That was quick! Israeli startup Base44, started by 31 year old solopreneur Maor Shlomo sold to Wix for $80M just six months from launch. This is Maor’s second startup, having previously founded venture-backed Explorium, At the time of the acquisition, Base44 had over 250,000 users and was profitable ($189K profit in May alone, according to Shlomo). Base44 is part of a rising wave of ‘vibe coding’ tools, which enable users to automate the process of software development to simply writing a prompt.
What Is Vibe Coding?
Vibe coding is revolutionizing software development by letting developers describe what they want in plain English, while AI writes the actual code. Popularized by Andrej Karpathy’s viral posts (he was previously the head of AI at Tesla), it shifts software creation from “how to do it” to “what to do” — enabling faster and more accessible app building.
At its core, vibe coding is a revolutionary software development methodology where developers articulate their desired outcome using natural language prompts, and AI tools then generate the actual code and implementation. It’s great for rapid prototyping especially of simple products/features, but it doesn’t replace entire software teams (yet). I highly recommend watching Andrej’s “Software in the era of AI” given at Ycombinator’s AI school:
Opportunities for Startups: Do more with less
Vibe coding tools empower startups to innovate faster, reduce development costs, make software creation more accessible across the team, and maintain flexibility in their technological choices.
Startups, especially ‘Day 0’ products looking to build a quick MVP from scratch should absolutely take advantage of these tools for rapid prototyping.
- Democratising Development Tools: Startups can create tools that make software creation accessible to a much broader audience, including non-coders, low-coders, technical product managers, and analysts. Examples include Lovable.dev, Bolt and Base44, which focus on user-friendly interfaces for app building via prompts.
- Providing Custom Project Generation (Generative Scaffolding): Instead of rigid templates, startups can offer platforms that generate personalized project scaffolds based on natural language descriptions (e.g., “a TypeScript API server with Supabase, Clerk and Stripe”). This approach can lead to a wider distribution of composable, stack-specific generations, reducing framework lock-in and encouraging experimentation. Tools like Tempo Labs, Bolt.new, and Replit are already active in this space.
- Building Full-Stack Application Generators: Startups can focus on tools that enable individuals to build entire full-stack applications quickly, sometimes within a day, using AI prompts. These tools can generate and visually present apps, allowing users to make targeted changes through prompts. Examples include Tempo Labs, Bolt.new / Bolt.diy, Lovable.dev, Base44, Gemini Canvas, and Rork.
- Developing AI-Native IDEs and Extensions: The shift impacts core developer tools. Startups are creating VS Code forks like Cursor and Windsurf that integrate AI deeply. Others are building VS Code extensions like Amp, Augment, Continue, Cline, and Cody (from Sourcegraph), which offer features such as autonomous coding agents, codebase analysis, chat modes, and task automation.
- Innovating in Version Control: With AI agents writing significant code, there’s an opportunity for AI-native Git systems that track “prompt+test bundles” as versionable units, along with rich metadata about which agent made changes or where human oversight is needed. This shifts the focus from line-by-line changes to expected behavior and outcomes.
- Creating Dynamic, AI-Driven Interfaces: Startups can build solutions where dashboards evolve from static displays to conversational, adaptive interfaces. LLMs can help users find controls, synthesize data, and even suggest relevant metrics, potentially leading to “dual-mode interfaces” optimized for both human and agent experience.
- Transforming Documentation: Documentation is becoming an interactive knowledge system. Startups can provide solutions that structure documentation as searchable databases and as grounding context for coding agents, effectively making documentation “for agent consumers” as much as for humans.
- Developing New Secret Management Solutions: The traditional
.envfile approach is insufficient in an agent-driven world. Startups can innovate with local secret brokers that provide scoped, revocable tokens or capability tokens to AI agents, decoupling secret access from the static filesystem and offering auditability. - Enhancing Accessibility as a Universal Agent Interface: Startups can leverage accessibility APIs to enable AI agents to observe and interact with applications semantically, moving beyond pixel positions or DOM scraping. This suggests a new “render surface” where apps define agent-accessible context through structured annotations.
