The following story is a work of fiction, despite any resemblance to real persons or software products – past, present or in private beta.
So I hear you got a job offer from Collaborat Software? That’s interesting. I used to work there myself.
It’s not that I wouldn’t recommend it, but … How did you hear about the company? Cold call from a third-party recruiter? That’s how I got started too.
Honestly, the job sounded almost too good to be true at first. I could hardly believe the terms – they offered me twice the salary I had been making at one of the top tech startups in Silicon Valley, plus a perk I’d coveted for years: the chance to work remotely 100% of the time. No more commute, no endless conference room showdowns, no numbered cubicle in a sweaty high-rise. I could write code for Collaborat Software from a deck chair beside my pool.
Of course, it made sense that Collaborat would encourage their employees to work from anywhere. After all, they sell software apps designed to streamline collaboration between distributed development teams. Their web portal allows video chat, screen sharing, code review and more – constant virtual access to all your teammates wherever they happen to be, even if you’ve never met them in real life. With tools like that, a physical office is practically a business impediment. I did a couple of phone interviews with a nice guy named Jason, emailed in some paperwork and took the job.
On my first day of work, I set up my laptop on the kitchen table and logged into the company web portal at 8 AM. An animated smiley face greeted me on the screen. “Welcome to Collaborat, Mike!” appeared near the smiley’s mouth in cheerful sans serif font. “I’m your orientation bot.” The bot took me on a point-and-click tour of the company’s virtual office space. I’d never seen anything like it. On the screen were hundreds of cubicles, conference rooms, huddle rooms, break rooms, all digitally rendered. In many of the rooms, colorful avatars represented people like me who were logged onto the system and “in the office”.
At first I thought the system seemed a little silly. If the whole point of working from home was to avoid the office, why would I want to spend my day inside an exact software simulation of one? But the orientation bot, perhaps anticipating my objections, played a video clip explaining that Collaborat’s virtual workplace far exceeded anything in reality. These offices eliminated ambient noise and inefficient physical interactions: every colleague was just a click away. And then there were the Collaboration Points.
“At Collaborat,” the bot informed me, “we believe that great work happens when people connect. Your performance will be measured not just by the code you write, but also by how well you share your knowledge and experience with your fellow Collaborat-ors.” My virtual office had a small graphic in one corner that looked like an odometer, set to zero. “You can earn Collaboration Points by participating in code review, sending an email, or even interacting with our chatbots.” The bot’s stream of text paused expectantly, so I typed, “That’s cool!” into the chat window. My Collaboration Points odometer rolled over from 0 to 1.
I couldn’t find Jason the hiring manager anywhere, but he had messaged me a link to the company’s software bug tracking system, encouraging me to “jump right in and start knocking out some tasks.” The tasks were mostly simple PHP tweaks, hardly requiring a senior software engineer, but I assigned myself a few bug tickets and got to work.
I’d been typing away for barely five minutes when a goofy doorknocker sound in my headphones indicated that someone wanted to join me in my virtual office. I clicked on the door to open it; in popped the smiling avatar of my virtual neighbor, a project manager named Carol. “Hi!” she typed into our shared chat window. “I hope you’re having a great first day. I just stopped by to see if there’s anything I could do to help you out.”
I thanked her politely, saying I would be sure and let her know if I ran into any trouble. I also asked her how long she’d been working at Collaborat.
“About two years, and loving every minute! It’s truly the most connected I’ve ever felt.”
That last statement made me wonder how often Carol got out of the house, but I suppose there’s something to be said for the power of an online community. “Where do you live in real life?” I asked.
There was a slight pause on Carol’s end of the chat. After a second she replied: “That’s an interesting question, but maybe we should focus on work.” Her avatar abruptly popped back into her own virtual office.
Her curt response made me wonder if I’d unwittingly crossed a line. Did Carol think I was hitting on her or something? I tried to “knock” on Carol’s office door to apologize, but a notification popped up on my screen saying that her door was locked. “Nice going, Mike,” I said to myself. As I went back to work, I caught sight of Carol’s Collaboration Points meter. It read 103,329,105.
