The Builder’s Mirage

Why the AI chatterati's biggest delusion isn’t about technology, it’s about people

Overwhelmed and feeling left behind by the AI tools hype machine?

It’s not you, it’s them…or maybe, just maybe, you are them!

MCP connectors, a suite of 30 tools and apps, agentic frameworks, ‘learn to vibe code in a weekend’…it‘s all so overwhelming. But let’s take a breath for a moment.

First, a personal story…

---

For the past six years, I’ve been serious about woodworking. Not dabbling — properly serious.

I’ve invested significant time understanding different types of work, joinery methods, wood behaviour, and tool selection. I’ve bought more equipment than my workshop (and skills) probably justifies. I can hold a genuinely informed conversation about the merits of hand tools versus machines, about grain direction, about which species suits which purpose.

Yes, I made this for my daughter

But last month I needed a bookshelf.

I went to IKEA.

Not because I couldn’t build one. But because I wanted a bookshelf on Saturday, not a project. The knowledge was there. The capability was there. But my need to apply them wasn’t. I just needed to ‘Get sh*t done’.

This is the insight the AI chattering classes can’t seem to absorb. The assumption baked into the constant, breathless energy of AI forums, LinkedIn feeds, and Substack comment sections is that, because building AI workflows is possible (and some people find it fascinating), EVERYBODY should want to do it and must, or they risk becoming irrelevant and replaced.

That assumption is wrong. It’s ignoring decades of research on how humans adopt technology.

The Builder’s Mirage

Spend time in any AI-focused community, and you’ll leave overwhelmed: everyone is building agents, orchestrating multi-tool workflows, automating their professional lives end to end. The energy is real. The implicit message is relentless: if you aren’t building, you’re obsolete.

You see, the mirage is structural, not accidental. Online communities follow the 90-9-1 rule: roughly

• 90% of users lurk,

• 9% contribute occasionally, and

• 1% generate the vast majority of content.

In AI spaces, that vocal 1% tends to be people who genuinely enjoy tinkering — technically skilled, intrinsically motivated to experiment, finding the complexity itself rewarding. There is absolutely nothing wrong with that. But the 1% are not representative of the median professional’s motivations, time budget, risk appetite, or threshold for investing in new skills and climbing a learning curve for each tool.

Two well-studied effects amplify the distortion.

The false consensus effect describes our tendency to assume our preferences are more widely shared than they are, especially when our reference group confirms them. In an enthusiast cluster, “everyone I know is building agents” becomes “everyone is building agents.”

Then group polarisation takes over: like-minded communities drift toward more confident positions, so “agents are promising” becomes “agents are inevitable,” and “it’s worth exploring” becomes “you’re behind if you haven’t started.”

The visitor (the 99%) feels urgent pressure from the visible minority, then hits a wall of friction when they try to replicate what that minority is doing. Exactly because the minority is not representative of what normal professional life looks like.

The Tinkerer’s Curse

There’s a companion phenomenon to the Builder’s Mirage: the Tinkerer’s Curse. This is the affliction that strikes the enthusiast who has spent long enough inside the echo chamber that they genuinely cannot understand why anyone would choose not to build.

The Tinkerer isn’t malicious. They’re often generous and evangelical, wanting to share something they find genuinely exciting. But they’ve lost access to a crucial insight: their love of the process is their own particular disposition, not a universal human characteristic.

They’ve confused their intrinsic motivation (the pleasure of figuring things out, of assembling something that works, of being technically deep) with a motivation that everyone should share.

In the woodworking world, I recognise this immediately: the craftsman who genuinely cannot understand why on earth you would buy cheap Swedish-marketed/Chinese-made cra*p…(Sorry, I mean furniture) instead of making it.

They treat a trip to IKEA as a significant moral failure. They assume that if you only knew what they knew, you’d make the same choices. That you are, in fact, more to be pitied than scorned!

The Tinkerer’s Curse is the inability to see that knowledge doesn’t automatically generate motivation, and that capability doesn’t translate into desire.

In AI circles, the Tinkerer produces a very specific kind of advice: learn these five frameworks, build this agent stack, master this toolchain. All technically sound.

But…none of it is grounded in a realistic understanding of what the average professional wants to spend their Wednesday afternoon doing.

Are you a Tinkerer or one of the cursed?
Tell us about it?

Necessity: the only reliable motivation to build

They say, “Necessity is the mother of invention.”

But what gets said less is that, without sufficient necessity, inertia wins.

EVERY time.

Most professionals aren’t running away from AI. They’re running a constant, largely unconscious cost-benefit calculation: is this problem painful enough to justify the investment of solving it?

Learning time, workflow disruption, cognitive overhead, the social risk of getting it wrong, the governance risk of getting it very wrong — all of that lands on one side of the scale. The current pain of the unsolved problem falls on the other.

Only when the pain rises sufficiently above the friction cost does action become rational.

The Tinkerer’s threshold sits near the floor. They’ll build with low necessity because experimenting, learning, and building are the reward.

Your average professional’s threshold, however, is considerably higher because, for them, that discovery process is purely a cost rather than a pleasure. When the enthusiast community insists that everyone should be building, what they’re actually saying, without realizing it, is that everyone should be like them, sharing their excitement and passion.

