I run a software development company. For years, clients came to us with requests like, “Can you build my MVP?” They’d have an idea, need validation, and we’d charge $20-50k to build a prototype. Most failed. That was just the reality of startups.

By mid-2025, those requests stopped. Completely. And I was freaking out.

I think the requests stopped, not because people stopped having ideas, but because tools like Lovable, Bolt, and Replit can now build that same MVP in a weekend for free. Anyone can vibe-code a basic app without writing a single line of code.

I lied to myself, saying this was fine. They’d validate with AI tools, hit limitations, then come to us for “the real thing.” Some did. Our remaining clients were better educated, post-validation, ready to invest properly.

But then I started seeing the numbers shift, and that shit was scary.

A flight simulator built in 3 hours making $87K per month. Solo founders hitting $100K MRR with vibe-coded products. A non-coder building a six-figure business on Replit in days. Apparently 25% of Y Combinator’s Winter 2025 batch is running on 95% AI-generated code.

These are no longer validation projects. They’re businesses. Stay calm Gavin… breath.

So I started asking a harder question, and probably the one many in my industry are asking too:

Where does my agency fit in this world? Or does it?

The validation revolution is genuinely empowering

From my perspective, here’s what happened. The barrier to testing a product idea dropped from $50,000 to $0. That’s true, because I’ve seen it first hand in my business with real clients.

You see examples of this all over now. Someone built a “Has the dog been fed?” tracker using Replit. It’s perfect for them. Market of one. When the build cost approaches zero, you can build for an audience of yourself. This is honestly kinda cool.

I’ve done this too. I built a spelling practice app for my kids in a couple hours. It scans their spelling sheet, reads words aloud, and lets them drill until they’ve got it. No searching for apps that almost work, no subscription fees, just exactly what we needed.

A friend built a hockey announcement app for his kids’ team using Base44. It generates audio for player names, handles goal announcements, team lineups, the whole production. A fun, specific use case for his family that no commercial product would ever serve.

When the build cost approaches zero, you can build for an audience of one.

Tools like Lovable and Bolt build about 70-80% of any basic app in hours. For validation, that’s perfect. Get user feedback quickly, prove the concept and learn what people actually want.

Most of these ideas will fail, and that’s fine. Bad ideas should fail fast. Better to learn in a weekend than after months of development and tens of thousands of dollars.

The uncomfortable evidence

Pieter Levels built Fly.pieter.com, a browser-based multiplayer flight simulator in 3 hours using Cursor. Within 17 days it hit $87K MRR. Now it has 320,000 users. He hasn’t rebuilt it. It’s still running on that original vanilla HTML/JavaScript. To be fair, Pieter has a huge network to sell too. But lets focus on the technical side here.

TrendFeed was built in 4 days with AI. Hit $12K revenue in 4 weeks. Now at $10K MRR. Sheesh.

Chatbase was built by a college student in 2 weeks using Supabase. He reached $8M ARR completely bootstrapped. Now its a polished, full-blown product with great branding.

A Brazilian edtech used Lovable to build a premium platform in 2 weeks. Generated $3M in revenue in 48 hours. $3M!? 48 hours. Thats unfair.

It’s painful to admit, but these aren’t outliers anymore. They’re a pattern, and the pattern contradicts the story I was telling myself.

The story I was telling myself

The story in my head went a little like this: vibe-coding for validation, professionals for “real” businesses. It felt true because it protected my business model. Go and have fun with your app idea. Call the big guns when you want something legit. Right? Riiiight!?

But look at the data:

• 83% of small businesses earn less than $100K annually
• Only 0.4% of SaaS startups reach $10M revenue
• 91% never hit $1M

Most businesses don’t need enterprise-grade infrastructure. They don’t need to scale to millions of users. They don’t need the last 20% that professionals provide.

I would often say “plan to rebuild,” but Steve Blank would call that “startup suicide.” His argument is that while engineers rewrite, competitors add features and gain market share. You don’t get to declare a timeout because your code is ugly. But being a software engineer who loves to code makes this painful. It’s part of the craft.

Airbnb started with founders manually taking photos of apartments. None of it was scaled. They iterated from there, they didn’t wait for professionals to build it “properly.”

Where my framework still holds

I don’t think everything I believed was wrong. Some of it holds up.

The Airbnb example still matters, not for early-stage startups, but for what a business becomes. Yes, you could build something that looks like Airbnb’s app using Bolt. But that app would have zero value. Because the app is maybe 10% of what makes Airbnb valuable.

The other 90% of Airbnb is trust systems, payment infrastructure, matching algorithms, fraud detection, multi-currency compliance, insurance frameworks, customer support at scale, network effects, brand reputation.

Same with Starbucks. Their app is the front-end to a closed-loop payment system. 17 million users locked into their ecosystem. Proprietary AI yielding 30% ROI boosts.

At this scale, the moat isn’t the app. It’s the ecosystem.

But lets look back at those earlier stats a second before we start lying to ourselves again. Most businesses aren’t trying to become Airbnb or Starbucks. And for them, the vibe-coded version might be enough. $8M ARR as a solo founder? Yea, I’ll take that.

What’s actually commoditized

Let me be honest about what died from our business.

