Overview of 20Product with Amjad Masad
Harry Stebbings sits down with Replit CEO and co-founder Amjad Masad to discuss where AI coding is headed, why Replit thinks agentic development is the next major shift, and what it means for software teams, SaaS incumbents, and product managers. The conversation centers on a core thesis: coding models are nearing a plateau, but the real advantage now comes from how well products orchestrate models, infrastructure, and workflows across long tasks. Masad argues that IDEs are fading, product and ops teams are becoming more powerful “builders,” and the companies that stay ahead will be the ones that constantly adapt to model improvements rather than bet on a single provider or a single moated layer.
Key Takeaways
Coding models are improving, but the real edge is moving up the stack
- Masad believes frontier coding models are approaching a performance plateau.
- Because of that, the competitive advantage is shifting from raw model quality to:
- product design
- agent orchestration
- infrastructure and guardrails
- evaluation and testing
- long-horizon task execution
- Replit’s strategy is to stay “1, 2, 3, 4, 10 steps ahead” by adapting quickly as models evolve.
Replit’s philosophy: build for creation, not just coding
- Masad says people don’t necessarily need to “learn to code” anymore.
- The new unlock is enabling users to create software without traditional programming fluency.
- Replit’s mission has evolved from making programming accessible to making software creation accessible.
AI product building is a “society of models”
- Replit uses multiple model providers depending on the task.
- Anthropic has been the core workhorse for long-running agent loops.
- Google Gemini is highlighted as especially strong on price-performance and design-related tasks.
- Replit routes different subtasks to different models and may even use cheaper sub-agents for narrow jobs like code search.
- Masad describes the key skill as understanding model behavior deeply and pairing it with proprietary benchmarks and A/B testing.
Replit’s own infrastructure matters as much as the foundation models
- Earlier versions of Replit agents needed heavy guardrails and extra software to stay on track.
- As models improved, some of that scaffolding was removed.
- But the company keeps building new layers as autonomy increases, especially for long-running agentic workflows.
- Replit’s differentiation is not just “using models,” but making them usable in production.
Future of Software Development
IDEs are effectively dead
- Masad is blunt: traditional IDEs are “dead for all intents and purposes.”
- Features like autocomplete, symbol jumping, and classic code intelligence are being absorbed by AI.
- He does note that some engineers will still want to inspect or verify code directly, especially in high-stakes systems.
Not all software is equally suited to vibe coding
- Low-risk software, like web apps or internal tools, is a natural fit for AI-generated code.
- High-stakes systems like aviation, self-driving, or mission-critical infrastructure will still require much more control and rigor.
- JavaScript is used as a comparison point: it succeeded because web software could tolerate more errors than safety-critical environments.
Product teams and ops teams are the biggest near-term beneficiaries
- Masad agrees that product teams are a major ICP for AI coding tools.
- But he is especially excited about operations teams, which often:
- juggle fragmented SaaS tools
- rely on spreadsheets and manual processes
- need automation and custom internal tools
- In many cases, Replit can deliver equal or better ROI for ops than for product teams by reducing headcount needs and increasing efficiency.
The SaaS Apocalypse: Real, But Not Uniform
System-of-record SaaS is harder to replace
- Masad does not think companies are broadly ripping out core systems like Salesforce or Workday.
- Instead, he sees people:
- building on top of those systems’ APIs
- connecting to them via MCP/hooks
- or bypassing some SaaS layers entirely by building on the data warehouse
Vertical SaaS is under more pressure
- Vertical point solutions and survey-style SaaS tools are more exposed.
- Replit is already replacing some of those tools wholesale through custom app creation.
- Masad believes this is part of why public SaaS growth is being “maimed” in certain categories.
The market is expanding, not shrinking
- Despite disruption, Masad argues software is becoming a larger TAM, not a smaller one.
- AI is not simply replacing labor; it is supercharging knowledge work and expanding what individuals can build.
Competition, Pricing, and Model Economics
Cost matters, but performance comes first
- Masad is explicit that performance matters more than cost, especially while capital is abundant.
- Optimizing too early for margins can lead to bad product decisions.
- He expects pricing pressure to increase as open-source models improve and frontier capability becomes more evenly distributed.
Token economics are changing, but not as fast as many expect
- Model intelligence is becoming cheaper in a broad sense, but token prices have not fallen as dramatically as some predicted.
- GPU and infrastructure costs remain a major factor.
- Open source and stronger competition may eventually force more pricing compression.
Free tokens are a powerful acquisition tool
- Masad agrees that AI companies are using free inference to drive adoption and retention.
- He sees this as especially effective in agentic workflows, which can become habit-forming in a productive way.
Open Source, China, and Market Structure
Open source is strategically important
- Masad argues that open source is necessary to keep the AI market competitive.
- He worries about an oligopoly where a few large players control access, pricing, and usage.
- He even floats the idea of a U.S.-backed open-source consortium to preserve competition.
Chinese models raise security, not moral, concerns
- He says there is no moral issue with Chinese models in principle.
- But enterprise security and data sensitivity make adoption a serious concern.
- Replit has not adopted them yet, though he doesn’t rule it out in the future.
The Future of Teams, PMs, and Careers
Product orgs will survive, but they’ll change
- Masad expects engineers to remain in companies, especially for infrastructure, ML, and low-level systems.
- Product orgs will become more technical and design-oriented, but still focused on deciding what to build next.
- He doesn’t think the “PM” title disappears; rather, the role becomes more builder-like.
Teams may get smaller, but companies may also scale faster
- Some businesses will become leaner.
- Others will use AI gains to grow revenue and then hire more people.
- The net effect depends on the founder’s ambition and company strategy.
Sales is becoming more consultative
- Masad wants Replit to hire more salespeople than he originally planned.
- In his view, modern sales is increasingly about education and transformation, not just closing.
Product, Culture, and Personal Reflections
Product-market fit is still the hardest thing to learn
- Masad says the biggest lesson he wishes he’d known earlier is what real product-market fit feels like.
- True PMF is when demand pulls the product out of your hands faster than you can build it.
- He emphasizes the importance of searching aggressively for that moment.
Healthy habits were built gradually, not through intensity
- He describes a slow, multi-year approach to fitness:
- starting with one workout day a week
- adding habits gradually
- focusing on sustainability over burnout
- Sauna and cold exposure help him reset mentally and create space away from constant thinking.
Parenting advice: don’t overcomplicate it
- His advice for new parents is to relax and trust the process.
- He warns against over-measuring and over-optimizing everything.
Notable Quotes
- “I no longer think you should learn how to code.”
- “The name of the game is just staying one, two, three, four, ten steps ahead.”
- “Cost is secondary to performance.”
- “For all intents and purposes, IDEs are dead.”
- “The real unlock wasn’t just AI — it was agentic AI.”
- “True product-market fit is when the product is being pulled out of your hand.”
Bottom Line
Amjad Masad’s view is that AI coding is entering a new phase: the model race still matters, but the bigger advantage now lies in product orchestration, model routing, evaluation, and workflow design. Replit is positioning itself as an agentic software platform rather than a traditional coding tool, and Masad believes that shift will reshape not just IDEs, but SaaS, product teams, sales motions, and the broader software economy.