Recently, I came across a very amusing discussion on social media. Some netizens complained that LLM hallucinations are decreasing, but their bosses' hallucinations are growing. The comments section was buzzing, with many people reporting that since the big craze for crayfish at the beginning of this year, their bosses' imagination about AI has become increasingly wild, believing AI is omnipotent. These "crayfish-brain" bosses keep making outlandish demands, almost treating AI as a silicon-based bodhisattva who answers all prayers and relieves all suffering.
At the same time, in a widely discussed article titled "The layoffs will continue till we learn to use AI," author Arnav Gupta faces potential layoffs on May 20. He analyzes that although AI has significantly boosted code output and token consumption, corporate revenue has not grown correspondingly, leading to a disconnect between "input-output-outcome." In the article, he criticizes a phenomenon where demand rapidly inflates after AI adoption, as AI amplifies the impulse to "just do it first." More and more bizarre demands from CEOs and product managers emerge, but there is little mention of concrete results; instead, layoffs are used to boost stock prices.
In fact, this kind of discussion is nothing new. Over the past year or so, similar complaints have been endless. There used to be a joking saying that the real money-making industries after AI are education and self-media. Recently, both OpenAI and Anthropic have released their ARR data, prompting the industry to boast about token monetization. However, much of this so-called token monetization comes from companies with inflated demands, burning tokens frantically to keep up with the "trend," while AI remains decoupled from their revenue.
What I find most astonishing in these discussions and complaints is why product managers, as demand managers, have become accomplices in the disorderly expansion of business?
From a product manager's perspective, many so-called "demands" are not really demands at all; at best, they are just "requirements."
The difference between the two is significant.
"Demand" means a real problem exists and is worth solving. "Requirement" is merely an expression of organizational members' own anxiety, KPIs, power, and sense of existence—an overflow of self-worth. This issue is very common in demand management: many people confuse the problems they need to solve with the things they want to do.
Before AI, these "requirements" would be intercepted by the high "transaction costs" within the organization. From proposal to PRD, scheduling, development, testing, and launch, a feature consumes a lot of communication, collaboration, and resources. Many immature ideas would naturally die off in this process.
After AI, these unworthy "requirements," due to their cheap implementation cost, are packaged as "product strategy," "long-tail opportunities," "AI transformation," and other high-sounding terms, piled up as everyone's workload. The ultimate question they point to is simply: "Am I keeping up with the AI trend?"
Strangely, product managers, who are supposed to manage all this, have not played a role in the business flow but have instead become producers of these unreasonable "requirements." This shows that for a long time, the role of "product manager" has been alienated into a mere node for demand transmission within the company.
We generally believe that product managers are primarily responsible for demand management: they define problems, constrain demands, and make value judgments, occasionally handling project management tasks. But in many companies, they have become cogs in project management, spending all day receiving demands, running processes, creating prototypes, writing documents, and passing messages up and down.
If product managers spend their days doing these "plumbing" tasks, no longer acting as a firewall against the company's disorderly expansion, and no longer standing from the user's perspective to fight for their value, it's no wonder they become a source of demand pollution.
Once they lose the core value of "demand managers," in the post-AI era, they will inevitably struggle to find their position within the organization. In the foreseeable future, organizational collaboration will inevitably settle into the context of agents, and product managers with only transactional attributes will be worthless.
As Arnav says in his article, the layoffs in these "AI transformation" companies stem from the inherent redundancy of human resources within large companies, not from AI actually reducing costs and increasing efficiency. These "plumbing" product managers are part of that redundancy. To survive, they have to go full throttle in producing "demands."
For a long time, self-media has loved to hype that "AI has replaced XX positions," and the "product manager" role, the most cursed by programmers, naturally made the list. If you ask me whether product managers are still needed in the AI era, I would say we indeed don't need so many people to "do the plumbing."
But if you ask me whether we still need people to manage demands in the AI era, I would say we need them more than ever, and their importance is growing. Whether that person is the boss, the CEO, or a product manager doesn't really matter. The easier it is to generate the illusion of "omnipotence," the more we need someone to pull on the reins and align the "hallucinations" within the organization.
We need product managers because we need someone who, when the boss is driven by a "crayfish brain" and burning tokens frantically, can calmly point at the pile of PPTs and say: "This is just a 'requirement,' not a 'demand,' because it doesn't solve any real problem."
In a sense, product managers are transitioning from "producers of features" back to "gatekeepers of value."
Users don't care whether you use AI; they only care whether you can help them solve their problems. This is the most fundamental product value, and the only value consensus that should be upheld in the AI bubble.