Based on the true story of Soham Parekh’s over-employment.
At 2:15 AM in Mumbai, which is 1:45 PM in Silicon Valley, Soham was in the zone. Several monitors glowed in his ocean-facing apartment, each displaying a different slice of his carefully orchestrated digital existence. Even 18 stories above the chaos below with a gentle seabreeze, it was hot.
Typically, every window AC would be blasting cold air, but that ‘blasting’ proved too loud for conference calls. He was deep in a live debugging session with Jake, a fellow engineer at a San Francisco startup.
The code was elegant; Soham was responsible for most of it. It was a React component that handled real-time data visualization. There had been a problem with it, but Soham had just cut the passive-aggressive blame filled debugging session short.
“Dude, you’re brilliant! How did you find the problem so fast?” Jake said over Zoom.
Soham’s fingers danced across the keyboard. “It just popped out at me, you know?” What he didn’t mention was that it had actually popped out at Claude 4, after it reviewed 800 pages about debugging this type of code.
The startup partnering Soham with a colleague was ironic. They either wanted young Jake, who kept referring to the founders like they were BFFs, to learn from Soham’s experience, or to verify he wasn’t a hack. Soham didn’t sign up to be a teacher and this kid looking over his shoulder was cramping his workflow.
Next to Jake, other screens hummed with automated activity.
One had a classic programming platform in dark mode, but unlike previous incarnations, this could code on its own for hours from a to-do list, like a real developer. Build, test, debug, move on to the next step.
On the other screen, his digital clone was attending a sprint planning meeting with his team at another tech company in Austin. His avatar nodded thoughtfully as the product manager explained user stories.
The fakery was surprisingly easy to set up. He fed a 20 sec video of him on a Zoom call into an open-source AI model he found on Hugging Face, and it created the lifelike Soham avatar. It even nailed his accent. It fooled his mom.
An LLM, with access to the project context and a database of Soham’s typical replies, listens in on the call and formulates the appropriate responses, which are piped through the avatar.
It wasn’t a scam, Soham told himself for the hundredth time. This was ‘efficiency’ in 2025, if you could stomach it.
His family had hit desperate times and Soham was the only one they had to hold everything together. His father’s textile business had collapsed during the pandemic downturn and his mother had needed surgery. His sister needed to be with her newborn twins, not working in a factory. He saved his uncle’s farm, which looks like it’s going to need saving again, and his incredible grandmother deserved a comfortable retirement. He also had a dozen nieces and nephews who would love a university education, though Soham was currently doubting that would be the best investment for their future.
But they all needed his help. They hadn’t exactly asked him for it, but now everyone was depending on Soham’s software engineering skills as their lifeline.
2 years prior, he was fired for telling his boss the truth about these circumstances: specifically the part about taking a second position to help his family. That’s when he discovered the over-employment Sub-Reddit forum. Engineers from around the world were sharing strategies for managing multiple remote jobs simultaneously.
The Americans called it “moonlighting,” but they were thinking small: one extra contract here, some freelance work there. Soham had reimagined the concept like a software architect of distributed systems.
His setup was a masterpiece of modern AI orchestration. MCPs – Model Context Protocols – connected everything via CursorAI, which he used more like an operating system than its original purpose as a programming platform. The Gmail MCP connected Cursor to all his email across 5 different accounts, allowing an AI to sort, label, and draft responses. The Slack MCP linked up dozens of channels, maintaining his presence across all the workspaces. The Google Drive MCP for file management; the Google Calendar MCPs for managing his schedule across the various employers.
This enabled his computers to do work autonomously. It was as if he had a secretary. Or more like 5 versions of himself, each focused on just one company.
Even before AI, Soham could handle the workload of two software engineers. The problem was dealing with everything else that goes with a job and keeping it all straight. Bosses, meetings, co-workers, slack channels, documentation standards, commit conventions… It’s hard to remember all of it for just one job.
