When everyone can build, what does building still prove?
I recently switched antivirus software, from McAfee to Norton. Not because McAfee was bad at being an antivirus. It was fine at that. I left because it kept doing the one thing that reliably triggers my tech aggression, a trick it took straight from the browser wars: hijacking my attention. The carefully crafted "click here" screens that quietly implement something I would never choose myself, like Yahoo with "secure search" suddenly living in my browser. The big flashy red warnings that my subscription is about to end in 100 days. One hundred days. Warn me if I am in the last week, maybe. Not when I am still three months away from that point.
So I switched to Norton, on the strength of reviews. And landed in an even worse hell of nagscreens, flashy notifications, and premium functions I would have to pay extra for. Do not get me wrong: premium functions are fine, as long as I can permanently disable and dismiss the notifications for not choosing to use them. You guessed it: that setting does not exist.
Somewhere along the way, simple functional software that just does what you pay for stopped being a product anyone sells. The bottom line of shareholders matters more than delivering good, functional commercial software. And as a consumer, I used to have exactly two options: accept it, or switch to a competitor doing the same thing in a different color scheme.
For some categories, that is still where it ends. Nobody builds their own fully integrated browser or a proper core antivirus engine at the kitchen table, so when Edge or Norton gets worse, you watch it get worse and you swallow it. But for a fast-growing set of everyday problems, the frustration finally has a pressure valve: you can build the tool yourself.
I know, because I do. I have many pieces of software running on my machine or in the cloud that I should not have been able to build alone. Not in the sense that they were forbidden, but in the sense that five years ago each of them would have required a small team: a developer or two, someone for testing, a few weeks of calendar time and a budget line that needed defending. I built them in evenings. Not by typing every line myself, but by orchestrating: one AI model acts as architect and validator, another generates the code, and a test suite sits between them as the referee. My job shifted from writing software to directing it, deciding what should exist, judging what came back, and refusing to accept anything the tests could not confirm.
That experience is genuinely new. For the first time, the ability to create working software has landed at the user level. You no longer need to be a programmer to produce a real, functioning solution to your own problem. You need to understand the problem, describe it precisely, and verify the result. The gap between "I wish this existed" and "this now exists" has collapsed from months to hours.
And yet the dominant emotion in every professional discussion I have about this is not excitement. It is suspicion.
And it goes deeper than the software itself. Wherever software gets built, and wherever businesses communicate at all, trust is part of the contract. The companies that spent decades earning that trust are squandering it in plain sight. Microsoft turned Windows and Edge into advertising surfaces. McAfee and Norton you have just read about. Adobe made its subscriptions so hard to leave that regulators stepped in. HP ships printers that refuse ink it did not sell. Sonos broke a product its customers loved with an app update they begged to have rolled back. Different companies, same pattern: user focus traded away for commercial squeeze, premium walls around functions that used to simply be the product, trust spent as if it were income instead of capital.
And the thing that could take their place, software built by users themselves, arrives with no trust at all. It deserves a platform. It starts with nothing.
That is the creative paradox: the moment creation became available to everyone, we stopped trusting what gets created. And, more quietly, we started distrusting each other.
Abundance kills the signal
To understand why, it helps to see what creation used to signal.
For most of history, a finished piece of work carried an implicit certificate. A well-built application proved that someone had spent years learning to program. A sharp analysis proved that someone had done the reading. A carefully written proposal proved that someone had sat with your problem long enough to understand it. The artifact and the effort were welded together, so the artifact could stand in for the effort. We never really trusted documents or code. We trusted what they told us about the person behind them.
AI cuts that weld. The polished output no longer proves anything about the effort, the understanding, or even the identity of whoever sent it. A flawless report can conceal an author who never read it. A working prototype can conceal a builder who cannot explain a single design decision in it. The signal that professionals relied on for decades, quality as evidence of competence, has quietly stopped working.
This is not the first time abundance has destroyed a signal. When photography became cheap, a photograph stopped being proof that something mattered. When email became free, a letter stopped being proof that someone cared enough to write one. Every time the cost of producing something drops to near zero, the thing itself stops carrying information about the producer. What is new this time is what the signal was attached to: not attention or sentiment, but professional competence itself.
The distrust does not stay with the artifacts
If this were only about artifacts, it would be manageable. We would build better review processes and move on. But the suspicion does not stay contained. It leaks into relationships.
Consider what has already changed in ordinary professional interactions. You receive a thoughtful reply to a difficult email and catch yourself wondering whether the sender actually engaged with your point or delegated the engagement. A partner delivers an impressively complete architecture document and you find yourself probing in the meeting, not to understand the content, but to establish whether they understand the content. A supplier's proposal is fluent, well-structured, and generic in a way you cannot quite pin down, and you notice that its fluency now counts against it.
You can watch this play out in real time in the AI communities themselves. I spend time in several of them, since my own inference strategy spans multiple providers, and the pattern is remarkably consistent: a new initiative or tool is met not with enthusiasm but with skepticism, almost by default. A great idea is assessed as exactly that, an idea, worth nothing until evidence arrives. Show it running, show the code, show who is behind it and what they have shipped before. The communities that live closest to this technology, full of people who love building with it, have become the hardest audiences to impress, precisely because every member knows firsthand how cheap an impressive-looking announcement has become.
I would like to claim I am above this, but my own first reaction to any new initiative is the same: not curiosity, but an immediate demand for proof. That reflex is the paradox in miniature. The more fluent creation becomes, the further the burden of proof moves upfront, onto trust and evidence that no amount of fluency can substitute for.
