Replit's CEO On The Only Two Jobs Left In The Company Of The Future

Predictions and Expectations:

• Amjad Massad expects that interaction with agents will eventually become multimodal, potentially incorporating video and audio, rather than remaining text-only.
• He predicts that the ability to build sophisticated software without technical knowledge will lead to significant wealth creation and productivity gains across many sectors of the economy.
• He expects that "computer use models" (agents that interact with GUIs) will eventually become better, though he currently views this progress as surprisingly slow.
• He believes that "true learning on the job" (agents that continuously improve based on specific organizational data) remains far away and has not yet been unlocked.
• He expects that the future of companies will consist largely of "generalist entrepreneurs" who identify problems and deputize agents to solve them.
• He predicts that agents will eventually be capable of being given high-level commands such as "build me a SaaS company and see what works," moving toward a post-prompting world.
• He expects that agents will eventually be able to generate ideas with high proficiency, though he believes human creativity for idea generation will remain important for a long time.
• He predicts that coding will continue to serve as a "hack or workaround" for computer use agents because it allows interaction with legacy software and APIs.
• He believes that while many tasks will be automated, the sales role will remain defensible, evolving into more educational and evangelistic functions.
• He expects that "vibe coding" and AI-native development will create a "neurological shift" in the population, changing how people perceive their ability to solve problems with code.

Timelines and Milestones:

• Replit has been on its mission for "10 years" as of the time of the interview.
• Amjad Massad expects AI capabilities to see massive step changes roughly "twice a year," which influences their release cadence.
• The company plans to release a new agent version "every six months" to align with predicted AI capability improvements.
• He mentions a "late 2025" timeframe for the "autonomous revolution" involving tools like Opus 4.6 and OpenClaw.
• He references "November and December" of the past year (implied late 2024 or 2025) as the period when "true autonomy" arrived for the market.
• The company is currently operating with a cycle of roughly "four weeks" between major agent releases, contrasting with the "three months" of their YC accelerator days.
• He expects that if a user cannot build something today, the platform might be able to build it in "a couple of weeks."
• He notes that the current period is a time where people are becoming "60 to 70% cheaper and more effective than traditional agencies" through vibe coding.

Technology and Product Direction:

• Replit plans to create "parallel agent" architectures to allow asynchronous work and design while the agent builds.
• The company is developing a "canvas" interface to allow users to design and explore features while agents work in the background.
• They are moving toward a "multi-agent architecture" that supports "teamwork" by allowing users to fork new VMs and work in parallel.
• The roadmap includes enabling the creation of mobile apps, websites, decks, and videos all within the context of a single project.
• They plan to continue vetting and integrating "MCPs" (Model Context Protocols) and skills to allow agents to expand their capabilities, similar to downloading skills in the Matrix.
• They are focusing on improving "computer use models" to enhance the agent's ability to test applications for both functionality and UX.
• They intend to move beyond writing skills to files toward enabling "continual learning" for agents within specific organizations.
• They plan to expand the role of "vibe coding" teams to allow internal staff to act as generalist entrepreneurs solving internal problems like support prioritization and HR onboarding.

Market and Industry Outlook:

• Amjad Massad expects a "billion new developers" to emerge, specifically "AI native developers" who do not worry about system components.
• He predicts that non-traditional roles (product managers, designers, entrepreneurs) will increasingly adopt software creation, shifting the market outside of traditional engineering.
• He believes the "developer tools" market is improving as tools become more joyful and focused on creation rather than accidental complexity.
• He expects that "vibe coding" agencies will disrupt traditional software agencies by being significantly cheaper and more effective.
• He predicts that the "PLG" (Product-Led Growth) motion remains the gold standard, with a shift where non-engineers are empowered to bring software into organizations just as developers did in the last few decades.
• He expects that "continual learning" capabilities will eventually allow agents to become more powerful within specific organizational contexts.
• He predicts that the "sales" function will transform from transactional to educational and evangelistic as AI handles more technical aspects.
• He believes that the "blind spots" in the economy where legacy software exists will be major areas for improvement by non-technical builders.

Company Plans:

• Replit plans to continue focusing on "creators" and "AI native developers" rather than traditional developers, though developers remain a part of the user base.
• They plan to expand their "DevRel" team to act more as "educators" to show non-developers what is possible, moving beyond traditional developer relations.
• They plan to maintain "enterprise sales" motions that include "hackathons" and "evangelism" to onboard champions within large organizations.
• The company intends to continue rewriting backends to support "long-running containers" for autonomous agents.
• They plan to implement "parallel agents" and "merge conflict" resolution to support multi-threaded, collaborative AI work.
• Replit plans to support the creation of "repl-native agencies" that leverage the platform to deliver software services at a lower cost.
• They plan to deploy "agents inside organizations" that can improve over time, though they note this capability is not yet fully realized.
• They plan to continue iterating on "computer use models" to improve testing capabilities and UX feedback.

Financial Guidance:

• No specific forward-looking financial metrics (revenue targets, profit margins, growth percentages) were provided by the speaker.

Risks and Caveats:

• Amjad Massad notes that "computer use models" are currently disappointing because they are "hard to build" despite seeming easy to train.
• He admits that the "current way we are hacking around continual learning" (writing to files) is a temporary solution until true on-the-job learning is unlocked.
• He warns against deluding oneself about "product market fit," noting that it can be an "explosive" thing that is easy to misjudge.
• He suggests that relying solely on "prompting" may become a flaw as the industry moves toward high-level goal setting.
• He cautions that "traditional agencies" will face significant disruption from cheaper, more effective "vibe coding" alternatives.
• He implies a risk that "legacy software" might be difficult for computer use agents to handle without the workaround of coding.

Confidence and Disagreement:

• Amjad Massad states he can "confidently say" to any entrepreneur that they can build a SaaS product, consumer product, or automation product on Replit.
• He is "not sure" if or when the world will achieve a "fully autonomous company," but believes some styles of companies could reach that point.
• He expresses surprise and slight disappointment regarding the slow progress of "computer use models," calling it a "little bit of a mystery" why it is so hard to build them.
• He believes with certainty that the industry is moving toward a "post-prompting world" where high-level goals are the primary interface.
• He is confident that "coding" will remain a necessary workaround for "computer use agents" for the foreseeable future.
• He expresses a strong belief that the "entrepreneurial mindset" is the key trait for champions who will drive adoption within organizations.