What Soham was really proud of, was his knowledge organization system: his databases. Every message, conference call, and email was tagged with the company and sender, plus a plethora of metadata, and finally analyzed and embedded to a knowledge graph and vector database. Then each new message would be analyzed and embedded for a fact and similarity search, that would feed his LLM the right context to respond appropriately.
Most of his work these days involved creating automations, double checking the work of his AI assistants, and remembering to sleep. It was a radical shift from his workflow in the past, but it was all for the best.
The quality of Soham’s work was as good as ever, as long as he kept the details straight. The only thing he was doing wrong, Soham repeated internally, was being so far ahead of the other developers, that they found it threatening.
Sure, putting Georgia Tech on his resume was a lie. But he did get a solid education at The Mumbai University. Plus, he calculated, adding a well known American University to one’s resume triples one’s likelihood of landing an interview. He was just compensating for the Americans’ bias.
He also found, if you live in Mumbai, they’ll offer you a third the salary than if you set up a VPN and say you’re in New York. Geographic arbitrage was a strategy that built the global economy: why would he not leverage that advantage for his family?
“Tech companies were like schools”, is how he explained it to his brother. “Everyone’s expected to proceed at a snail’s pace regardless of proficiency.” With AI’s help, he could complete an entire sprint’s worth of work in a day, then spend the rest of the days maintaining the illusion of normal productivity across multiple employers, so no one would fire him.
The moral calculus was simple: should he really not take work he could do, just because it was ‘moonlighting’?
He did admit, ‘Moonlighting’ feels like the perfect term when your colleagues think you are in their country, but you were actually working all night in Mumbai.
The system wasn’t flawless, especially on the 20th hour running only on coffee and samosas. But he was currently working 5 software engineering positions at 5 different tech companies, all convinced they had secured the exclusive services of a brilliant engineer. The combined income – about $43,000 this month – had transformed his family’s circumstances. Soham was the golden child. His mother was receiving proper medical care. His sister was home with her newborns. His grandmother had regained her dignity.
He couldn’t stop.
“Wait, Soham,” Jake said, his voice carrying a note of professional curiosity that made Soham’s chest tighten. “This is way different than before. Looks like you refactored it entirely. What did you use to do this?”
Soham’s fingers froze over the keyboard. He fixed the code too fast! Jake, of all people, had caught him.
Professional developers had grown accustomed to coding with autocomplete. You could ‘TAB’ your way through a block of code without peers batting an eye. Of course, autocomplete requires the programmer to hover over the code and verify each step.
Using the latest AI tool to do the bulk of the work, entirely unsupervised, was apparently even too edgy for developers right out of school. Was he being tested?
For an MVP like this, most of Soham’s process was automated. The Figma MCP connects the AI with the designer’s concepts. The hottest AI currently available writes the code in Cursor, with the Context7 MCP for the latest documentation. Testing is a matter of prompting a complete list of possible use cases, then prompting it to use the Playwright MCP to run all the tests, fixing any problems as they’re encountered. 97% of the work is done by AI using MCPs.
Apparently, when the AI fixed the code, it did a full overhaul.
From the beginning of the debugging call with Jake, the Californian had been talking non-stop about what he thought the issue could be, which could have been an essay called ‘it’s not my fault’. Soham opened three instances of Cursor, popped the buggy program in all of them, and prompted each to attempt to solve the problem using a different approach.
Instead of sitting there listening to Jake, he opened Whatsapp where hundreds of unread messages from a dozen family chat groups were waiting.
After several minutes, one of the Cursors had a version that was passing all the tests. The site was now loaded quickly. The basic functions all worked and it looked right.
Getting Jake to end his unfiltered self-exoneration was growing in priority, plus he wanted to turn the AC back on. With sweat beginning to bead on his brow, Soham had simply declared the success to Jake and pushed it to the server. He assumed Jake would be relieved and that would be the end of it. But he was wrong.
