Optimism vs. The Market

The engineer in me is amazed by AI. The entrepreneur in me is suspicious.

One sees technology that can digest PDFs, emails, and contracts and spit out structure in seconds. The other sees a market moving like a bubble: noisy, crowded, and over-funded.

Bubble Dynamics

We're at the late stage of hype. The clouds are settling, the noise is cooling, but the rat race is at full sprint. Everyone wants to be the next Notion, Slack, or Stripe — but “with AI.”

Most will fail. The lucky ones will get absorbed.

Big Tech Can Kill Directions Overnight

Need proof?

• Google hyped Stadia, Inbox, Reader… then killed them. Gemini could meet the same fate if revenue stalls.
• Microsoft & Meta sold us the Metaverse dream, then buried it in silence.

Lesson: when big tech pivots, entire startup categories vanish overnight.

What Has Stuck vs. What Hasn't

Stuck:

• General chatbots (ChatGPT, Claude)
• Coding copilots and vibe-coding IDEs (Cursor, Windsurf)
• Vertical AI (healthcare billing, legal reviews, manufacturing line monitoring)

Hasn't stuck:

• AI note apps
• AI browsers
• AI productivity wrappers
• AI-powered search clones

Some products solve pain. Others just… look shiny.

The Economics Problem

Past revolutions drove costs down. PCs, mobiles, GPS, Linux — they scaled because they got cheaper.

AI? It's driving costs up.

• GPUs are the new oil.
• APIs come with surprise bills.
• Context windows are a silent tax.
• Agents burn money with every query.

Even OpenAI and Anthropic, with billions in revenue, are bleeding billions in losses. The only guaranteed winner? NVIDIA, selling the shovels in this gold rush.

The Platform Play

Here's the truth: chatbots themselves don't monetize. 95% of users are free-tier.

That's why the labs aren't really building “products.” They're building platforms. OpenAI, Anthropic, Google — their strategy is clear: build the model, let startups pile on top, and hope some survive long enough to create real value.

Startups are the guinea pigs. Infra is the cash cow.

Why Mostly B2B

I pulled data on the last 500 YC startups.

• 2023 → 44.5% AI.
• 2024 → 72.5% AI.
• 2025 → ~80% AI.
• Of those, ~85% are B2B.

Because B2C is brutal. Consumers won't pay $200/month for “AI email helper.” Enterprises will — if you cut costs in law, finance, healthcare, or dev workflows.

That's why YC bets on vertical AI: Amber reducing claim denials by 55%, Linewise preventing million-dollar manufacturing downtime, Lora building AI workspaces for lawyers.

Vertical, niche, high-value.

The Trust & Adoption Gap

Here's the reality: most AI features impress in demo day but fizzle in real life.

• How many meeting summaries do you actually read?
• How often do you trust an AI email draft without rewriting it?
• How often do agents deliver ROI beyond the wow-moment?

Demos create excitement. Real adoption demands trust. And AI hasn't earned that yet.

The Data Fear

And then there's the elephant in the server room: data.

Companies hesitate to pipe sensitive data into a model hosted by OpenAI or Anthropic.

• Will it be used to retrain the model?
• Could it leak into someone else's output?
• Who owns the derivative knowledge?

Self-hosting is expensive. Trusting providers is risky. For many enterprises, the fear of data leakage outweighs the benefit of automation.

The Human Cost

It's not just money. It's people.

Developers are asked to pivot, learn, pivot again — building “AI-for-X” products that rarely survive. Employees babysit bots that create more work than they save. Careers get tied to hype, then cut when the bubble bursts.

The collateral damage of hype isn't just failed startups. It's burnt-out builders.

The Real Value of AI

Here's what I do believe: the true breakthrough isn't text generation. It's consuming arbitrary, messy information and turning it into structured data.

That's why vertical AI makes sense.

• Reading contracts.
• Parsing medical records.
• Digesting resumes.

Clear pain points. Clear ROI.

Not another AI note app. Not another AI browser.

ROI Problems

The core problem: with AI, the cost curve scales with usage.

Traditional SaaS: build once, scale cheaply.
AI SaaS: the more your customers use it, the higher your bills.

And ROI takes months — sometimes years — to show. Clients get impatient. Startups die in the gap.

The Silent Tax

Every prompt has a cost. Every token adds up. Every context window is a toll booth.

That's why AI agents burn money at scale. A single fat query can cost $5. Imagine scaling that across thousands of employees. No CFO is signing that invoice.

Coding Bots: Reality Check

On paper: one bot replaces two developers.
In reality: one dev babysits two bots, doing twice the work.

More lines of code get written. But business value? Questionable.

Should You Add AI to Your Product?

Clients often ask me:

• “Can you build us an AI chatbot?”
• “Can you add AI to our workflows?”

Yes, we can. But should we? That's the harder question.

The Illusion of “Just Add AI”

Adding AI is not a growth hack.

• An AI menu won't make diners order more food.
• An AI marketing site won't boost conversions without a working CTA.
• A chatbot won't double your revenue unless it actually reduces support costs.

AI is a tool. Not a business model.

When AI Does Make Sense

That said, it's not all bad. AI can shine when ROI is measurable:

• Customer support bots that cut tickets by 40%.
• Workflow automation that reduces hours to minutes.
• Vertical AI in healthcare or finance where efficiency gains save millions.

These use cases stick because they save money, not because they look shiny.

My Promise as a Builder

I won't say no to adding AI. But I also won't bolt it on for hype.

My filter is simple:

• Will it cut costs?
• Will it increase revenue?
• Will it improve customer experience in a way worth paying for?

If yes, we'll build it. If not, let's save the money.

AI as Bottled Water

Here's my analogy: AI is water. Essential. Revolutionary.

But the current market is the bottled water economy — endless clones, endless wrappers, endless branding. We already have ChatGPT and Claude. Do we really need 500 clones?

So Why Not Bet?

Because AI today is powerful, but not yet sustainable. Costs are high, trust is low, ROI is shaky, and hype is everywhere.

I believe AI will win eventually. It'll become as cheap and invisible as Wi-Fi — something we don't think about, but can't live without.

But until that day, I'd rather build boring software that customers pay for than gamble on bottled hype.

Because hype doesn't pay bills. Customers do.