The AI Pragmatist's Survival Guide

Something interesting keeps happening to me when I deliver AI literacy training. During the training, people follow along, ask questions, and genuinely try to understand where, how, and whether they should use LLMs in their daily workflow.

But I get the real dirt outside the formal training. A subset of learners always come to me with a similar complaint. It goes something like this:

"It's just not that useful in my workflow, but everyone from the top down insists we use it as much as possible, so I end up sending it off on research tasks that I don't really care about or just avoid using it at all."

I get this comment literally every time I run a training program. And oftentimes these people aren't wrong. When we dig into their workflows and responsibilities, it doesn't make sense to use an LLM.

This post isn't about whether or not you should use LLMs; it's about how you survive in an organization that has drunk too deeply of the Kool-Aid.

The Power of Peer Pressure

"If all your friends jumped off a bridge, would you too?"

That's what a lot of mothers have asked their kids at some point as they were growing up. Humans are social creatures, and most of us want to belong to a group.

In a business environment, peer pressure can be downright destructive. Have you ever heard a leader critique a worker as "not being a team player"? Unfortunately, in some companies today, that means "accepts any wild claims about AI that the organization makes."

And this is one of the main issues with surveys and self-reported data. It's also one reason why we aren't seeing any impact of AI on GDP and overall productivity metrics. There is an honesty problem. And for me and my business, if an organization has an honesty problem, I generally avoid working with them. Here are some signs:

Faith Tests

This is the "team player" test. In many organizations, advancement and continued employment require a belief in the culture and its initiatives. This means if executives in the organization start promoting the "transformative power of AI" or coworkers with good rapport with leadership start talking about how they're "10x'ing their productivity", you realistically have two options:

1. Nod along and say nothing.
2. Speak truth to power.

Given these two options, most people will take option 1 because it's safe. And can you really blame them? Last time I checked, it still takes money to live, and that requires an income, and most people get their income from their job. While it can feel good to speak truth to power (or fantasize about telling people exactly how you feel), that's often not the... expedient path.

I see this in my advisory services, especially during discovery. Someone will say the company is now "AI first" but can't tell me exactly what that means. Or they'll talk about the vague "transformational power of AI", but all they've done is bought some Copilot subscriptions that people are only using performatively.

Token Leaderboards

Thankfully, with the move to metered billing, this particular idiocy is starting to die down. Companies have been measuring employees on AI spend, where more is better.

If you've read my post about workslop, you already know how this story ends: whatever you measure, you'll get more of, and usage is not value.

Edtech is another place where something similar happens. L&D professionals will tout "course completion" and "NPS (Net Promoter Score)" without any other outcomes. Those stats in a vacuum mean nothing. Ask anyone who's been forced to sit through HR training. 100% completion, absolutely miserable experience.

AI Washing

This one is happening a lot. Leadership doesn't know, because it's tied to the faith tests. Every week, some engineer reaches out to me on LinkedIn and tells me about their AI reality:

"It's useful in some areas, but it can't actually complete all of my tasks."

But that's not what they say in the standup meetings. In public, they say that they've been using AI heavily and it's great. In their real work, they use AI sparingly for what makes sense, then send it off for performative tasks while they finish the job themselves because they care about quality.

But that's not what a faith-test organization wants to hear. Management is unhappy when AI isn't in the story. This has spread through some organizations all the way to the board level, and it's not how engineering works. In engineering, you bring problems, and the engineering team figures out solutions. In the worst organizations, we see the equivalent of "here is a hammer, use it for EVERYTHING." And that right there is one of the reasons we aren't seeing much ROI in AI tool adoption.

There's a reason the hype persists even where the results don't. I said this on social media a while back and I'll say it again here: a lot of incompetent people are VERY excited about the prospect of knowledge and skills not mattering.

Now, Let Me Ease Up a Bit

I'm an engineer by training and a skeptic by nature. You can convince me of anything if you bring the data. And if you follow me on social media, you'll note that I'm often critical of AI initiatives and companies. And yet, I deliver AI literacy training.

