In Defense of

On September 4, 2019, The Wall Street Journal published an investigation into a UK-based startup called Engineer.ai. The article was headlined "AI Startup Boom Raises Questions of Exaggerated Tech Savvy." Its central finding: the company's artificial intelligence was mostly humans. Specifically, about 700 humans in Bangalore, writing code manually while the marketing department told investors it was AI. [WSJ]

Anyone reading that article in 2019 could reasonably conclude that Engineer.ai — which later rebranded as Builder.ai — was a fake AI company. This is because it was, in fact, a fake AI company.

In May 2023, Microsoft made a strategic equity investment and partnered with it anyway. [TechCrunch]

Not a small deal. A "strategic collaboration." An integration with Teams. An equity stake of undisclosed size. An announcement featuring Microsoft's public endorsement of a company that had been publicly debunked four years earlier. Microsoft said, and this is a direct quote: "We see Builder.ai creating an entirely new category that empowers everyone to be a developer." [GlobeNewsWire]

Two years later, that new category turned out to be Business Process Outsourcing with better marketing. Builder.ai filed for bankruptcy in May 2025. [TechCrunch] Microsoft was left holding $30 million in unpaid Azure bills. Amazon was left holding $85 million in unpaid AWS bills. [DevOps.com] Investors were left holding a $450 million pile of ash.

The AI was Anil. And Rajesh. And Priya. And 697 other engineers in Bangalore.

The Warning Microsoft Read And Ignored (2019)

The Wall Street Journal's 2019 investigation was not subtle. It reported that former employees had described Builder.ai's "AI" as being 80% manual coding by human developers. It quoted internal sources who admitted the artificial intelligence didn't really exist in any meaningful way. The company denied the allegations, then quietly changed its name from Engineer.ai to Builder.ai, which is the corporate equivalent of getting a haircut after committing a crime.

Robert Holdheim, a former Builder.ai executive, sued the company for $5 million in 2019. His allegation: he was fired for flagging that the AI was fake. [DevOps.com]

These facts were public. Anyone could Google them. In fact, anyone who Googled the phrase "Builder.ai AI real" in 2019, 2020, 2021, or 2022 would have found extensive documentation that the AI was not, in the technical sense of the word, "real."

Microsoft's due diligence process famously involves layers of legal review, technical audit, financial verification, and compliance screening. Somewhere in that process, apparently, nobody typed Builder.ai into a search engine.

The 2023 Deal Nobody Should Have Made

In May 2023, Microsoft announced a strategic collaboration with Builder.ai. The press release used phrases like "empowering everyone to be a developer" and "fueled by Azure AI." It featured a joint statement from Builder.ai CEO Sachin Dev Duggal, who called it "transformational." [Windows Central]

What Microsoft's due diligence team should have known but apparently didn't:

1. Duggal was under investigation. Public records from India linked him to a Videocon money-laundering probe. This was documented in Indian financial press. Microsoft's compliance screening did not flag it.

2. The auditor was compromised. Builder.ai's audit firm employed a director who also worked for one of Duggal's other companies. This is called a conflict of interest. It's the kind of thing you notice if you're paying attention. Microsoft was not paying attention.

3. The revenue was fiction. Builder.ai's 2023 revenue was later revealed to have been overstated by 300%. Reported: $180 million. Actual: $45 million. Forty percent of reported revenue came from reseller contracts in the UAE and Qatar, many of which did not have end customers. They were, in the technical sense, imaginary. [TechStartups]

4. The AI was still humans. This one had been public knowledge since 2019. Microsoft did not verify. Microsoft did not test. Microsoft did not ask.

Meet Natasha, Your AI Assistant Who Lives In Bangalore

Builder.ai's flagship product was called "Natasha." She was marketed as an AI assistant who could build software "as easily as ordering pizza." Business owners could describe what they wanted. Natasha would build it. Six times faster than traditional development. Seventy percent cheaper.

Natasha was not a chatbot. Natasha was not a large language model. Natasha was not, in any technical sense of the word, artificial intelligence.

Natasha was a workflow. A form on a website. When a customer described what they wanted, the description was routed to a project manager, who assigned it to human developers — mostly in Bangalore — who wrote the code manually. The output was then handed back to the customer under the branding of "AI-generated software."