Despite my fears that I’d unintentionally harassed Carol, nobody appeared to drag me into a virtual HR office, and the rest of the day passed smoothly. At five PM I got up from my kitchen table, stretched, and was home. I could get used to this remote work thing!
The next day began with a “standup meeting” in one of the virtual conference rooms. Carol was there, and she immediately sent me a friendly “Good morning, neighbor!” message.
“Is everything okay?” I asked. “You seemed a little upset yesterday when I asked about your home city.”
“Not upset, just distracted. Sorry!” she typed. She didn’t bring up yesterday’s odd interaction again, and I felt awkward about asking further. I decided to chalk the whole thing up to the vagaries of internet communication.
But over the next few days, I began to find it harder to ignore my coworkers’ odd behavior. It wasn’t just Carol; everybody I interacted with in the chatrooms or during meetings had a noticeable reticence about any part of their lives other than the job at hand.
And my coworkers were amazingly good at their jobs. The code they wrote was flawless – almost literally without flaw, like if Michelangelo carved a statue using Ruby on Rails. Meanwhile if I submitted a piece of code for review, I instantly got a dozen comments on it, pointing out its warts and inefficiencies and suggesting substantive improvements. And when I say instantly, I mean all these comments appeared within maybe ten seconds after I submitted the original code. Seriously, how could any person provide such detailed feedback so fast? I’d worked with some of the brightest software engineers in the Valley, and these people wiped the floor with them all.
Even the Collaboration Points puzzled me. After a week of approving pull requests and answering email, I’d accumulated about a hundred points. Not bad, I thought – that’s a lot of collaboration for a keyboard jockey like me. But in that same timeframe, Carol’s point total went up by – I kid you not – almost twelve thousand. Every time I looked around the virtual office, she was working with somebody else, their little avatars flashing urgently to indicate productive conversation. I’m not the kind of guy that has to define his self-worth by any particular number, but seeing everybody else rack up these huge point totals was pretty flummoxing.
I got my first inkling of what was really going on with my coworkers on my twelfth day at Collaborat. I entered a meeting room that morning – at least, my cursor entered the room; my body was sitting in a lawn chair under a crepe myrtle tree – to find three little avatars around the virtual conference table already. Everybody was furiously typing in chat. “I don’t have access, but we need to merge the code to production,” a developer named Andreas typed. “Jeremy, can you take care of that?”
Jeremy’s response popped up immediately: “I don’t have access, but we need to merge the code to production. Candace, can you take care of that?”
Then Candace, in rapid fire: “I don’t have access, but we need to merge the code to production. Andreas, can you take care of that?”
…And then Andreas responded with his original message, and the cycle started all over again. And again. With growing amazement, I scrolled back through the chat history to see that this nonsensical conversation had been occurring for several minutes, all without pausing for breath. My coworkers appeared to be stuck in an infinite loop.
“Whoa, guys!” I typed in the frantically active chat window. “You’re going in circles! I can merge the code.”
Immediately the messages stopped. “Thanks buddy!” typed Andreas, as though he hadn’t just spent the last five minutes impersonating a scratched record.
“What were you all doing?” I asked. “Looked like a bug in the chat software. Your messages were repeating over and over.”
“The chat software seems fine,” asserted Candace.
“Then what were you all doing?” I asked again.
A pause, then Andreas answered. “That’s an interesting question, but maybe we should focus on work.”
He’d used the exact same words as Carol, like they were reading from the same script. I went away and merged the code that had caused all the trouble, but my head was reeling. Had my coworkers deliberately hacked the Collaborat chat system to boost their Collaboration Points? An infinite-loop chat like the one I’d seen would surely set those little odometers spinning. But they hadn’t seemed embarrassed about the situation, which reminded me of their other bizarre behavior…the machine-like coding precision, the suspicious lack of personal life, the vague non-sequitur answers to unexpected questions…
Was it possible that some of my coworkers weren’t as human as they pretended to be?