Well, that's because I'm not a zealot. I'm a pragmatist. So what do I really think?

• AI has significant legal and social challenges: Massive copyright theft, environmental concerns, the wave of slop covering the internet. These are legitimate issues that need addressing. But regulators and courts move slower than tech startups. I try to steer my clients in directions I believe are legally and ethically sound.
• Is AI useful?: Absolutely. I would rather ask an LLM for syntax I don't use every day than comb through the cesspool that is Google Search or Stack Overflow. I'm also technically competent and know when I need to fact-check it.
• Is AI a god-machine that renders humans obsolete?: No. I don't believe LLMs can achieve true AGI, and I've spent a lot of time with talented people across fields who use it; zero percent of them claim it can do their job without intervention.
  
  However, some types of workers will certainly be replaced. But with technology, that isn't unique to LLMs; we've been automating and replacing jobs throughout human history.
• Does AI increase productivity?: Sometimes yes. But it's not simple, because many output gains depend on the human operator's skill, the desired quality level, and the type and complexity of the task.
  
  Even with my technical expertise, in my AI experiments it rarely "one-shots" anything. Over time and with refinement, it can get into the ballpark on quite a few mundane tasks, but I often point out that today AI tools are a multiplier on expertise, and multiplying by zero is still zero.

And that brings me to the reason for this post.

The Nuance

If you take one thing away from this post, take this, because both of these statements are true, backed by my experience with dozens of organizations delivering AI advisory and training:

If you're not getting miraculous value from AI, you're not crazy. The people claiming 10-100x productivity transformations are misleading you. Sometimes deliberately, because they have a profit motive, and sometimes because they can't tell the difference between "good" and "looks good".

You see this in the recent outcry about Anthropic watermarking AI output. Suddenly a bunch of people who were evangelizing AI are embarrassed to admit how much they were using it? I'm sorry, but if the output was both high quality and efficient, they wouldn't care about being "exposed".

If you're getting zero value out of AI, that's likely a training issue. Not always, because these are probability engines, so if you work in something niche, then yes, LLMs probably won't be super useful. There's also an expertise gap. The more skilled you are and the better you can communicate and structure the work, the better outputs you are going to get.

Nuance dies on social media. The boosters want to say that everyone should be getting 100x productivity because they have something to sell, so if you're not, you're the problem. The skeptics say that nobody gets any value at all, so don't bother. They're both wrong. But the only way for you to learn whether it provides value is to run experiments and collect data.

Now, let's talk about survival.

You Can't Fix the Culture, Stop Trying

If you're an individual contributor, listen up. You cannot fix this. You cannot reason people out of beliefs that they didn't reason their way into. We have a serious herd mentality in business right now. Yes, it will feel good in the short term to fire off that article, be snarky on Slack, or whatever power fantasy you're having right now (I have them too).

If your company is in the grip of AI mania, they are not going to look at your dissent as information; they're going to look at it as a defection. In public meetings, if a coworker says something like "AI is making me 10x more productive", they're likely just performing because they think that's what management wants to hear.

It's not your job to talk them off the ledge. It's a trap, and it will backfire later, because when the numbers finally surface, bad leaders will blame the people who "just didn't buy in." It'll work for a time, but eventually investors will demand real returns, and you'll already be laid off by then.

Instead, focus on the things you can control.

Guard the Quality of Your Output

Your work is your reputation, and reputations last a long time. The mania is loud now, but it is not permanent, and when the pendulum swings back, someone is going to have to clean up the workslop. The people who kept their standards will be the ones trusted to do it. Keep your head down and do your job well. If management wants you to use AI, then use it; they're the boss, but wherever possible, try to maintain a quality bar with practical usage.

Learn the Tools

At this point, I don't believe LLMs are going away. Anthropic and OpenAI may not survive, but I do think that local models are here to stay. You do need to figure out where the lines are so that when you do walk into an interview and you're asked about AI, you can speak from experience.

