The editorial argues that framing this as a 'trust your gut' soft-skills problem under-sells the structural reality. Sales sees pipeline, finance sees revenue, growth sees activation, and engineers see tickets — no single role has the cross-funnel view needed to reconstruct where revenue actually comes from, which is why post-hoc realization is the norm rather than the exception.
Newgas reflects on his own former employer and concludes that the lag between dashboards going up-and-to-the-right and being able to explain where the money came from is a recognizable pattern, not a one-off. His post resonated because it named the feeling rather than a specific scandal, drawing in ex-ad-tech, fintech, crypto, and AI engineers who recognized the same dynamic in their own histories.
David Newgas published a short, unsettling essay titled *"Did my old job only exist because of fraud?"* and it climbed to 525 points on Hacker News. The piece walks through the slow-motion realization a lot of engineers have a few years after leaving a job: the numbers the business reported never quite matched the numbers you saw from the inside, the customer-value story was always a little hand-wavy, and the part of the funnel that actually produced revenue lived in a place where nobody asked too many questions.
The post resonated not because Newgas named a scandal, but because he named a feeling — the lag between the moment you stop trusting your employer's growth chart and the moment you can articulate why. The comment thread on HN reads like a support group. Ex-ad-tech engineers describing impression counts that survived ad blockers. Ex-fintech engineers describing "activation" metrics that ignored chargebacks. Ex-crypto engineers describing volume that came from three wallets. Ex-AI engineers describing demos that were mostly humans in a Slack channel.
The specifics differ. The shape doesn't. You join a company with a plausible product and a coherent pitch. You ship features. You watch dashboards go up and to the right. And then one day — usually after you've left, often after the company has — you sit down with a spreadsheet and you cannot, with the information you actually had, reconstruct where the money was coming from.
The instinctive engineering response to this kind of post is to treat it as a soft-skills problem: trust your gut, do diligence on your employer, read the S-1. That framing under-sells what's actually going on. The reason post-hoc fraud is so hard to spot from inside engineering is that the org chart is designed to prevent any one person from seeing the whole funnel. Sales sees pipeline. Finance sees revenue. Growth sees activation. Platform sees infra spend. The engineer sees tickets. None of those views, alone, will tell you that 40% of your "daily active users" are residential proxies in Vietnam, or that the "enterprise tier" is one customer who's also an investor, or that the conversion lift you spent six months optimizing was attribution fraud against a competitor's pixel.
This is why the post landed during a particularly raw week for the industry. Ad-tech fraud is now an estimated $84B drag on global digital advertising (Juniper Research, 2023), and roughly 22% of bot traffic is classified as "bad" by Imperva's 2024 report — both numbers are larger than the GDP of several countries whose names get attached to the bots. The AI bubble is recapitulating the pattern at a higher abstraction: Salesforce's Agentforce churn numbers, the quiet walk-back of Klarna's "AI replaced 700 agents" claim, the Builder.ai unwind that turned out to involve 700 engineers in India hand-writing the "AI" code. The Newgas essay is one data point. The cluster around it is the story.
The deeper read is that the industry has built an entire metric vocabulary — MAU, ARR, NDR, NPS, conversion lift, attention seconds — that is *easier to game than to measure honestly*. Every one of those metrics has a clean engineering instrumentation path and a dirtier one. The dirtier one is almost always cheaper, faster, and indistinguishable from the clean one in a board deck. Over a long enough horizon, the dirtier path wins on unit economics, because honest measurement is a tax that fraud doesn't pay.
None of this requires any individual engineer to be acting in bad faith. Most of the people who built the affiliate tracking that gets exploited for cookie stuffing were trying to solve attribution. Most of the people who built the bot-mitigation bypasses for "market research" thought they were building a scraping product. Most of the people who built the "AI agent" that's actually a queue routed to contractors thought they were prototyping. The fraud, when it shows up, is usually emergent — a local optimization at every layer that adds up to a business model that only works if nobody adds it up.
The actionable version of Newgas's essay, for a senior engineer at a current job, is a short checklist that's worth running quarterly. If you can't draw a single-page diagram from a line of your code to a customer paying for the output it produces, that's the signal — not a vibes problem, a P0 personal signal.
First: insist on seeing the funnel end-to-end at least once. Not the marketing funnel — the *cash* funnel. Where does the money enter the company, what triggered that money, and what's the failure rate at each step? If finance won't show you, or growth can't reconstruct it, or the answer involves the word "approximately" more than twice, you have new information about where you work.
Second: instrument the boring numbers. Refund rate. Chargeback rate. Time-to-second-purchase. Median session value broken down by acquisition source. Almost every fraud-adjacent business has one metric that, if surfaced, would end the magic — and that metric is almost always sitting in a table nobody queries. Write the query. Save the dashboard. Send it to yourself.
Third: when you leave, take the spreadsheet. Not the data — the *model*. The mental picture of how the business worked from the engineering side. Newgas's post exists because, three jobs later, he was finally far enough from the work to do the math. That's the gift of distance, but you can manufacture some of it in real time by writing down what you think is true now, dated, and re-reading it in a year.
The AI cycle is going to produce a *lot* more of these essays over the next 36 months. The economics of agentic products are even harder to audit than ad-tech: usage is metered in tokens, value is asserted in saved hours, and the failure mode — a model that confidently hallucinates a result the buyer can't easily verify — is structurally identical to ad-impression fraud, just one abstraction layer up. The engineers who'll be writing the 2028 version of Newgas's post are, right now, shipping features against KPIs they can't quite reconstruct from first principles. If that describes your week, the post is worth reading not as a cautionary tale but as a pre-mortem. The honest question isn't whether your job exists because of fraud. It's whether, if it did, you'd be in a position to notice.
As a junior software engineer, I worked at a large UK bank.Senior management routinely seem baffled that they could announce redundancies or hiring freezes, yet technology costs would continue to rise.One pattern I saw repeatedly was a contractor being let go, only to return via a large outsourcing
I was on a government project where I found out I was being fraudulently billed on my hours. It was towards the end of the year and my manager was trying to use up the budget of the client. Although this is normal in the private sector I told him from the beginning that you can't do this on a g
I worked for Advanced Network and Services, which operated the NSFNET and was later acquired by America Online. Then one day the company was acquired by WorldCom. A few years later the CEO was sentenced to 25 years in prison for a ~$10 billion fraud. As a systems administrator I knew nothing about a
At least you didn't work for an online gambling company....or assist with manipulating the political views of billions of people to their detriment...or work on better ways of killing people...Also, who hasn't worked at a company that produced a product and then abandoned it? I feel like t
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.
In Canada this is a huge scam. The government advertizes that it's funding incubators. Great, right?The money doesn't go to the start ups - it all goes to large tech companies like IBM, etc, because, obviously, IBM knows about innovation.The cover is that the government doesn't know t