Stop building this kind of AI startup

In partnership with

Hey there

On a scale of 0-10, how well are you using AI in your company? (I’m not talking about those ‘clever’ ChatGPT hacks your friend told you about).

While you were busy debating AI's role, your competitors weren't just using it; they were living it.

Sarah Wang from Andreessen Horowitz just shared what CTOs are actually reporting: AI productivity gains jumped from 10-15% last year to 30-50% this year. One team? They're running 90% AI-generated code with 10x productivity.

But here's what really caught my attention...

While we're celebrating these productivity gains, 80% of AI startups seeking funding from Silicon Valley's top VC firm are using Chinese open-source models.

Not American models. Not closed-source premium solutions.

Free. Chinese. Open-source.

What does this tell you about where the real value is heading?

It tells me we're witnessing the democratization of AI – and most people are sleeping through it.

While established companies pay premium prices for GPT-4 and Claude, scrappy startups are building billion-dollar ideas on DeepSeek and Qwen models that cost pennies to run.

Speaking of Claude – if you're using it, you need to act before September 28th. Anthropic just changed their policy to use your conversations for training by default. You have to manually opt out now.

Pretty convenient timing, isn't it?

But here's where things get really interesting...

OpenAI employees are collectively claiming they've achieved AGI and that "aliens are among us". Some synchronous, mystical tweets suggesting breakthrough territory.

Either they're having a collective breakdown, or something significant is happening behind closed doors.

Meanwhile, China isn't waiting for permission. Alibaba just announced they're developing an AI chip to replace Nvidia's presence in their market. If successful, they'll join Google as one of the few companies combining AI chip development with advanced LLMs.

Talk about controlling the entire stack.

But here's what should really worry you...

ESET just discovered PromptLock – the first AI-powered ransomware that uses the gpt-oss-20b model to generate malicious Lua scripts in real-time. It's not active yet, but it's a proof-of-concept that shows how AI can automate cyberattacks across Windows, Linux, and macOS.

So while you're building AI tools to help people, someone else is building AI tools to exploit them.

This brings me back to my original question – is your AI business already obsolete?

If you're building:

• Another ChatGPT wrapper

• A content generation tool

•  An AI assistant that anyone can replicate

• Something that requires expensive API calls when free alternatives exist

Then yes, you might be in trouble.

But if you're building:

• AI-native workflows that can't be replicated without deep domain expertise

• Solutions that combine multiple open-source models for specific use cases

• Tools that leverage local AI models for privacy and cost efficiency

• Platforms that help businesses transition from expensive APIs to open-source alternatives

Then you might be onto something.

The future isn't about building on top of OpenAI and Claude.

It's about understanding that the real power is shifting to open-source models, local computing, and specialized implementations.

While everyone's paying premium prices for API calls, the smart money is learning to run Qwen3-coder locally, fine-tune open-source models for specific tasks, and build systems that don't depend on external APIs.

The companies winning in 2025 won't be the ones using the most expensive AI.

They'll be the ones using AI most efficiently.

So let me ask you again - what are you building?

Are you following the hype, or are you anticipating where this is actually heading?

Because while everyone's debating AGI timelines, the real disruption is happening in the models you can download and run yourself.

What do you think? Are you pivoting your strategy, or doubling down on what got you here?

I'd love to hear your thoughts.

-Aashish

P.S. If you want to discuss this further with other AI builders, join our community where we share practical insights about local models, open-source alternatives, and real business applications.

Employ Anywhere (without risk or hassle)

Hiring internationally doesn’t need to be complicated—or expensive. As an Employer of Record, RemoFirst lets you legally employ talent in 185+ countries, managing all international HR and payroll aspects from one platform, starting at $199/month.

Skip the legal entity setup, surprise fees, and compliance risks. Our transparent pricing and local expertise make it simple to scale globally.

Whether you're hiring your first engineer in Brazil or your 30th contractor in India, we’ve got you covered.

Start hiring globally