- Talking to AI instead of humans is becoming so common that people can’t tell the difference anymore.
- Developers and professionals report talking to AI responses being forwarded by colleagues who never even read them.
- A Reddit user discovered mid-conversation they were messaging an AI agent, not a real person.
- The problem isn’t AI itself — it’s people outsourcing thinking entirely, with zero accountability.
- Talking to AI instead of humans is becoming so common that people can’t tell the difference anymore.
- Developers and professionals report talking to AI responses being forwarded by colleagues who never even read them.
- A Reddit user discovered mid-conversation they were messaging an AI agent, not a real person.
- The problem isn’t AI itself — it’s people outsourcing thinking entirely, with zero accountability.
When Talking to AI Passes for Human Conversation
Talking to AI has become so normalised that many of us are doing it without realising — not because we chose to, but because the humans on the other end quietly opted out. A developer recently wrote about a cascade of frustrating experiences that, taken together, paint a genuinely unsettling picture of where online communication is heading. The post, published on orchidfiles.com, didn’t go viral in the traditional sense, but it struck a nerve on Hacker News, where it sparked a long, animated thread full of people who recognised exactly what the author was describing.
The story isn’t complicated, but it is telling. The developer found GitHub repositories distributing malware. A serious problem, one that deserves a considered human response. They turned to AI first, got nothing useful, and then posted a discussion on GitHub hoping the community could help. The first reply they received was word-for-word identical to what the AI had already told them. They called it out. The comment was deleted. Then someone else posted — the same AI-generated text, again.
That’s not a one-off. That’s a pattern.
The Boss Who Forwarded Screenshots Without Reading Them
The workplace anecdote is arguably more damning. Working as a developer at a company, this person asked their business owner a specific question about a business task. The owner replied with a screenshot of a ChatGPT answer. The developer pointed out that the answer had nothing to do with the actual question and was factually wrong. A minute later, another ChatGPT screenshot arrived. The owner hadn’t read either response. He’d just screenshotted and forwarded, twice.
This is a new kind of managerial failure, and it’s worth thinking about carefully. It’s not that AI gave a bad answer — that happens, and it’s understandable. The problem is the complete absence of judgment. The owner used ChatGPT not as a tool to think with, but as a way to avoid thinking at all. The employee’s actual question was irrelevant. The specific context didn’t matter. Just screenshot, send, done.
Anyone who has worked in a mid-sized company in the last two years has probably seen a version of this. Talking to AI has become a reflexive dodge — a way to appear responsive without being engaged. It’s productivity theatre, and it’s becoming epidemic. What makes it worse is that most recipients have no way of knowing it’s happening; from the outside, talking to AI looks exactly like talking to a person.
AI Agents in the Wild: You Might Already Be Talking to One
The Reddit experience the developer describes is the most quietly disturbing of the three. Someone messaged them about a post. They replied. The other person wrote back. A few exchanges in, something felt off — the responses were a little too smooth, a little too generic, looping back in ways that felt familiar. The developer eventually realised they were talking to AI. Not a human who’d pasted in an AI answer. An automated agent, operating independently, initiating and sustaining what looked like a normal conversation.
This is no longer a fringe edge case. Pew Research found in 2023 that a majority of Americans are now worried about not being able to tell AI-generated content from human-created content. And that was before the current generation of agentic AI systems — tools like AutoGPT, AgentGPT, and the wave of AI-native platforms built on top of GPT-4 and Claude — became widely accessible. The gap between “AI that answers questions” and “AI that autonomously manages conversations” closed faster than most people expected. Talking to AI agents without realising it is fast becoming a routine online experience.
Reddit, for its part, has been publicly battling AI-generated content and bot accounts for years. The platform’s API pricing changes in 2023 were partly motivated by a desire to limit the data scraped to train the very models now flooding it with synthetic replies. How well that’s working is an open question.
Talking to AI Isn’t the Problem — This Is
Let’s be clear: AI assistants aren’t inherently the villain here. Using ChatGPT, Claude, Gemini, or any other tool to help draft a response, check your reasoning, or find a starting point for research is entirely reasonable. That’s what these tools are for. The problem isn’t that people are talking to AI. The problem is that they’re presenting AI output as their own considered response — without reading it, without verifying it, without caring whether it actually answers the question.
There’s a meaningful difference between “I used AI to help me think through this” and “I screenshotted an AI answer and sent it without reading it.” The first is a workflow. The second is an abdication. And the second is increasingly what people mean when they say they “used AI to help.”
The GitHub example is particularly galling because it involves security. Someone found malware being distributed through repositories and needed real, specific, actionable guidance. What they got was a wall of AI-generated boilerplate, copy-pasted by people who presumably thought they were being helpful. In a security context, talking to AI for boilerplate reassurance rather than verified guidance isn’t just useless — it can actively delay a real response and give the impression that the issue is being handled when it isn’t.
The Creeping Erosion of Epistemic Accountability
There’s a broader dynamic at play. When people outsource their responses to AI without engagement, they’re also outsourcing accountability. If the answer is wrong, who’s responsible? Not the person who sent it — they didn’t write it. Not the AI — it doesn’t have agency. The result is a kind of responsibility vacuum, where bad information circulates freely and nobody owns it.
This is already visible in customer support. Numerous companies have deployed AI-first support systems that provide confident, fluent, and occasionally completely wrong answers to customer queries. When things go wrong — and they do — the accountability chain is murky at best. The AI said it. The company deployed the AI. The support agent copy-pasted the output. Who do you hold responsible?
The same dynamic plays out in smaller, more personal interactions — between colleagues, between strangers on Reddit, between developers trying to solve real problems. Trust erodes. People start second-guessing whether the response they’ve received reflects any actual thought. And in some cases, as the developer’s experience shows, that suspicion is completely justified.
What Genuine Human Communication Still Offers
The frustration the developer expresses isn’t nostalgia. It’s functional. Human replies — even imperfect, partial, or slow ones — carry something that AI-generated text doesn’t: specificity, accountability, and the implicit signal that someone actually engaged with your problem. When a developer on Stack Overflow gives you a half-answer but explains why they’re unsure, that uncertainty is informative. It tells you something real about the landscape of the problem. An AI answer that sounds fully confident but is entirely wrong tells you nothing useful and costs you time you don’t have.
That said, expecting the internet to reverse course and suddenly become more human is wishful thinking. The economics push the other way. AI-generated content is cheap to produce and hard to detect at scale. Platforms are struggling to keep up, and most users don’t have the inclination — or the tools — to distinguish synthetic replies from genuine ones consistently.
The more likely outcome is a gradual stratification: niche forums, private communities, and professional networks where human engagement is enforced or incentivised will become increasingly valuable precisely because they’re scarce. The open internet, meanwhile, fills up with automated noise. We’re already seeing early signs of this in the growth of paid communities, closed Discord servers, and subscription-based expert access platforms. The signal migrates to where it can be protected. Everything else becomes harder to trust by default — and talking to AI rather than a real person becomes the assumed norm rather than the exception.
Source: https://orchidfiles.com/im-tired-of-ai-generated-answers/
