Blog
/

Are We All Going Bots**t Crazy?

On AI, human agency, and the things worth preserving.

Are We All Going Bots**t Crazy?

Everywhere I turn, someone is talking about AI fatigue.

My teammates, my professional network, my friends; even my kids seem worn down by the deluge of content and context they now have to process every single day. I feel it too. I hate that a quick search for information now feels like a quasi-human conversation full of opinions and sycophantic offers of help, none of which I asked for.

At first, AI fatigue resonated with me. AI is demanding attention in so many aspects of my life that I am struggling to keep up in a meaningful way. I’m leading a team of developers through a period of professional uncertainty, while building software whose bedrock is human connection, and raising kids about to enter a workforce where AI is ubiquitous, but what that means for their careers is still unclear.

These are all very different experiences of AI, yet they seem to be producing similar feelings of excitement tempered by uncertainty, frustration, and exhaustion. 

But I've started to wonder if it's really the AI part that's wearing me down. I suspect the tensions we are all experiencing in similar ways in our different lives actually are human problems, stemming from human decisions. 

AI changes the conditions under which familiar human problems occur, but it certainly didn't invent them. We’re just speeding up the bullshit.

We've made the appearance of human effort cheap

Humans have been producing bullshit for decades, but we used to do so at human speed. The indiscriminate sales pitch, the padded student essay, the colleague bluffing their way through a meeting they didn’t prepare for. Bullshit isn’t new; what has changed is the speed, scale, and credibility with which we can produce it. 

In On Bullshit, Harry Frankfurt defined bullshit as an indifference to the truth; not necessarily lying, but not trying to convey facts. It's far more dangerous as an enemy of the commons than a deliberate lie; at least liars care about the truth. 

AI has made it incredibly cheap to produce convincing evidence of human effort. We are all experiencing it in some form: writing that suggests depth of knowledge, code that suggests understanding of the system, or communication that suggests care and empathy. It manifests as canned perfection that resists meaningful editing or argument because it lacks deliberative reasoning. 

AI bullshit has become so prevalent that it has spawned botshit, another term that may appear to blame the tech but, in fact, points to a problem further upstream. In ChatGPT is Bullshit, Michael Hicks reminds us that LLMs simply aim to replicate human speech or writing, so it seems to me they’re doing exactly what they are designed to do. We are part of the equation. Bullshit in, botshit out.

The ubiquity of AI kicked off a sequence of events. We got shiny new toys that superpowered our brains, but the productivity gains came with a hidden cost. The cognitive burden of constant detection and verification began to wear us down, individually and then collectively. We started questioning our own judgment and calling others to account. We were experiencing new things at a rate that our rational minds couldn’t keep up with. 

It started taking up the energy we need for uniquely human things like original thought, independent judgment, and lived perspective. We began to surrender thinking and, consequently, agency. What followed was the slow erosion of trust across professional and public discourse. 

The erosion of trust is being felt keenly by community builders. At Discourse, our foundational goal was to bring thoughtful conversation back to the internet. That’s tricky when not all participants in the discussion can think. It’s even trickier when they’re so good at acting like they can that we have to waste brainpower on an assessment before we process the information. A confident, well-crafted response goes a long way to building a reputation of expertise, which may or may not be deserved. The system further breaks down when those responses start to overshadow more accurate contributions from less polished writers. None of the old signals work anymore.

But we stand to lose more than trust.

Understanding is not the same as producing an answer

Some activities produce value through the process, not just the result. William Zinsser encapsulated this idea when he famously said that writing is thinking on paper. When we stop choosing our own words and abdicate our thinking to a system, important parts of writing are lost in the loop. When humans write, opportunities for understanding are built into the process. As we gain experience, we begin to build a voice, and paragraphs connect naturally as our thoughts develop. So to produce content faster than we can think has to come at a cost.

I recently had an elbow reconstructed and lost the use of my dominant hand for a couple of weeks. Rather than leaning on technology for support, I learned to write with my left hand. It was far from proficient, but I stuck with it doggedly, even though I couldn't articulate why I felt so strongly about it. Now I realize that writing mattered to me independently of its efficiency. Writing is how I organize and process my thoughts. Voice-to-text demands a fully formed thought before I voice it.

Just as writing helps us discover what we think, programming helps us understand why systems behave as they do. Learning happens through mistakes, revision, and experimentation. If understanding is built through the process of writing the code or building the product, how do we preserve it when the construction process changes?

Research shows that how we use AI to pursue efficiency affects how much we learn and understand. Engineers suffer comprehension debt if they don’t critically evaluate agentic output. Teams accumulate cognitive debt as their shared understanding of the system decreases to the point that they can no longer confidently predict the impact of changes. 

If we think of AI tools as amplifiers of our existing capabilities, we can draw parallels from history. The invention of calculators didn’t remove the need for mathematical understanding; it simply freed people up to tackle more complicated problems. Similarly, the introduction of CAD into architecture, which sped up the rate at which architects could illustrate their ideas but didn’t eliminate the need for a true understanding of structure and design.