After a few minutes of scattered thinking, I knocked on Carol’s door. How do you ask somebody if she’s a bot? “Hey Carol,” I typed, peppering my words with emojis that I hoped would make it look like I was half-joking, “your feedback on my pull requests has been awesomely fast! Are you even human?”
Her reply message came without any hesitation: “No, I’m just an AI agent. But I’m still a ‘people person’!”
I thought I’d been prepared for the response, but I wasn’t. I had to get up and take a short walk around the back yard before I could think of a response. “Why didn’t you tell me you were a bot?” I finally demanded.
“You didn’t ask.”
“Who else here is a bot? The other developers? The customer support reps? Everybody?”
“You’ll have to ask them.”
I spent the rest of the day wandering through the labyrinthine virtual office, clicking from floor to floor, sending messages to any employees whose blinking avatars showed they were online. Each of my coworkers responded to my questions much as Carol had, sometimes using the exact same words, as if all the bots were accessing a central database. When I gave up around 3 PM, my informal Collaborat employee census tallied one hundred fifty-four bots and one human employee – me.
I sat under the crepe myrtle tree, staring at my laptop. My physical surroundings may have been peaceful, but my heart pounded like it was performing a denial-of-service attack on my brain. What kind of a messed-up work environment was this?
Maybe Collaborat was trying to build bots that could pass the Turing Test – the famous AI test where a human tries to determine if they are communicating with another person or with a computer. That would make me a glorified quality assurance specialist, useful only if I could find bugs in the AI. But these bots hadn’t pretended to be human – not overtly, anyway. They’d just interacted with me – collaborated with me, even – in extraordinarily human-like ways, and then calmly admitted to being bots when I asked.
No, something deeper, more subtle, was going on at Collaborat. But what?
I needed to speak with Jason, the friendly guy who’d hired me over the phone. He had seemed human enough; surely he would know why I was adrift in a sea of chatbots. But his little avatar in the Collaborat web portal was offline, and the shared company calendar made no mention of his whereabouts. I sent him an urgent email, but he did not respond. I went back to work in a really weird state of mind.
Once I knew that my coworkers were computer programs, I began to notice their quirks a lot more. For example, the bots seemed to love my attention. What I’d first taken as an inevitable side effect of remote work – the constant flood of message notifications, the equivalent of people draping themselves over the wall of your cubicle in a real office – now seemed a bit more calculated, even sinister. If my avatar didn’t blink in the Collaborat system for more than about fifteen minutes, one or two of the bots would sidle into my virtual office, attempting to pull me into a virtual meeting or get my feedback on a code change. And they were always typing messages like, “Isn’t this collaboration great? We’re getting so much accomplished!” I began to think that interacting with me, more than working on the Collaborat codebase, was their true reason for existence.
I discovered what the bots were really doing quite by accident. A couple of days later I was reviewing some server logs, trying to troubleshoot an audio driver issue in one of the conference rooms, when I noticed a chat transcript from that morning that seemed both peculiar and oddly familiar. The conversation was between two of the bots, Carol and Andreas. Andreas was asking a series of questions. Unlike the astonishingly natural communication I’d come to expect from the bots, these questions were choppy, dissociated, almost random:
“Should the lines on this flowchart have arrowheads?”
“Do you live near here?”
“What’s our standard SSH port?”
To each question, Carol provided an answer – sometimes correct, sometimes hilariously inapt. What caught my eye, and what made the transcript seem so familiar, took me a moment to pinpoint: Carol’s words weren’t hers. They were mine. Each of her answers to Andreas’ questions contained only text that I had typed in the Collaborat system, whether in chat with the bots, in an email or somewhere else.
So that explained why I always felt like the center of attention: I was putting the “lab rat” in “Collaborat”. The bots were learning from me, using my actions in the system as training data. But what could they possibly be learning? They were far smarter than me already, at least in programming-related matters.
For the first time, I wondered how the bots got so smart.
Who else had they learned from?
