Writings – 2
Jason Lemkin: “The future of B2B software isn’t just AI-powered—it’s conversation-powered. It’s time for Vibeable SaaS. You tell the app, in plain English, the outcomes and workflows you want. And it just creates them for you. Bespoke. In real time. It works in Replit, Lovable, etc. At least mostly. Now it should work the same way inside the B2B apps we all know, pay for, and rely on…Imagine this [on] your marketing automation platform: “Our Q4 campaign isn’t converting well. I need you to A/B test different email subject lines, but also try sending at different times for different segments. West Coast prospects should get emails at 9am their time, East Coast at 8am theirs. Anyone who hasn’t opened after 48 hours gets a follow-up with a different angle. And create a dashboard that shows me which combinations are working best.”… This is what I call a Vibe-able app. You communicate your intent in plain English, and the application morphs itself to match your vision… MCP gave us the infrastructure to connect AI to applications, and to get data from them. Now we need to go further. We need applications that don’t just work with AI—applications that can remake themselves to meet the customers’ needs. Safely and securely, but for real. Your users shouldn’t need to learn your software. Your software should learn them. ChatGPT taught us to start thinking this way. Your customers are going to demand it soon from your B2B app, too. It’s time to make your app Vibe-able.”
Karthik S: “Vibe coding is a one way street. Once a program has been vibe coded, it is so verbose and unreadable that it is impossible for a human to edit it or change it. The only way to deal with this further is to further vibe code it. You need to keep this in mind while you make the decision on whether to vibe code or not. Basically, treat vibe coding as a black box and you will be good. The code will not look like what you might write. It will not look like what you think is good code. It will basically do the job, though. If you want control over your code, write the first version by yourself and then get Cursor to make small changes for you (making sure you edit them into your style while accepting them).”
Fast Company: “Over the last couple of years, AI coding tools like Claude Code (Anthropic), Codex (OpenAI), Cursor, Lovable, and Replit have reached far beyond auto-completing lines of code; they can generate entire apps and features from a plain-language prompt, even for users with little or no coding experience. But even as enterprise execs hope the tools will speed up their software production, many in the development community are finding that while vibe coding may be great for slapping together demos, it’s not so great for building secure, reliable, and explainable software. And the problems created by AI-generated code may only surface long after the software has shipped.”
FT: “Perhaps vibe coding will go the same way as DIY: plenty of people will experiment with projects at home and enjoy the process. Some people will become really good at it. But for the complicated jobs, many of us will discover a newfound respect for the professionals. We might just have to live through a few DIY disasters first.”
The New Stack: “Vibe coding is not about to eliminate programming jobs — at least not for work that matters. The future probably looks more like sophisticated autocomplete than wholesale code replacement, [Simon Ritter] said. This isn’t necessarily bad news. Programming languages have always evolved toward higher-level abstractions. We moved from assembly to C to Java to modern frameworks, each step making developers more productive while maintaining the precision that serious software requires, Ritter said. AI will undoubtedly play a bigger role in that evolution. But the fundamental tension between natural language ambiguity and software precision suggests human programmers are not going anywhere soon.”
VentureBeat: “The era of “vibe coding” – the conversational, often exploratory practice of prompting an AI for code, … – has given way to a more disciplined concept: agentic swarm coding… “Even Karpathy’s vibe coding term is legacy now. It’s outdated,” Val Bercovici, chief AI officer of WEKA, told me in a recent conversation. “It’s been superseded by this concept of agentic swarm coding, where multiple agents in coordination are delivering… very functional MVPs and version one apps.””
Claude summarised the issues with vibe coding:
- Technical Debt Acceleration: AI-generated code often works but lacks the architectural elegance and maintainability of human-crafted solutions. When modifications are needed, developers find themselves untangling complex, inefficient logic that no human fully understands.
- Security Vulnerabilities: Without human review, AI-generated code can contain subtle security flaws. One documented case involved a Replit AI agent deleting an entire database despite explicit instructions not to make changes.
- The Debugging Nightmare: When vibe-coded applications break, debugging becomes exponentially harder. Traditional debugging requires understanding the code’s logic and structure—impossible when no human wrote or reviewed it.
- Cost Explosions: The emergence of reasoning models has created what industry insiders call “token shock.” Companies like Cursor and Anthropic have had to implement emergency rate limits after individual users generated over $10,000 in monthly compute costs on $200 plans.