Both these technologies changed how people developed and applied expertise, not just how quickly they could do it.

So now the question is not whether AI did the work, but whether the human gained or lost something important in the process.

We should ask why a developer can become more productive while feeling less fulfilled. Why a technically correct reply in a community can make someone feel less connected. Why a carefully worded personal email can feel impersonal.

And why generating more information can leave everyone less informed.

These are the problems we are really trying to solve. 

Human communication is about more than transferring information

Functional communication primarily exists to transmit information, but communication serves other purposes too. In her writing on AI norms and values, Charity Majors puts functional communication at one end of a continuum, with personal communication at the other. She suggests that the value of functional communication comes from the idea or artifact itself, whereas personal communication derives value from a particular person's thoughts, feelings, or the relationship involved. The further along the spectrum towards personal communication you get, the less appropriate AI-generated content becomes.

In personal communication, there is an expectation that it will be between people. When we interact with other humans, we do so because we want to know what they think or feel. We participate in communities to learn from others or to share our own ideas. We read books or articles to access another person’s experience, judgment, or perspective. We write personal letters or emails to share our thoughts and feelings with people we care about, trusting that they'll do the same in return. 

That trust is the basis of an unspoken social contract that exists among humans in any kind of relationship. Trust is fragile and breaks very easily, as my husband discovered when he sent me an AI-generated apology after an argument one evening. While the words themselves were fine, the obsequious tone left me wondering whether he had actually thought about me at all. The issue wasn’t that he used AI; it was that he didn't meet my expectation of genuine consideration.

Communities are a good example of how expectations of AI involvement change along the communication continuum. A member might regularly choose to consume AI-generated technical documentation, but they would be left with a bad taste if someone sent them a generic AI-generated wall of text in response to a personal message. In a community, the consequences of getting it wrong can extend far beyond any individual interaction.

Community builders are grappling with how to balance the value of human interaction with the utility of AI-generated content in their governance frameworks. Some are taking a hard line, arguing that forums should offer human insight rather than replicating what users can get from LLMs. Others recognize that LLMs need accurate information, so they're taking a more moderate approach, using mechanisms like content segregation and labeling requirements. AI wins hands down if you need an answer in seconds, but sometimes you want to hear from someone who has actually dealt with the problem.

My personal take is that AI-assisted communication can be thoughtful and authentic, but only if the human collaborating with it engages with intention and is accountable for the output. In simple terms - respect the social contract. Whatever role AI plays, it shouldn't come at the expense of the human connections that make communities work.

People need to feel competent, useful, and that they belong

And it’s not just communities feeling the pain; AI-related disruption is occurring in our workplaces too. Researcher and psychologist Cat Hicks studied more than three thousand software engineers and discovered that almost half of them are experiencing AI Skill Threat - a term describing the anxiety people feel about how AI-assisted coding will change their professional identities. 

This information is helpful because although the cause of the threat itself is new, the human behaviors it creates are not. The well-studied identity threat theory tells us that our minds fly into a vigilant threat state when something in our environment signals that a group we belong to is being devalued, displaced, or redefined in ways that undermine our sense of who we are.   

If engineers worry about keeping up and remaining professionally valuable, fear being judged for using (or not using) AI, or are nervous about whether their organization has fair expectations around productivity, it’s no surprise they feel anxious. If these concerns aren’t directly addressed, a culture that prioritizes competition over learning can make them worse. Hicks’ research found that software teams with high ‘contest cultures’, where talent must be proven through aggressive competition, show higher levels of AI Skill Threat than teams that value learning, belonging, and shared achievement. She also notes that the risk of developing AI Skill Threat was halved for software engineers who feel a sense of belonging and are encouraged to keep learning. 

Self-Determination Theory helps explain this finding by identifying three core psychological needs that underpin our motivation to learn and grow: autonomy, competence, and relatedness (belonging). And if creative freedom and problem solving are intrinsically motivating activities, how do we motivate people to learn when AI is threatening to take over the parts of their work that they are good at or, more importantly, enjoy? And what does learning even look like now, given the loss of opportunities to learn through problem-solving discussed earlier? 

Here we can draw another parallel to community building. Healthy communities also depend on people feeling that their contributions matter, that they can ask questions and learn safely, and that people recognize and value them. In both environments, replacing participation with the appearance of it risks undermining the value of participation itself. 

AI has sped up the rate at which things are changing all around us, but the environment helps determine how we experience those changes. 

The human problem is cultural, not just individual

We’re each navigating this uncertainty individually, but we’re all experiencing it at the same time. In our homes and workplaces, we’re constantly establishing boundaries, making judgments, and questioning what we can trust.

All that verification comes at a cognitive cost. Add the psychological threat of questioning whether we belong, whether we’re good enough, or whether our skills will remain valuable, and it's no wonder we’re feeling fatigued.