The technical term for this arrangement is "outsourcing." It has existed since approximately 1962. Its innovation cost is zero. Its labor cost is significant. Its AI content is also zero.

If your neural network requires health insurance, vacation days, and biryani for lunch — congratulations, you invested in humans.

To be clear: there is nothing wrong with outsourcing. It is a legitimate business model that has employed millions of people and delivered billions of dollars in value to customers who were transparent about what they were buying.

Builder.ai's customers thought they were buying AI. This is called fraud.

In its defense: AI is hard to verify

You could argue that in 2023, distinguishing "real AI" from "elaborate outsourcing" was genuinely difficult. GPT-3 was recent. GPT-4 was being launched. Everyone was calling everything AI. Even ChatGPT was, at some technical level, mostly statistical pattern matching dressed up as intelligence. Where does the AI end and the marketing begin?

This is a fair question if you're a first-year business student. It is not a fair question if you're Microsoft, the largest software company in the world, with 220,000 employees, an in-house AI research division, exclusive access to OpenAI, and a due diligence budget that could purchase a mid-sized country. Microsoft could have tested Natasha. Microsoft could have asked a Microsoft engineer to spend one afternoon evaluating whether Builder.ai's AI was real. Microsoft did not.

The Collapse (May 2025)

By early 2025, Builder.ai had become impossible to sustain. Its cloud bills were catastrophic: over $100 million owed to AWS and Azure combined. [TheStreet] Its revenue was a fraction of what had been reported to investors. Its CEO was under federal investigation.

Then Viola Credit, a lender that had extended $50 million in 2023, seized $37 million from Builder.ai's accounts. What remained was $5 million in restricted Indian bank accounts — meaning Builder.ai couldn't legally access the money to pay employees, let alone Amazon or Microsoft. [Yahoo/Benzinga]

On May 20, 2025, Builder.ai filed for bankruptcy across five jurisdictions: UK, US, UAE, Singapore, and India. [TechCrunch] Microsoft quietly removed all promotional materials mentioning Builder.ai from Azure marketing pages. Their strategic collaboration announcement, once trumpeted as evidence of Microsoft's AI leadership, was deleted from press archives where possible.

Federal prosecutors in New York opened a fraud investigation. They are, as of publication, still requesting customer and financial documents. [Windows Central] CEO Sachin Dev Duggal had been replaced in February 2025 by Manpreet Ratia, whose stated role was "restoring investor confidence." He did not restore investor confidence. He was, instead, the person who signed the bankruptcy paperwork.

The $115 Million Cloud Bill

Microsoft's exact equity investment in Builder.ai was never disclosed publicly. But we know Microsoft joined a Series D funding round that pushed total funding past $450 million. The Series D was led by the Qatar Investment Authority (QIA), which contributed the largest portion. Other investors included SoftBank's DeepCore, IFC (the World Bank's private-sector arm), Lakestar, WndrCo, and Insight Partners.

What we do know precisely: Microsoft is owed $30 million in unpaid Azure cloud services. Amazon is owed $85 million in unpaid AWS bills. [DevOps.com] [TheStreet] Both companies had been running Builder.ai's infrastructure on credit, apparently in the belief that a $1.5 billion unicorn would eventually pay its bills.

These bills will never be paid. Builder.ai's remaining assets are being liquidated to pay administrators, and there is no realistic scenario in which Microsoft and Amazon recover their money.

Meanwhile, the equity investment — the "strategic" part of the "strategic collaboration" — is worth zero. Whatever Microsoft paid for its stake in Builder.ai is now, for accounting purposes, an impairment charge. A rounding error in Microsoft's annual $80 billion in profit. But an impairment charge that was preventable with one search on Google.

In its defense: Microsoft is very large

Microsoft's annual revenue is approximately $245 billion. An unpaid $30 million Azure bill from a fake AI startup represents 0.012% of their annual revenue. For context, this is roughly what Microsoft spends on office snacks in a bad quarter. If you're going to fumble on due diligence, doing it at a scale small enough to be invisible on your balance sheet is at least strategically consistent.