He should have looked it over. He should have gotten some sleep before his nightly day shift. It was these little screw-ups that kept Soham under $50k per month. His heart was racing, but his face hadn’t given anything away yet.
“I was trying out this new tool. I’m surprised it worked at all.” Soham smiled, hoping it didn’t look as forced as it felt. “Now I gotta figure out how that thing fixed it, then rewrite it properly.”
It was another little-white-lie: the code was ready and he wasn’t going to work on it anymore, except prompting the AI to add Jake’s comments back in. Anyways, how many little-white-lies did the tech bros tell to score venture capital for their mediocre app idea?
Jake seemed frozen; might be the connection.
Soham hardened. If Jake was going to rat him out, that was fine. These jobs were expendable.
Soham had found a free, customizable n8n automation that searched out ideal job openings, applied for them, and lined up interviews. He got the impression his new avatar was more likeable than his real self, and figured it was probably time to automate the interview process too.
Jake’s serious gaze finally broke. “Slick bro! I needed that last night! What AI’d you use?”
Soham’s relief lasted only a second as his phone began buzzing aggressively. It was his avatar warning system!
“I’ll send you a link. And I’ll take care of the code, but I have something urgent right now Jake.” He ended the call, cutting off Jake’s “cool man”, and rolled the chair toward his other computer.
Marcus, the project manager in Austin looked pissed. He was leaning in, saying, “what are you talking about”, which had triggered the AI alarm. Soham unmuted the avatar who was calmly responding with its head tilted. The transcripts looked right: build features, recommendations for better UX…
No! This was the wrong project!
He was piping in the SF project, to the Austin guys who were expecting details about a payment system. He’d finished that system an hour after it was assigned, a week ago, but he must have saved it in the wrong file. If only he’d automated that step too!
Retaking live control of an avatar was doable, but jarring. He yanked off his t-shirt and pulled on the white button-down that his avatar always wore. He ran his fingers through his hair, whiped his brow, and composed himself. No time for further inspection. A keystroke toggled from the avatar to the live camera, as he tilted his head to mimic his double.
“Sorry about that sir. I got confused, but I’ve finished the payment system.”
“Soham, you just spent five minutes explaining a front-end system that has nothing to do with our project” Marcus insisted.
“I know sir, but your assignment is completed. I was just confusing this with…” He couldn’t think of anything to say but ‘another project’. Last time he said such a thing to a boss, he was let go the next day.
Marcus finished it for him. “Another project? We need you 100% focused on this Soham.”
“I actually finished the payment system last week and was just waiting for..” He realized this excuse didn’t make anything better.
Marcus was already on to the next red flag anyhow. “Weren’t you clean-shaven at the beginning of this meeting? Now you have a beard and you’re all sweaty!”
Soham tried to remember that last time he shaved. He wanted to turn the avatar back on, but he still hadn’t switched over the context. And now the beard!
“I think we need to have a conversation with HR,” Marcus said sternly.
Soham couldn’t handle it any longer. He nodded in agreement and killed the video feed. He rolled, reached, and twisted a knob to ‘AC Max’. Then straightened up in the cold air and felt his stubble; maybe a week?
His turnover rate when he first started this game was high, but it was the first time he was busted over a beard. Maybe an alarm to shave every other day? Better yet, update his avatar with a beard.
A third screen, which had been turning out code for 20 mins, was now motionless and proclaiming “Project complete. Zero errors.” He closed that window and opened an n8n automation, typed ‘$100,000’ in a field titled ‘salary’, set ‘confirmed interviews’ to ‘3’, and toggled the automation to ‘on’.
Then turned to his primary machine. Another meeting was scheduled in 20 minutes with a Seattle startup and he needed to make sure there were no crossed wires. And he had to work fast, as he also needed a new, sweatier, bearded avatar.
Soham didn’t care what some tech bros on the other side of the world thought of him. He was doing this for his family, and they deserved the best. Just a few more tasks to automate and he might also be able to spend time with them.