- Building Asynchronous Agent Workflow Orchestrators: Developers are increasingly delegating tasks to agents that operate in the background. Startups can create orchestration layers for these asynchronous agent workflows, allowing agents to interpret designs, respond to feedback, and triage bugs across various platforms.
- Driving MCP (Model Context Protocol) Adoption: MCP is emerging as a universal standard for agents to interact with the real world, providing context and enabling modular integrations. Startups can build MCP clients and servers and optimize existing services for agent consumption by shipping with MCP surfaces by default.
- Providing Abstracted Primitives for Agents: Just as human developers rely on services for payments or authentication, agents need clean, composable service primitives. Startups can provide these services, optimising them for agent consumption by exposing APIs, schemas, and capability metadata, potentially even shipping with MCP servers by default.
Vibe coding is indeed more than simple low-level automation of software.
The Promise and the Pitfalls of vibe coding
Despite these significant innovations, it’s important to remember that vibe coding is still evolving. Key challenges include:
- Prompt Engineering is Key: The quality of the AI-generated code heavily depends on the clarity and specificity of the natural language prompts. “Quality in = quality out”. Ambiguous prompts can lead to poor code.
- Code Validation and Trust: Even with advanced models, human oversight and review remain crucial. AI-generated code needs careful inspection for security, performance, and correctness. Experts like Martin Casado note that while AI is “great at doing dazzling things,” it may not be “good at doing specific things”. Relying too much on AI could lead to “buggy and hackable code” or “masses of broken code, full of security vulnerabilities”.
- Context Management Challenges: As codebases grow larger and more complex, AI tools like Cursor may struggle with maintaining constant context, requiring developers to maintain rules and context files to guide them effectively. Large language models (LLMs) currently “can’t reason their way around” complex dependencies in software projects.
- Non-Deterministic Output: The output of AI tools can be unpredictable and may vary even with the same prompt, leading to distrust among some developers. As computer scientist Daniel Jackson points out, “almost no applications in which ‘mostly works’ is good enough”.
- Scaling Complexity: Handling large projects with many dependencies remains difficult for current AI models.
- Cost and Practicality: Many AI coding tools are expensive and best suited for starting new projects (“Day 0”) rather than maintaining or improving existing ones. They are less effective for “Day 1+” tasks, which involve improving, fixing bugs, adding features, or working on existing, large codebases with multiple team members.
- Versioning and Documentation: AI-generated code still needs to adhere to team practices for versioning and documentation.
- Compliance: In regulated industries like finance, human oversight is essential for compliance, even when AI scaffolds the code.
Experts like Martin Casado caution that while AI is impressive, it can generate “masses of broken code” without proper supervision.
Despite these challenges, many experts believe AI will fundamentally change how software is built, primarily by abstracting lower-level tasks and allowing developers to focus on higher-level design and architecture.
Below is a non-exhaustive list of vibe coding tools and IDEs/code editors:
And also this vibe coding landscape by Nufar Gaspar:
Overall, vibe coding is just tip of the iceberg when it comes to the automation of software development. Think of it as the “prompt engineering for coding”. In reality, the agentic AI workflows making their way into software engineering teams are far broader across development tools, documentation, templating, etc. A16Z recently published a good overview of developer emerging patterns.
I’m not sure we expected the job loss coming from AI automation to start with software developers, an area that for long has been perceived as ‘safe’ for jobs, but the pace of innovation in the coding automation space is fast and very real.
I'm a former general partner at google ventures, head of Google for Entrepreneurs in Europe and founding head of Campus London, Google's first physical hub for startups.
I'm also the founder of Techbikers, a non-profit bringing together the startup ecosystem on cycling challenges in support of Room to Read. Since inception in 2012 we've built 11 schools and 50 libraries in the developing world.
- Remagine Ventures Ranked Among Israel’s Most Active VCs in 2025 - March 18, 2026
- The Global Race for AI Sovereignty: Where Does Israel Stand? - March 16, 2026
- The Web’s Next Customer Isn’t Human - March 13, 2026