Basic CRUD apps. Shit like ToDo lists, Landing pages, Marketing sites. Standard authentication systems. Generic productivity tools. The $20-50k MVP.

AI wrappers are another. I said they were dying. But Cursor hit $500M ARR with zero marketing spend. Harvey AI reached $100M ARR serving law firms. PDF.ai makes $6M ARR as a solo founder project. These are “thin wrappers” that won through distribution, not technical moats.

The moat was never the code. It was distribution, network effects, timing, and brand. My inner engineer was dying inside.

The gap keeps shrinking

The 80/20 gap isn’t stable anymore.

In August 2024, the best AI solved 33% of SWE-bench coding challenges. By late 2025, Claude Opus 4.5 solves 81%. Context windows now ingest entire codebases. Autonomous agents take high-level plans and produce complete programs while you sleep.

What required professionals last year doesn’t this year. That trend continues at an alarming rate.

When I say “the last 20% requires professionals,” I might be describing a temporary limitation, not a permanent truth. The AI army marches ever forward.

Where does my agency fit now?

This is where I stop having confident answers and start having questions.

Here are my thoughts around that, in no particular order:

Scale-up specialists. Only take clients who’ve hit real walls that iteration can’t solve. Post-PMF, proven revenue, actual infrastructure problems. We’re not building MVPs anymore, we’re solving problems that emerge at scale. Higher bar, fewer clients, bigger budgets. This is closer to “devops engineer”. This market is smaller. And AI might solve scale problems too.

Distribution partners. When technical barriers are low, distribution beats product. What if we pivoted from “we build your product” to “we get your product in front of users”? Growth engineering, not software engineering. We already do this, but maybe we put our foot on the gas? Do we want to be competing with dedicated marketing agencies.

Vibe-coding accelerators. Instead of building for clients, teach them to build themselves. Charge for strategy. Show them what to build, which problems matter, how to sequence features. The building becomes their job, we’re the guide.

Vertical specialists. Go deep in regulated verticals where compliance is the moat. Focus on areas we know well like healthcare, finance, legal. HIPAA, SOC2, FDA approval. These don’t get vibe-coded away easily. There’s hoops to jump through and getting it wrong is catastrophic.

AI-augmented boutique. Embrace the tools fully. Smaller team, faster delivery, lower prices, higher volume. If vibe-coding does 80%, our value is the 20% plus quality control plus accountability.

I don’t know which of these is right. I’m probably going to try several.

What I’m watching

The 80/20 line. Every quarter the tools get better. If the gap closes faster than I expect, options 1 and 5 become less viable. If it stabilizes, they might be the play.

Distribution as moat. Calendly is a $3B company solving a trivially simple problem. Every Calendly link is an unpaid ad. Pieter Levels’ $138K/month comes partly from 600K Twitter followers. If distribution is the real moat, maybe we should be building audiences, not apps.

The rebuild rate. How many vibe-coded businesses actually hit walls that force rebuilds? If most iterate successfully on their original base, the “plan to rebuild” advice is wrong. If most hit breaking points, we’re the ones they call.

Solo founder success stories. Realistic solo founder timelines seem to cluster around $500-$2K month one, $5K-$25K by month six, $10K-$50K by month twelve. If those numbers hold, the “market of one” expands into “business of one.” That’s a different competitive landscape.

What remains defensible

There’s a few areas that still feel durable and are worth noting.

Regulated moats. Healthcare ($187B by 2030), banking licenses, FDA approval. Compliance complexity is a feature, not a bug. This might be the last safe harbor. It’s also a pain in the ass.

Network effects. Not all networks are equal. Uber’s proved weak, Airbnb’s proved strong. But when they work, they’re still defensible.

Deep vertical integration. Embedding into workflows creates switching costs. Implementation complexity creates barriers. This is where “the other 20%” still matters.

Brand. When product is commoditized, brand becomes differentiation. Generic AI apps are forgettable. But brand isn’t something you hire an agency to build, it’s something you earn over time.

My honest conclusion

I started writing this post to explain why companies still need professional software development. I was going to tell you that vibe-coding is for validation, and real businesses hire people like us. Fat chance.

The evidence didn’t support that clean story, as much as I tried to read around the facts.

Some vibe-coded apps become real businesses. Most businesses don’t need enterprise scale. The technical gap keeps shrinking. Distribution matters more than code quality. Professional development has its own problems like cost overruns, delays, technical debt.

So as of today, here’s where I sit:

The $20-50k MVP is dead, and it’s not coming back. Good riddance, it was a bad deal for founders anyway and boring to work on.

Some percentage of businesses that would have hired us will now succeed without us. That’s fine and they should! Being a founder, I love seeing others win. Its infectious.

There’s still a market for what we do. I’m just not sure how big it is, or how long it lasts, or what shape it takes. That ball is still very much in the air.

I’m not writing this post to pitch my services. I’m writing it because I think a lot of people in my industry are telling themselves comforting stories. I was too.

The barrier to entry dropped to zero.

The barrier to building something that matters? I used to think that stayed exactly where it was.

Now I’m not so sure. But I’m also too fucking stubborn to throw my hands in the air. So lets see how deep this rabbit hole goes. ✌🏻