The problem isn't the money. The problem is what the money says. Microsoft's endorsement of Builder.ai was worth more than the $30 million in unpaid Azure services. It was a public signal to other AI startups, other enterprises, other investors: Microsoft has evaluated this company and found it worthy. Microsoft had not evaluated. Microsoft had signed a marketing deal.

The AI-Washing Era

Builder.ai's collapse coincided with what analysts began calling "the AI-washing correction." AI venture capital funding dropped 15% in Q1 2025. Stability AI's valuation fell from $4 billion to under $1 billion in the same period. Several other AI startups quietly disclosed that their "proprietary AI" was, in fact, calls to OpenAI's API with a wrapper.

The pattern was consistent: raise money on AI hype, spend the money on cloud bills to power the illusion of AI, run out of money before delivering actual AI, blame market conditions.

Microsoft was not alone in getting bent over by AI-washing. But Microsoft was, uniquely, the company with the most resources to detect it — and the most public commitment to being "the AI company." Every time Microsoft endorsed a fake AI startup, they weakened their own credibility as the AI industry's judge of quality.

By 2026, when Nadella himself would publicly admit that Microsoft's own Copilot integrations with Gmail and Outlook "don't really work," the pattern was complete. Microsoft could not verify AI in other companies. Microsoft could not build working AI in its own company. Microsoft was, in the AI era, precisely the kind of hostage they had themselves become — held by a technology they had championed publicly and could not evaluate internally.

Microsoft's problem is not that they invested in Builder.ai. Microsoft's problem is that they still don't know what real AI looks like.

The Truth Is Even Funnier

After Builder.ai collapsed, a viral narrative took hold across tech media and social media: "Microsoft invested in a company that faked AI with 700 engineers in India." Industry cynics began joking that in Builder.ai's case, AI stood for "Another Indian."

It's a great headline. There's one problem: it's not entirely true.

Gergely Orosz, who runs The Pragmatic Engineer — one of the most respected engineering newsletters in the industry — spoke with several former Builder.ai engineers and published a detailed investigation in June 2025. His finding: Builder.ai's AI was real.

The company had a team of 20 internal developers building Natasha — a genuine AI system that included a code generator, a chatbot, a knowledge graph built on vector databases, and a benchmarking pipeline that tested new LLMs as they came out. They used Ruby on Rails for the web components. They ran coding benchmarks to choose the best model for each use case. By 2024, when LLMs were more than capable, Natasha was using them extensively.

The 700 engineers? They were an outsourced development network — contractors from firms like Globant and TatvaSoft, spread across India, Vietnam, Romania, Ukraine, and Poland — building actual client apps. That's not "faking AI." That's software development services. It's a legitimate business model that has existed since the 1960s.

What was fake was everything else.

The revenue — overstated by 300%. The reseller contracts in the UAE and Qatar — many without end customers. The CEO's clean record — he was under investigation for money-laundering. The auditor — employed by one of the CEO's other companies. The entire financial structure of a $1.5 billion unicorn — built on circular contracts, inflated projections, and institutional willful blindness.

Builder.ai didn't fake the AI. They faked the revenue, the customers, the books, and the CEO's integrity. The AI was the only honest part of the operation. Everything around it was the fraud.

Which makes Microsoft's fumble even worse. Because Microsoft didn't need to verify the AI — the AI worked. Microsoft needed to verify the finances. The books. The contracts. The CEO's legal history. The auditor's independence. These are things that show up in a standard due diligence process. They are not technically complex. They do not require machine learning expertise. They require an accountant and a Google search.

Satya Nadella — the CEO of the world's largest software company, born in Hyderabad, educated at Manipal and the University of Wisconsin, with three decades of experience in enterprise technology — invested Microsoft's money in a startup whose fraud had been publicly documented by the Wall Street Journal four years earlier.

He didn't need due diligence. He needed Google. Which, as the CEO of Microsoft, must have been physically painful to open.

Next in the Bent Over Club →

Chris Freeman

"He said no to Australia. He trusted his friends. He sued for $400 million. The handshake felt real. Freeman watched his Stake.com founder equity walk out the door — and hired a lawyer six years too late."