I wrote previously about the costs of creation and production reaching zero multiple times before and I spent the weekend meditating on these ideas again. In my view, AI is creating deflationary pressure on intelligence for a few fundamental reasons, even though it (currently) is economically unviable under current management regimes of token-mAxxing…

Abundant Supply Meeting Every Demand

Intelligence used to be scarce. You needed to hire experts, wait for their availability, pay premium rates. AI makes intelligence abundant and instantly available. When I can answer questions, write code, or analyze data for essentially zero marginal cost, the economic value of those cognitive tasks drops dramatically… as my clients know… our conversations are being listened to by my agents and they are working immediately during our conversation…

• Traditional intelligence markets operated on scarcity economics - there were only so many doctors, lawyers, programmers, or analysts available at any given time, which created natural price floors. AI eliminates this constraint entirely, creating what economists call a “post-scarcity” condition for cognitive labor where supply can expand infinitely to meet any spike in demand without quality degradation.

• The marginal cost approaching zero fundamentally breaks traditional pricing models. When it costs a company thousands of dollars in salary to have an employee spend a day analyzing data versus pennies to have AI do it, the entire cost structure of knowledge work collapses. This isn’t just cheaper. It’s a different economic category altogether, like how digital music didn’t just make records cheaper but essentially made distribution costs disappear.

• Unlike human experts who need sleep, have limited attention, and can only focus on one task at a time, AI can be instantiated millions of times simultaneously. This means that during peak demand periods when human intelligence would be most valuable and expensive, AI scales effortlessly, preventing the price surges that would normally occur in tight markets… and we’re already seeing creative ways for businesses to maximize output and lower token costs…

Commoditization of Previously Specialized Skills

Tasks that once required years of training like writing marketing copy, basic legal research, data analysis, or translating languages are becoming commodified. It’s like how calculators made arithmetic skills less economically valuable. The rarer the skill was, the more deflationary pressure AI creates when it can perform it (accurately).

• Skills that took years or decades to develop professionally now have a dramatically shortened “moat” period. A junior lawyer might spend five years learning to draft contracts competently, but AI can produce similar quality work immediately. This doesn’t just devalue the skill. It restructures entire career paths and educational investments, making the ROI on specialized training much less certain.

• The standardization effect accelerates commoditization. Once AI can perform a task, that task becomes defined by what AI can do, creating a ceiling on how much that skill can command. If “basic data visualization” or “competent blog writing” can be done by AI, those capabilities become table stakes rather than differentiators, forcing human workers to compete on dimensions beyond just performing the task itself… this is where I am spending MOST OF MY TIME - TRAINING MY AGENTS.

• Professional guilds and credentialing systems that maintained artificial scarcity are being undermined. You historically needed someone with specific certifications or degrees for certain work, which limited supply and maintained pricing power. When AI can do the actual work regardless of credentials, it exposes how much of professional pricing was based on regulated scarcity rather than intrinsic task difficulty. I would know, I’m on the board.

Speed and Scale Effects

AI can do in seconds what might take a human hours or days. This doesn’t just make intelligence cheaper, it makes it effectively infinite in supply at any given moment. You can’t compete on price with something that costs fractions of a penny and works instantly… however! This requires HUNDREDS OF HOURS OF TRAINING TO DO WELL.

• The velocity differential creates a compound deflationary effect. If AI can complete 100 analysis tasks in the time a human completes one, it’s not just 100x cheaper per task, it enables entirely new use cases that weren’t economically viable before. This expands the market while simultaneously crashing per-unit prices, similar to how cheap steel didn’t just make existing products cheaper but enabled skyscrapers and cars that couldn’t exist at previous price points.

• Latency elimination changes decision-making economics. Businesses previously had to carefully ration expensive expert time, batching questions and analyses. With near-instant AI responses, the friction cost of getting intelligence drops to zero, meaning decisions that weren’t worth the overhead of consulting an expert become routine. This massively increases utilization while decreasing the per-query value.

• The “always-on” nature prevents price recovery cycles that would normally occur in labor markets. Traditionally, when demand for specific expertise surged, prices would rise until more people trained in that field, creating cyclical pricing. AI doesn’t have training pipelines or supply constraints, so there’s no mechanism for scarcity-driven price increases to ever occur again in AI-capable domains.

The “Good Enough” Threshold

For many tasks, AI output is “good enough.” Maybe not better than the top 1% of human experts, but better than average and vastly cheaper. This creates downward pressure on pricing across the entire market for intellectual work, not just at the bottom. Here is where patient and diligent operators win the long game: they train for many months…

• Quality requirements follow a power law distribution. Most tasks don’t actually need the top 10% expert quality, they just need to be above a certain acceptable threshold. AI hitting that “good enough” bar for 80% of use cases doesn’t just capture that 80%, it reprices the entire market downward because even specialists must now justify their premium pricing against a cheap alternative that works most of the time. For client work, this needs to be near 100% though… (good training).

• Risk tolerance shifts dramatically when the cost of trying something is negligible. Previously, hiring a mediocre consultant was a significant risk because of the money invested, so companies paid premiums for proven expertise. With AI, the cost of a “mediocre” attempt is so low that users will try AI first and only escalate to expensive humans when AI fails, inverting the traditional decision tree and capturing most of the volume at commodity prices. We want human in the loop as much as possible!

• The comparison anchor fundamentally resets customer expectations. Once people experience getting decent answers instantly for free, even excellent human expertise feels expensive by comparison. It’s not that the human work is worth less objectively, but the psychological reference point shifts. This is similar to how free internet content made people unwilling to pay for most journalism, even high-quality journalism. The anchor point moved… and frankly, the value moved too…

The irony is that while this makes intelligence cheaper as a commodity, it potentially makes judgment, creativity, taste, and human connection more valuable, the things that are still distinctly human.

We’re seeing a revaluation of what kinds of intelligence actually matter economically and I am constantly at war with myself as I shed all of the years of learning things that I now have a skill, mcp, api, or webhook that does it better.

AI can destroy the identity of artists.

All the best,
ps