AI Has Broken the Implicit Proof of Work Model

It’s 2019 you’ve read an article. It’s crap. Regardless of how crap it was, save for the worst of the worst SEO spam (and even then, some underpaid person probably had to write that spam), someone sat down and took time out of their life to write it.

Proof of work is a computing term with real world analogies everywhere. In its most basic form, it’s a type of system where producing something is expensive and verifying is cheap.

Most humans place some value on their time. In the United States, we’ve even valued the absolute minimum you can be paid for that labor at around $7.25 an hour (as of 2026).

AI has very quickly and very thoroughly driven the cost of plausible looking writing in most forms (including code) to 0.

You can see people grasping for some value signal when they read articles now. If you have an accidental em dash, or some hint of Claude or ChatGPT speak, most people will dismiss your article without a further thought. This isn’t because all articles that are written either fully or partially by AI have no value, but because without someone on the hook, the reader is now burdened with the full cost of a bad piece of writing.

This goes doubly for code, in which reading has always been much harder than writing. Your 100 line PR had to be written and understood by a reviewer, but he could count on you having taken at least a few hours to write that code.

It’s the same asymmetry that allows DDoS attacks to work: the burden of proving something is useless is significantly higher than the burden of sending useless information.

It’s not just writing though. As AI goes “agentic” (whatever that means), it is moving beyond just text. Institutions have relied on complex forms and bureaucracy to stop everyone but the most persistent people from breaking through. Got a parking ticket? Well you probably wouldn’t hire a $300 an hour lawyer, but you might use ChatGPT to waste everyone’s time. Immigration forms, insurance claims, school disputes, customer service complaints, housing cases: anywhere dogged persistence was once expensive, AI makes it trivial.

What does the future look like? Likely bleak. Digitally, we can likely look at email and social networks. Both deal with similar problems of infinite sending and limited reading. Both deal with it similarly. Black box trust systems, shadow banning (showing your stuff to nobody and not telling you) and rate limiting.

Even then, it doesn’t work well and only applies to platforms that have a system in place for this. Your mom is sending you AI messages she hasn’t read, your VP is sending slack messages he never wrote. In the past, you could trust your mom to not send slop, and your VP implicitly had trust. Not anymore.

In the real world, the outcomes will be worse. The likeliest case is an arms race. Your resume? It will be filtered by opaque systems and rejected. The institutions? They will be forced to add more and more hoops. Customer service? Well, good luck.

Whats the solution? I don’t have answers. Open source projects are quickly switching to proof of trust (can others vouch for you). But it won’t be enough and it disenfranchises newcomers or those without connections. There is a world where pay to play might work. Pay some money, and regardless of if it’s written with AI if you think it has value you can pay some nominal fee to have it “posted” or “sent”. For obvious reasons this isn’t perfect, but it might be better than the black box systems. I think people and platforms will converge on a combination of solutions (most somewhat horrible), but until they do, this problem will only grow worse.