Some communities have stopped relying on individual judgment altogether and turned the skepticism into policy. Several 3D printing communities, for example, now simply prohibit sharing software you built yourself, even when it is released as open source. Not because the software is bad. Because there is too much garbage to sort through, and because nobody can tell whether the tool you adopt today will still have a maintained future tomorrow. The features can be excellent and the license can be open; the one thing no reviewer can verify is a future. So the door stays closed for everyone.
Think about what that means. At the exact moment our freedom to build explodes, the dynamics push in only two directions: back toward the big software houses that are busy squandering the trust they built, or toward an endless splintering of self-built solutions, each with a consumer base of one, maybe a handful. The middle ground, where a good tool finds a real audience on its merits, is exactly what the missing trust makes unreachable.
None of this is paranoia. It is a rational response to a real shift: in any interaction, you can no longer tell who, or what, is at the helm. And that matters, because the value of a professional interaction was never in the words exchanged. It was in what the words committed the other person to.
When a colleague said "I reviewed this and it is sound," the value was not the sentence. It was that a specific person had put their judgment, their reputation, and their future accountability behind it. When a partner said "we can deliver this," the value was that an organization was staking its name on the claim. Strip away the certainty about who produced the words and whether anyone stands behind them, and the interaction becomes hollow even when the content is excellent. You are left holding a well-written sentence that commits nobody to anything.
This is the question I keep returning to: if you do not know who is at the helm, how do you establish the actual value of the interaction? Not the value of the deliverable, which you can test, but the value of the exchange itself, the thing partnerships are actually made of.
What my own practice taught me
Building with AI myself has sharpened this rather than softened it.
The setup I described earlier, models generating and validating each other's work with tests as the arbiter, taught me something uncomfortable: inside my own pipeline, I trust nothing by default. Every generated component is treated as a claim to be verified, not a result to be accepted. The code does not get merged because it looks right or because the model sounded confident. It gets merged because an independent check, one I designed and control, says it behaves correctly. Confidence is cheap. Verification is the product.
Now turn that lens outward. If I apply zero-trust discipline to the output of my own tools, tools whose configuration I chose and whose limits I know, what should I apply to a deliverable from a party whose process I cannot see at all? The honest answer is: the same discipline, or more. And most professional relationships are not built for that. They were built on the old signal, where quality implied competence and competence implied accountability. That inference chain is broken, and most contracts, partnerships, and working relationships have not yet noticed.
But my practice also taught me the counterpoint, and it is the way out of the paradox. The value I add to my own projects did not disappear when the typing was delegated. It moved. It moved into the questions I ask, the requirements I refuse to relax, the tests I insist on, and the responsibility I take for what ships. The machine produces; I answer for it. Anyone who has worked this way knows the difference immediately between someone who directs AI and someone who merely forwards its output. The first can defend every decision. The second can only hope nobody asks.
Trust becomes the scarce asset
Here is where the economics of the situation become interesting. Markets price scarcity. For decades, the scarce asset in software and knowledge work was capability: the ability to produce good work at all. Capability is now abundant. Anyone can produce something that looks like good work. Which means capability is no longer what commands a premium.
What remains scarce is everything AI cannot generate on demand: a track record that took years to accumulate, transparency about how you actually work, the willingness to be accountable when something fails, and skin in the game. Trust, in other words. Not trust as a soft value on a company slide, but trust as a hard economic asset: the one input into a professional relationship that cannot be prompted into existence.
This is something that's being picked up on more broadly. With terms like 'HX' (human experience) entering the arena (as posed by Amy Abatangle of Gartner,ref; in the ai era trust-scarcity is rewriting the rules of brand-growth).
Also; i came across this really interesting paper of Robert Dogonowski which is for the most pretty technical, but what caught my eye in particular was it's conclusion; "The analysis locates the failure not in AI capability but in the information structure of oversight". In plain terms; all things being equal; if you can't control what is thrown at you, you can only control what you verify. the limitation becomes exactly that; what you can verify.
You can see the repricing happening already. The question in vendor selection is shifting from "can you build this?" (almost everyone can, or can appear to) toward "will you stand behind it, and how do I know?" References, long relationships, and demonstrated ownership of past failures are worth more than portfolios, because portfolios can now be manufactured. So can reviews, as my antivirus switch reminded me the hard way. The professionals gaining ground are, perhaps surprisingly, the ones who are most open about their AI use: "this analysis was AI-assisted, here is what I verified myself, and here is what I will personally guarantee." That sentence does something no polished deliverable can do anymore. It re-attaches a person to the work.
And this is why the paradox resolves the way it does. AI did not destroy the value of professional relationships. It relocated that value. The deliverable used to carry the trust; now the relationship has to carry it, because the deliverable no longer can. Interactions where a named person takes verifiable responsibility become more valuable precisely because everything around them became cheap.
Working in the paradox
So what does this mean in practice, for anyone building, buying, or partnering right now?
It means saying how you work, before you are asked. Disclosure of AI involvement is rapidly becoming what citing sources became in research: not a confession, but a mark of professionalism, and its absence a quiet warning.
It means judging partners on verification, not fluency. The interesting question is no longer whether the work is polished but how the other party checks their own output, and whether they can walk you through their reasoning without the document in front of them.
It means putting accountability in writing where it used to be implied. The old signal did the work of a warranty; now you need the actual warranty.
And it means accepting that trust has become slow in a world where everything else got fast. You can generate a working prototype in an evening. You cannot generate a reputation in an evening, and for the first time in years, that asymmetry works in favor of the people who have one.
The creative paradox, in the end, is only a paradox if you believe the value of work lives in the artifact. It never did. The artifact was always a proxy for something human standing behind it. AI has merely forced us to stop confusing the two. Everyone can now hold the helm's wheel. The scarce thing is knowing whose hands you would trust it to in a storm, and that question has always had only one kind of answer: earned, slowly, and on the record.
