The AI Startup That Was Just Humans in a Trench Coat

Back in 2019, an Indian startup called Engineer.ai was making some pretty bold claims. It said it had built an AI-assisted app development platform that could whip up 80 percent of a mobile app from scratch in about an hour. The founder, Sachin Dev Duggal — who also went by the title “Chief Wizard” — was out there on conference stages selling a vision of automated app creation that sounded almost too good to be true.

It was.

A Wall Street Journal investigation revealed that Engineer.ai wasn’t really using AI to build apps. Instead, it was mostly relying on human engineers in India and elsewhere to piece together the code manually. The AI part of the equation was, at best, aspirational. At worst, it was a straight-up fabrication designed to attract the nearly $30 million in funding the company had already raised from SoftBank’s venture arm and other investors.

The company’s own chief business officer, Robert Holdheim, sued Engineer.ai earlier that year, claiming Duggal was telling investors the product was 80 percent done when it had barely been started. That’s a pretty damning accusation coming from someone who was supposed to be running the business side of things.

When the WSJ pressed Engineer.ai on how it actually used machine learning, the company said it employed natural language processing to estimate pricing and timelines, and used a “decision tree” to assign tasks to engineers. Neither of those is modern AI in any meaningful sense. No AI agent was compiling code. No neural network was designing user interfaces. It was just humans with project management tools, wrapped in a layer of tech buzzwords.

This kind of thing is more common than you’d think. A 2019 study by MMC Ventures found that startups claiming some AI component could attract up to 50 percent more funding than other software companies. The same firm estimated that 40 percent or more of those companies didn’t actually use any real AI at all. The .ai domain — from the British territory of Anguilla — doubled in registrations over the preceding few years. Slapping that on your company name was basically a shortcut to investor attention.

And the money followed. Global funding for AI startups hit $31 billion in 2018, according to PitchBook. SoftBank alone had pledged hundreds of billions in AI investments. The logic was simple: if you’re building a traditional app development platform, you’re competing with every other startup in a crowded space. But if you say you’re building an AI-powered app development platform, suddenly you’re a visionary solving a hard problem. Investors eat that up.

The Engineer.ai story also highlights a darker truth about a lot of “AI” in the real world. It barely exists. Think about content moderation on platforms like Facebook and YouTube. They use some AI to flag problematic content, but the real work is done by armies of contractors — many of them overseas — reviewing videos and posts for hours on end. The AI is a thin veneer over a massive human operation.

Same thing happens in customer service chatbots. A lot of those “AI-powered” bots are actually humans typing responses behind a curtain. The AI just routes the conversation and occasionally suggests a canned reply. When the bot gets confused, a human jumps in. That’s not automation. That’s a call center with a nicer name.

Getting AI to work at scale is brutally hard. You need massive amounts of high-quality training data, which is expensive and time-consuming to collect. You need top-tier researchers who command six-figure salaries. And even then, you might end up with a system that works well in the lab but falls apart in production. Companies like Google and Facebook have entire research organizations dedicated to this, and they still struggle to deploy reliable AI in many areas.

For a startup with $30 million and a Chief Wizard, the temptation to cut corners is huge. Why spend years building real AI when you can hire humans, call it “human-assisted AI,” and collect the checks while you figure it out later? It’s a strategy that works until someone sues you or a journalist digs into your claims.

Engineer.ai isn’t an isolated case. It’s a symptom of an industry that rewards hype over substance. Investors want to fund AI companies because AI is the hot thing. Founders know this, so they stretch the truth. The result is a market flooded with startups that are heavy on buzzwords and light on actual technology.

I’ve been watching this space for years, and the pattern is always the same. A startup announces some amazing AI breakthrough. The press writes breathless articles. The funding rounds get bigger. Then, eventually, someone looks under the hood and finds a bunch of humans doing the work. The startup either pivots, gets acquired, or quietly fades away. Rinse and repeat.

The real tragedy is that genuine AI research gets tainted by association. When people hear “AI startup” now, a lot of them roll their eyes. They’ve been burned too many times by companies that promised the moon and delivered a spreadsheet with a chatbot interface.

So what’s the takeaway? If you’re an investor, be skeptical of any AI claim that sounds too good to be true. Ask for specifics. Ask to see the model. Ask how much of the process is actually automated versus human-assisted. If the answer is vague or defensive, walk away.

And if you’re a founder building real AI, don’t take shortcuts. It’s harder, slower, and less glamorous. But at least when someone looks under the hood, they’ll find something that actually works.