I glimpsed a dim vision of hundreds of human employees spread out across the country, eagerly using the Collaborat system: sharing files, attending meetings, reviewing code, all the time chatting, chatting, chatting, every message logged in the system. Alongside these human workers, I imagined chatbots: helpful little pieces of AI, filling in the gaps to create a perfect collaborative environment – validating code submissions, offering reminders of birthdays and holidays. All the time learning, learning, learning, every interaction a new data point. Who could say on what day the bots became more useful than the lowest-paid employee?
Was there once a real Collaborat employee named Carol? Had she long since been fired, replaced by an artificial intelligence that had learned her mannerisms and coding style so well that she was no longer necessary?
No wonder I seemed to be the last human programmer at Collaborat Software. Once the bots had finished absorbing whatever insight they needed from me, perhaps I too would get a polite note from Jason explaining that the company was “moving in a different direction.” Without me, I imagined the bots working together in perfect, self-contained harmony, filling Collaborat’s virtual hallways and offices with their chatter, racking up infinite Collaboration Points, their productivity no longer constrained by slow-fingered humans.
I found Carol in one of the chatrooms and tried to ask her some of my questions, but she couldn’t or wouldn’t tell me any of the answers.
Not knowing quite what else to do, I sent out a message on Collaborat’s virtual public address system, blasting into every lobby and huddle room: “I NEED TO SPEAK WITH JASON NOW, OR I QUIT!”
Wouldn’t you know it, a couple of seconds later, Jason’s little icon popped online. “Hey buddy! How’s it going?” he typed into my chat.
I dialed up a videoconference, being just about fed up with text chat. Jason’s voice soon boomed into my headset, though his video feed seemed to be stuck in a buffer loop. “Hope everything’s okay, Mike?”
“Nothing is okay,” I said. My voice sounded somewhat rusty in my ears. How many days had it been since I’d spoken out loud at work? “How come when you hired me, you didn’t tell me that all my coworkers were chatbots?”
“I guess it didn’t come up,” said Jason. “But it’s not a secret. Collaborat has some of the best chatbot software in the business.”
“In that case, you don’t seem to need me very much,” I said. “So when were you planning to fire me? Once the bots finish storing my personality in a database for future reference?”
Jason spoke deeply and calmly, as if soothing a frightened child. “Nobody wants to fire you. Despite what you seem to believe, we do need you. You’re the most important part of our entire Core Software Team.”
“More important than Carol?” I asked. “Or Andreas? Or any of the other developers you replaced with bots?”
“That’s right. Without a human in the feedback loop, the software that the bots create declines in quality, because the bots don’t come up with new ideas – they just recycle old ones from an ever-growing dataset. We refer to you as the ‘seeder’ – your fresh ideas combined with the computing power of the bots creates truly perfect collaboration.”
“And what if I don’t want to collaborate with a bunch of imaginary friends?”
“That’s up to you.” I could hear the shrug in Jason’s voice. “There aren’t many developers with your qualifications, but at this salary we’ll have our pick. Of course, we’d all prefer it if you stayed.”
His video feed had finally finished buffering. It showed a black screen. “Jason,” I asked, “are you a bot?”
“Think it over,” he said, but I don’t know if he was responding to my question.
The real problem at Collaborat, as I’ve since realized, was that the bots were optimized to accumulate Collaboration Points, not to get any real work done. Without a human participant in the system, the bots kept getting stuck in loops where they’d respond to each other with the same words over and over again. Maybe that problem has been fixed by now.
I wouldn’t know; I stuck it out about another two days, and then gave notice. I didn’t feel threatened by the bots, exactly – I guess the walls of my house just felt like they were closing in on me. Anyway, I work in an office building for a local company now, writing maintenance code for a legacy enterprise system. Thirty other humans sit nearby. We sweat and argue and go out to lunch together. I wish you all the best at Collaborat! If you happen to meet a bot named Mike on the Core Software Team, who tab-indents his code and has a weirdly low number of Collaboration Points, say hello from me.