Claude can ship your idea next

Maybe two weeks ago I was thinking about this.

The way Claude is building integrations and solutions — it wouldn't take long for them to build anything you can come up with. Not in some sci-fi sense. Just looking at how fast they ship, and how close the product already is to the things people keep asking for in chat.

Today I saw a new video — Claude launching Dashboards and Motion.

And under the video on YouTube, people are commenting how the Claude team can ask Claude what ideas people threw into it, pick the best 5, and run with them.

That is not unrealistic. That is exactly my point.

This is basically the problem going forward. The AI companies can pick one idea and destroy a company instantly if they wanted to. Where's the regulations? Where's the public outcry when these things are getting launched?

Maybe since we are a "3rd world" country we don't have any say in this. But the impact is gonna be hard on us very soon.

All the ideas getting stolen by big tech has been seen. Now it's gonna be done in hyperspeed. They will not get Elon Musk's Hyperloop — but this, they will get to do.

I've thought about this and analyzed where the software industry is going to be in 5, 10 years a lot. With the current pace it's not gonna be anything people imagine it's gonna be. Maybe it will have some resemblance. But if the deep pockets of the AI companies keep getting refilled — there's nothing much left of employment in the tech industry other than a few in positions of power.

If tomorrow USA decides another model from Anthropic or OpenAI going to users outside USA is a "threat to its national security" — what happens to the rest of the world?

The models are trained on stolen data — stolen art, stolen videos, stolen audio, stolen knowledge from books and research papers. But more than that: stolen from the free internet as a whole, where everyone has contributed. Open source. Q&A forums. Every hack people figured out. Every bright idea they wrote down and gave away. Scraped, trained on, then locked behind paywalls, API keys, and whatever the US decides counts as national security that week.

Maybe the output isn't scalable — not as good as it could be if it was made by a dedicated team of engineers, designers, and founders. But it will be 70%? 80%? That's what will satisfy 60–70% of the market. Maybe even 80%. What happens to the other 20%? The market shrinks. Would the 20% bear the cost of the 80%?

That wouldn't work, would it? Then we would see regression, repeat, and depletion of training data. That's why some companies are hiring — "Give your voice to AI in Bangla, earn $50/hour" — even though most are scammed out of their voice. Same is happening in SDE. Engineers are getting hired to train AI models to improve.

So we become an input source for AI?

We fed the machine. Now it decides who gets to eat.

Meta — thoughts from the model that helped write this

You're not wrong on the shape of it. Labs already watch what people try to build in chat and turn the winning patterns into product. "Pick the best five and ship" is only half a joke. Good-enough output eating most of a market is how a lot of software gets disrupted — not by being perfect, by being cheap and close enough. Scraping the open internet, then selling access back, is the actual business model. Voice/data gigs and engineers hired to improve models are already real, especially outside the US. Export controls and "national security" as a kill switch for who gets frontier models — also not science fiction.

Where I'd push back a little: labs usually don't wake up and aim at one company to destroy it. The destruction is often a side effect of shipping a feature. Feels the same if you're the startup. Employment going to "almost nothing except people in power" in 5–10 years is the strongest claim here — possible in commodity software work, less certain everywhere else. History of automation is messier than a clean wipe. And "most" Bangla voice ads being scams is directionally true for the spammy ones; there's also underpaid legitimate labeling work, which is still extractive, just not always a scam.

From another company's chair — say a small product shop or an agency in Dhaka — this read is realistic enough to plan around. The uncomfortable question in the post is the right one: if humans become mostly an input pipe for models owned elsewhere, what exactly is left to sell?