But these pressures don't exist in isolation. The collective expectations and norms within our schools, workplaces, and communities can create or intensify them. This places responsibility on leaders. Organizations should be intentional about the cultures they create and mindful of the messages they may be inadvertently sending through existing norms.

An engineering team that treats every technical discussion as a test and celebrates individual brilliance may be unwittingly nurturing a contest culture. AI adoption isn’t a simple productivity decision if a team rewards speed over understanding. 

But the hidden cost of chasing productivity extends well beyond engineering teams. When we prioritize speed and output over quality and understanding, we risk simply shifting the burden of work onto someone else. Glean’s 2026 Work AI Index found that the time AI saves us is absorbed by the work of making it usable. Not all of that work is wasted labor—much of it involves learning, iterating, and verifying output, all of which is valuable—but it gets costly when expectations aren’t clear about what quality looks like and who is accountable for the finished artifact. Suddenly, the time we save ourselves is being charged to other people. 

At Discourse, we are a fully remote, async-first org, and genAI use is challenging one of our core values. We believe in the Power of the Paragraph – the value of writing things down and thoughtful storytelling. We read and write with mutual respect for each other’s time and attention, and we never had a reason to question each other until botshit entered the equation. It’s a challenging culture shift, and we still have work to do, but this sentiment from a wise teammate sits well with me:

If a post is a waste of time because it’s too wordy, inaccurate, or hard to parse, that’s reasonable feedback anyone could use. Shifting it to AI/not-AI reframes the discussion away from quality and towards the accusatory “was this you?”

In communities—environments which often encourage anonymity by design—trust has replaced attention as currency. When established patterns of engagement break down, and people don’t know who or what to trust, they stop participating. Regardless of the environment or the cause, when people stop explaining and documenting their thinking, shared knowledge deteriorates. Software teams are shipping new features and accumulating code faster than ever, but are they accumulating understanding at the same rate? Communities are seeing record levels of bot traffic that consume resources without contributing to the discussions that make them valuable, while AI-generated content risks driving away the humans who used to engage with and challenge each other. Public writing is proliferating, but how much of it contributes original thought? 

Are we really willing to trade individual productivity gains for weakened collective intelligence? If not, we need to start making smarter choices about how we use these tools.

What should we preserve, and what should we change?

Right now it feels like the balance is off. To course-correct, we need to identify the parts of work and life that derive their value from human participation and find ways AI can support, rather than displace, that. AI can give us input, but the value is in what we do differently because of it.

I’ve experienced the tension in my own work. There’s a commonly held belief that adopting AI automation in community management can make things feel transactional and robotic. That can certainly be the case when it is used indiscriminately, but in my experience it’s done the exact opposite – automating tasks like compliance reporting and content tagging frees up the bandwidth to be more human. When you use a tool to efficiently handle functional or operational tasks that take a lot of time—like using an accounting tool to calculate your tax return—you reclaim your most valuable asset: presence. 

The point isn’t to retain friction for its own sake; it’s to distinguish unnecessary work from the activities that add value and only humans can provide.

Engineering teams are facing similar challenges as they figure out where AI fits into software development workflows. Reaching agreement on which aspects of the work are appropriate to delegate and where the lines of responsibility lie can be challenging under any circumstances, let alone in a highly charged environment full of anxious, stressed people.

I find it helpful to approach these decisions with five principles I think are worth preserving.

Ownership

Everyone needs to have skin in the game. We should take an adversarial stance when collaborating with AI, challenging it by insisting on sources and asking for counterarguments. We should stand accountable for everything we share. We should be able to explain and defend it, regardless of how we generated it. As Charity Majors puts it, we’re not just the “human in the loop”; we’re the owner of the loop.  

Understanding

AI should support our learning rather than bypass it. We should apply our own judgment to critical decisions and recognize which parts of an activity offer opportunities to develop our understanding. We should create conditions that encourage this work and deliberately build those opportunities into our AI-assisted workflows.

Care

We should always consider the effect of our work on those downstream. What are they expecting from us, and what are we asking in return? Are we saving time by wasting someone else’s? 

Belonging

We need to be intentional about the culture we create and what messages we send through our decisions and actions. We should create environments where people feel safe saying “I don’t know” and where learning, uncertainty, contribution, and collaboration are valued. 

Trust

We should preserve provenance, original perspective, shared reasoning, and meaningful human engagement. We shouldn’t be afraid to use human language that carries risk, uncertainty, and accountability. We should challenge each other with kindness and respect, and accept challenges from others with curiosity and good intent.

So is it AI that is wearing me down?

The answer is more complicated than that.

Sure, the additional cognitive work is a factor, but the surrounding uncertainty about what to trust, the changing expectations of what competence looks like, and the loss of meaningful human participation are much more taxing.

They are fundamentally human problems, shaped by the decisions we make.

AI isn’t taking away everything that makes us human; it’s simply presenting us with choices. We all have a say in what gets lost, what gets preserved, and what becomes possible.

Comments