability and suitability depend on the task.

Masad expects automation to shift developers’ attention from environment setup and library plumbing toward customer needs and business logic. He speculates that a strong engineer could be as productive in ten years as 100 engineers are today. This is an ambitious prediction, not a demonstrated result.

He says fully automating creative business judgment would require capabilities he associates with general intelligence. In 2023 he regards AGI within ten years as a low-probability outcome, while acknowledging that others disagree and that the consequences would be difficult to predict.

Stebbings asks whether more coding supply will reduce developers’ wages. Masad predicts some highly paid, low-effort roles could pay less, while unusually creative engineers may gain leverage and earn more with AI assistance. He illustrates the idea with John Carmack and Elon Musk; his wage and productivity claims are forecasts, not measured labor outcomes.

Masad worries that small teams could produce very valuable companies and that wealth inequality and resentment could grow. Stebbings raises the possibility of social unrest; Masad agrees that division and even violence are risks he hopes do not materialize. Their discussion is speculative and Masad’s interpretation of inequality is contested.

On knowledge inequality, Masad offers a different prediction: AI tutors could broaden access to education and shrink knowledge gaps, even if economic outcomes remain unequal. His comment that ChatGPT might add ‘IQ points’ is a metaphorical personal claim, not an established measurement of intelligence.

Asked what he would do at Meta, Masad supports shifting some resources toward AI, which he sees as already gaining traction. He does not reject the long-term possibility of immersive technology, but thinks the timing and available capital for the metaverse may have been overestimated. This is his 2023 view of another company’s strategy.

Stebbings raises concerns about Chinese-owned consumer apps and access to US user data. Masad says he values democracy and open markets and is concerned about the Chinese government’s values, but argues that a ban should depend on evidence of harm. Stebbings replies that the possibility of government access itself warrants action.

The two disagree over how much evidence is needed and how a law would define the countries or systems covered. Masad points to the difficulty of applying a rule consistently, including to other countries with restrictive governance. Their claims and policy positions are presented as a debate, not an independent finding about any app’s data practices.

Masad predicts pressure for TikTok to increase transparency, US operations or separation of data, and speculates that those steps might avert a ban. He thinks an outright ban less likely without sufficient political will. This was a 2023 forecast, not an account of subsequent legislation or events.

In the quick-fire round, Masad says he is optimistic about broader access to the software economy. Having children has made him more concerned about the future. He values the entrepreneurial community in San Francisco while acknowledging that similar communities can exist elsewhere.

He says he would like to lead a very different business for a day to understand an unfamiliar industry. One belief he expects many people to dispute is that Bitcoin could eventually become a global reserve currency. He offers a long-run argument about neutrality and portability, without specifying that it will happen in his lifetime.

Masad would tell his younger self not to sacrifice as much health and friendship for work. He says he has been rebuilding habits and relationships, and wishes he had spent more time with his mother before her death. He reflects that pursuing success can incur a personal debt whose costs are difficult to reverse.

For 2028, Masad hopes a person in rural Jordan or elsewhere can learn to code, earn money and start a business using Replit. He wants the product to support the journey from a first line of code to a first company and revenue. This is the founder’s ambition for his business, not an established outcome.