I knew Wall Street would be an early adopter of AI, but even I didn’t expect Goldman Sachs to start onboarding robot coders before most of us figured out how to use ChatGPT without embarrassing ourselves. The bank is now piloting an autonomous software engineer named Devin, created by a startup called Cognition, which claims it’s the world’s first AI capable of handling complex, multi-step coding tasks. Let that sink in: Goldman, a place where junior analysts famously work 100-hour weeks to manually tweak PowerPoints, is suddenly outsourcing real engineering work to machines. The irony is almost too perfect.

Here’s the part that should terrify anyone who’s ever debugged a line of code for a living. Marco Argenti, Goldman’s tech chief, casually mentioned that this AI might scale from hundreds to thousands of ‘Devin’ instances, depending on use cases. That’s corporate speak for ‘we’re figuring out how many jobs we can vaporize without collapsing morale.’ The bank insists this is about ‘augmentation,’ not replacement, but tell that to the engineers now babysitting algorithms that could eventually make them obsolete. I’ve seen this movie before. First, AI ‘assists’ with tedious tasks like updating legacy systems. Next, it’s handling entire projects ‘with supervision.’ Then, one Tuesday morning, some VP declares a ‘pivot to AI-first workflows,’ and suddenly half the team’s badges stop working.

The cognitive dissonance in tech right now is staggering. On one hand, companies like Microsoft brag that AI writes 30% of their code, framing it as a triumph of innovation. On the other, they refuse to acknowledge the obvious corollary: if AI can do 30% today, it’ll do 70% in two years. Salesforce’s Marc Benioff already claims AI handles half of his company’s work. At what point do we admit we’re not just ‘augmenting’ labor but actively constructing its replacement? Goldman’s experiment is particularly poignant because finance has always been ruthless about efficiency. If a $4 billion AI startup can shave basis points off trading latency or reduce IT costs by automating grunt work, human coders become a variable expense to optimize. Just ask the 200,000 global bank employees Bloomberg predicts will lose jobs to AI by 2030.

What fascinates me most is the narrative gymnastics required to sell this transition. Argenti describes a ‘hybrid workforce’ where humans ‘supervise’ AI agents. Translation: employees will now spend their days prodding black boxes to do their former jobs, like gym teachers overseeing dodgeball games played entirely by robots. The real skill set shifting isn’t in coding anymore, but in prompt engineering and damage control when the AI inevitably hallucinates a stock trading algorithm that tries to short the entire S&P 500. Remember when ‘learn to code’ was the refrain for displaced factory workers? Now it’s ‘learn to coax coherent output from an LLM before it eats your career.’

Let’s not pretend this is unique to Goldman. The same week this news broke, Amazon admitted it’s using AI to auto-fire underperforming Flex drivers. Ford’s CEO openly discussed ‘rightsizing’ headcount thanks to AI efficiencies. The velocity of corporate adoption here is breathtaking. Cognition, the maker of Devin, doubled its valuation to $4 billion in a year, bankrolled by silicon valley sharks like Peter Thiel. The incentives are painfully clear: automate fast, cash in faster, and let society figure out the fallout later. I don’t blame the engineers building these tools—they’re playing the game as it exists. But when the history of this era is written, we’ll marvel at how quickly we normalized replacing skilled labor with probabilistic parlor tricks.

The bottom line? Goldman’s Devin pilot isn’t just a tech milestone. It’s the canary in the coal mine for knowledge workers everywhere. If complex programming jobs aren’t safe, neither are consulting reports, legal briefs, or financial models. The only real question is who’s next. And whether we’ll even need human journalists to write about it when the time comes.