Another factor is tool volatility. When tooling frameworks evolve weekly, features are retired quarterly, and the “best practice” you learned three months ago is already contested and evolved, the friction cost of building rises on a constant upward curve.

Meanwhile, the problem being solved hasn’t changed. So the necessity calculation keeps coming back the same way.

Not yet. Maybe later. For now, tolerate the problem.

Most professionals aren’t waiting to be convinced that AI is useful. They’re waiting for a problem urgent enough to justify the investment of tools, time, and the competition for their energy and brainpower.

What the numbers actually say

Time out.

It probably helps to separate three very different behaviours that AI constantly collapses into one:

1. using AI (chatbots, copilots, embedded features);

2. configuring AI (templates, prompt libraries, lightweight automations, minimal tooling);

3. and engineering AI workflow systems (multi-agent frameworks, tool orchestration, integration pipelines, monitoring, security, the 40+ tool stack).

Most of the “everyone needs to learn this” energy is pitched at category three. But so far, the usage data lives in categories one and two.

• Pew Research Center found in September 2025 that 21% of U.S. workers said at least some of their work involves AI, and 65% said they don’t use it much or at all.

• In the EU, Eurostat’s 2025 data shows 32.7% of people aged 16–74 have used generative AI tools, with only 15.1% using them for work.

• On the agentic AI front: McKinsey’s 2025 global survey found that 62% of organisations are experimenting with AI agents, but no more than 10% are scaling them in any single business function.

• Gartner predicted in June 2025 that over 40% of agentic AI projects will be cancelled by the end of 2027. Their word for the current state was pointed: hype-driven experiments. They also flagged “agent washing” — vendors rebranding existing capabilities as agentic to ride the wave — as a significant distortion of the real picture.

What this is saying, despite the frothy chatter, is that for the majority these are not the signs of a technology that has crossed the necessity threshold for most of the working population.

We’ve seen this movie before

The technology adoption pattern is old and reliable. The early phases of transformative technologies are dominated by enthusiasts who enjoy assembly, configuration, and troubleshooting. Mass adoption arrives later, only after someone packages the capability so that the majority can benefit without having to rebuild their lives around it.

Radio in the 1920s required you to build your own receiver. The U.S. Bureau of Standards published guides titled “Construction and Operation of a Simple Homemade Radio Receiving Outfit.”

The enthusiasts built. Everyone else waited. What unlocked mass adoption wasn’t better building guides; it was the transistor radio. Portable, affordable, no assembly required. The content boom followed the productisation.

Home computing followed a similar arc. I bought a Sinclair ZX80 computer kit in the early 1980s, got properly frustrated while assembling it, and abandoned the project (a waste of 80 pounds!). In the end, I waited 3 years until I could afford to save up the money for an IBM XT (with an amber monitor, no less!) to run proper business software (SmartOffice). This was a machine that came configured and ready to do actual work.

For most people outside of software development, Agentic AI is in its crystal radio and Sinclair ZX80 phase. The AI chatterati communities buzzing with agent-building energy are doing genuinely important work — mapping the territory, discovering affordances, stress-testing the tools— but their experience is not necessarily a productivity template for the other 90%.

The right question: Person, Pain, Promise

There is a product management framework that cuts straight through the noise, and it applies here with uncomfortable precision. Good product thinking starts with three things: the Person you’re building for, the Pain they actually have, and the Promise you can credibly make about how it gets better. Miss any one of the three, and you’ve built something for yourself, not for them.

• The AI chattering class has been getting all three wrong. It has misidentified the Person, pitching to the average professional as though they have the motivations of a Tinkerer.

• It has misread the Pain, assuming the problem is building faster, when the actual pain is a specific, pressing task that is not being done well enough.

• And it has made the wrong Promise, “you can build anything you need”, when most people are screaming, “I just need to get sh*t done, HELP!”

A plausible “transistor radio” phase for AI looks exactly like this.

Narrow, high-reliability applications embedded in tools people already use OR that have simple adoption paths; solving problems specific enough to be genuinely painful; with promises simple enough to be believable.

• Draft this email.

• Summarise this meeting.

• Extract these figures from this document.

• Find me this document.

• Help me research and structure this argument/report

Not “build an autonomous agent that manages my entire workflow.”

The boring future

Here’s the thesis in full: the biggest delusion among the AI chattering class isn’t that AI is powerful; it absolutely is.

It’s that the average professional wants to become an AI workflow engineer.

They don’t. Most of the time, they want the Ikea bookshelf.

And they will reach for it by the most frictionless path available.

The Builder’s Mirage makes this hard to see from inside the echo chamber. The Tinkerer’s Curse makes it hard to understand the majority’s reluctance.

And the failure to ask Person, Pain, and Promise in the right order will continue to keep the conversation pointed at the wrong people with the wrong message about the wrong solution.

— • —

Stuart Miller is Managing Director of Haverin Consulting, a healthcare IT strategy consultancy. He has spent 20+ years watching organisations confuse production with progress, and he remains stubbornly optimistic that product discipline will eventually be fashionable again. #IamPragmatic