Replit Triples to $9B

Six months ago, Replit was worth $3 billion.

Last week, they raised $400 million at a $9 billion valuation. Same product, mostly. Same team. The difference is that they renamed their core offering "Replit Agent" and leaned hard into a single, audacious claim: that software creation is now something anyone can do, not just engineers.

You describe what you want. The AI builds it. You tweak the output. You ship it.

They call it agentic software creation. The rest of the internet, following Andrej Karpathy's lead, calls it vibe coding.

I've been sitting with this story for a few days now, and I have to be honest: I don't know whether to be impressed, threatened, or both. My money is on both.

Connect your email and calendar, and Attio instantly builds your CRM. Every contact, every company, every conversation, all organized in one place.

Then Ask Attio anything:

No more digging and no more data entry. Just answers.

Start your free trial →

The Vibe Coding Wave Is Real, and the Numbers Show It

Here's the thing about vibe coding: the term is only a few months old, but the thesis it represents has been building for years. Karpathy used it casually to describe his own late-night side project workflow, and the idea caught fire instantly because every engineer recognized it. You stop reading the code carefully. You describe what you want, accept the output, fix the edges, and ship. The code is almost incidental.

The VC world grabbed it and turned it into a multi-billion-dollar thesis overnight.

Replit at $9 billion is one data point. Cursor raised at $9 billion in January. Anysphere crossed $100 million ARR. GitHub Copilot now has 1.8 million paying subscribers. Replit itself is targeting $1 billion in ARR by year-end. The market is sending a signal about what a platform shift looks like when it's actually happening, not just being forecasted in a conference deck.

I've invested in software tools before, and I've missed a few I should have caught. I don't think this is hype. The category is real.

What I'm not sure about is what it means for people like me.

When Anyone Can Code, the SWE Edge Gets Fuzzy

Let me say the quiet part out loud, because I've been hearing it in partner meetings and refusing to name it directly: if a non-engineer can build working software using natural language, the "I can read your code" superpower that SWE-background VCs have been selling for years starts to look more fragile than it used to.

This is the anxiety in the room. I've had it myself.

The argument goes something like this. SWEs became valuable in VC because we could cut through technical hype with hard analysis. We could tell a real moat from a wrapper startup. We could spot architectural landmines that would blow up in 18 months. That value came from being able to look at code and understand what was actually happening under the hood.

But if vibe coding democratizes code creation to the point where anyone can produce a functional application in an afternoon, does the SWE lens stop being a differentiator?

I've spent the last week arguing with myself about this. Here's where I've landed.

The argument is wrong, but it's wrong in a subtle way that's easy to miss.

The Demo-to-Durability Gap Is Getting Wider, Not Smaller

Here's what vibe coding actually does to the due diligence problem.

It makes demos significantly easier to produce and significantly harder to evaluate.

Two founders with Replit Agent and three months can now produce something that looks, feels, and behaves like a $2 million engineering project from 2022. The user flows work. The dashboards look polished. The demo lands well in a partner meeting. A non-technical investor walks out impressed.

And then you look under the hood.

The codebase is AI-generated spaghetti with no test coverage. The authentication system would fail a junior engineer's code review. The "core AI feature" is a prompt pipeline stapled to a UI wrapper. The company has two people and could not survive a real production incident, a security audit, or 10x growth without triage and a near-complete rewrite.

I've started calling this the demo-to-durability gap: the delta between how impressive something looks in a 30-minute walkthrough and how ready it actually is to handle production reality.

This gap existed before vibe coding. Vibe coding is making it wider, faster.

And here's the thing: closing that gap is exactly where deep technical judgment matters most. Not reading code for the sake of reading code, but asking the right questions. Does this team understand their own system's failure modes? Can they explain what happens when their primary AI provider has an outage? What's their incident response history? Have they stress-tested the core flow, or have they only ever lived in the happy path?

A SWE who has shipped production systems knows what to ask. A generalist doesn't. Another AI doesn't, either.

Here's What I've Actually Updated

I'll be direct about what this week changed in how I think about deals.

I'm paying less attention to the code itself and more attention to the builders' relationship with the code. Do they understand what they built? Can they walk me through their architecture without the demo running? Do they know where the bodies are buried?

A team that built something with AI assistance and deeply understands every component of it is a completely different investment from a team that built something with AI assistance and couldn't explain how their own authentication works. Both teams can produce the same demo. The diligence is in the conversation that comes after.

I'm also thinking harder about what "technical team" means going forward. The old heuristic was: do they have strong engineers? The new question is: do they have someone with the judgment to know when the AI-generated output needs to be thrown out and rebuilt by hand? That's a rarer skill than it sounds, and it's not always the person with the most impressive GitHub profile.

Does all of this mean the SWE-to-VC path still makes sense? I think it does. More than ever, actually. But the edge is evolving. It's less about code review as a mechanical act and more about the accumulated intuition that only comes from having been in production incidents, from having shipped things that broke, from knowing viscerally what a system under pressure looks and feels like.

You can't vibe-code that intuition. You earn it.

Whether I've earned enough of it is a question I'm still sitting with.

- SWEdonym

The DTC beauty category is crowded. To break through, Jennifer Anniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

Learn more