Primary data · sourced from public filings·700+ listed companies · India-first·
Open screener
ἀλήθεια · aletheiaAncient Greek for truth — literally “un-forgetting”: the act of revealing reality, not merely stating it
← All posts
Sector Thesis·4 min read·Week 26

Why AI Is Not Your Moat (But How to Use It Anyway)

Every founder in 2024 is wrapping GPT and calling it a startup. YC's recent batches show this doesn't work. The moat isn't the model—it's what you do with it that users can't replicate in five minutes.

ByAmit Tyagi·Fitoor Capital
Aletheia Insights · Weekly

Get 1 unfair insight every week from India's startup ecosystem.

Read by serious founders and investors. No fluff.

The Wrapper Trap

You've seen the formula. Take GPT-4, add a UI, charge $29/month. Ship in weeks. Get users.

It works for exactly six weeks.

Then Claude 3.5 launches and your product is a feature of a free tool. Baseline Labs analyzed YC S24: 34% of AI-first companies showed no defensible differentiation. They're still alive, but they're not building moats—they're burning runway.

The hard truth from Michael Seibel at YC: "If your product is a thin layer on an LLM, you don't have a company. You have a side project that will die the moment the model improves."

What Actually Works: Three Patterns

1. Data as the Real Moat

You need proprietary data the model can't access. Not "we'll collect it later." Now.

Example: Humind (YC S24) doesn't sell you ChatGPT with a resume parser. They've indexed millions of real job transitions. The AI is the interface. The data is the product.

For Indian founders: You have an unfair advantage. Your customers will give you data competitors won't access for three years. A fintech SaaS collecting repayment behavior? A supply chain tool tracking real-time shipments? That's your moat, wrapped in AI.

2. Workflow Lock-in

AI should disappear into your user's existing process, not replace it.

Example: Jasper (founded 2021) didn't win because they had better prompts. They won because they built integrations into Slack, Figma, and HubSpot. Users never left their tools. AI happened in the background.

Non-obvious insight: The best AI products feel boring. They don't demo well. The user doesn't say "wow, look at that AI." They say "our team shipped 3x faster this quarter." That's when you have something.

3. Domain Expertise as the Filter

You know something specific—law, medicine, construction, Indian tax codes—that the model doesn't.

Example: Spellbook (YC S21) sells to lawyers. ChatGPT can write contracts. Spellbook learned from 500,000+ real contracts. It knows what clauses matter. It knows what your jurisdiction cares about.

Why this works for India: You understand regional compliance, local payment behavior, and unmet needs. Build for that. Use AI to scale your expertise, not replace it.

The Framework: Is Your AI a Moat or a Feature?

Ask this before you build:

1. Can a user replicate this with a $20/month Claude subscription? If yes, you're a wrapper.

2. Does your value come from data, integration, or domain knowledge? If no, you're a wrapper.

3. Will your product get harder to replicate as you scale? If no, you're a wrapper.

If you answer "wrapper" to all three, pivot. Seriously.

How to Use AI Properly

Invisible integration: Embed AI where users expect automation. Not a chatbot tab. A smarter search. Faster filtering. Better recommendations.

Specificity over generality: Don't build "AI for everyone." Build "AI for accountants processing GST invoices." Solve one costly problem perfectly.

Own the feedback loop: Use your product to collect data that makes the AI better. Better AI makes the product stickier. That's the flywheel.

The India Playbook

1. Find a $500M+ problem in a regulated market. Tax filing, supply chain, lending, compliance. AI helps, but you're solving the problem.

2. Collect 10,000 examples of that problem solved. Your data.

3. Build the simplest integration into your customer's workflow. One API. One Slack command. One Excel sheet that just works.

4. Charge for speed and accuracy, not the AI. Users pay for outcomes, not technology.

5. Now scale. Your data moat is 18 months thick. Competitors are still figuring out the problem.

The Real Insight

AI isn't your product. AI is your manufacturing plant. The better your inputs (data, domain knowledge, workflow integration), the better your output.

Founders who ask "How do I use AI?" fail. Founders who ask "What costly problem can AI help me solve faster?" win.

YC's winners in 2024 aren't the ones with the fanciest prompts. They're the ones with the smartest business models.

What to Do Monday

Step 1: List three problems in your industry that cost users money and time. Not "writing is slow." Specific. Measurable.

Step 2: For each, ask: Can I collect proprietary data about this? Can I build integrations around this? Do I have unfair expertise here?

Step 3: If you answer yes to at least two, you have a company. Build that. Use AI as the tool.

If you answer no to all three, you have a feature idea. Sell it to someone else or pivot.

The startup graveyard is full of founders who thought AI was the moat. Don't be one.

Amit Tyagi

Founder, AletheiaAI & GP, Fitoor Capital

Veteran of India's startup ecosystem. Writing about fundraising, investor psychology, and what it takes to build fundable startups in India.

Run a fundability check

India's only MRE-backed platform for founders and investors. Analyse your deck, find investors, and validate your raise strategy.

#AIstrategy#productmoat#YClessons#technicaldecisions

Don’t miss the next one

One insight every week. No fluff.

Aletheia Insights · Weekly

One contrarian insight. Every week. No generic startup advice.

Join founders and investors building with better information.

Why AI Is Not Your Moat (But How to Use It Anyway) · Aletheia Insights