Be Tactful

Yeah, I came out of IT. Tact is not something engineers are known for. So here are some rules you can follow:

1. Don't debate the religion: When people make vague but supportive noises, let it pass. Again, you're not going to reason anyone out of a belief they didn't reason their way into. Plus, they may just be toeing the corporate line to protect their job. It happens.
2. One-on-ones, not groups: If you have concerns, raise them privately, through proper channels. Group settings plus peer pressure can force people to take positions they might privately doubt.
3. Ask questions instead of making statements: This is the Jedi mind trick to use on boosters. I know you want to argue; you want to tell them they're full of it, but that makes you look like a defector. Instead, ask questions. "Wow, that sounds really great. What metrics should we collect to prove this works? We'll want this well documented so we can celebrate the win." Data is like kryptonite to boosters.

If the boss demands X, do X

Yes, malicious compliance is a thing.

If your company has genuinely gone psychotic with token quotas, AI-washing as policy, and slop worship, then you only have two options: malicious compliance or leaving.

In a tough job market, most people are choosing malicious compliance. And if you care deeply about quality, like me, I know it hurts your soul to not produce the best quality you're capable of. But you need to let it go.

If your boss demands you do X, then you do X. You do it exactly as specified, documented, visible, without sabotage. If the demand is theirs, the outcome is also theirs. And when the predictable result occurs, the paper trail tells the story.

Martyrdom pays nothing and teaches the organization nothing. If your boss demands 10x output and outsourcing your whole brain to AI, then polish your resume and let the slop flow.

You can't imagine how much it hurts me to say that. But some companies have gone absolutely off reality's rails.

Just be aware that the hype cycle is still in full effect, so the next job might be more of the same under a different logo. But the best time to find a new job is when you already have one, so take the time to be picky before you jump ship. During the interview, if they ask about AI, ask how they measure AI competency. If the answer is adoption, metrics, and token spend, you have your answer: run.

Stop Doomscrolling

Consume exactly as much AI news as you need to stay informed, and nothing more. This is one area where I find a lot of value in AI. I set up a little cron job with a cheap model to grab news feeds, filter down to what I'm interested in, and summarize them with original links.

The firehose of outrageous claims and slop is designed for engagement, and rage-baiting works. Stop scrolling. Seriously.

Instead, find like-minded peers, get together once in a while, have a drink, and complain. We laugh, we cry, we share trauma. Then, go find something productive to do.

For the Leaders

I mentor leaders too. But this post is getting long, so I'll keep this short. Culture is what you tolerate.

• Measure outcomes, not adoption. Ditch the vanity metrics.
• One of a leader's responsibilities is to be a shield, so your best people can be productive. That means not telling them how to do their job. You set goals; they figure out how to deliver.
• Train your people properly. "Hey everyone, you now have an AI subscription, figure this shit out." is not leadership. I'm genuinely shocked by how often I see companies do this.
• Keep it real. Celebrate the wins, illuminate the challenges. It's much easier to defend "we're seeing some gains, but not what we hoped" versus whatever exaggerated claims you can twist the data into. Eventually, reality will pay everyone a visit.

The Bubble Pop

The market can stay irrational longer than you can stay solvent. While there may be a bang, I'm predicting more of a slow, embarrassing deflation outside of a few companies (looking at you, Oracle, OpenAI, Anthropic, and Coreweave). The blockchain and crypto people are still around; they're just a bit quieter now.

When it happens, organizations will need people who can legitimately judge where AI is useful and where it isn't, and whether the cost is worth the ROI. That means people who held their quality standards, learned the tooling, and remained sane.

That's the pragmatic long game. You don't need to win every argument or try to die on every hill. You just need to still be standing as competent, credible, and non-psychotic when the boosters finally have to admit reality.

You're not crazy. Do good work. We'll get through this.