Charles Miller argues that standalone AI techniques have a short shelf life because successful ones are likely to be absorbed into the products themselves. In an essay published on October 26, 2026, on The Fishbowl, he challenges the idea that learning the latest prompting tricks creates a lasting professional advantage.
His argument targets a specific kind of knowledge: techniques that someone can learn in weeks, use to get better results from a model, and then see packaged into a default experience a few months later. If the product starts handling that work automatically, knowing how to perform it manually becomes less distinctive.
Miller also makes an important concession. People still need to keep up enough to do their current jobs well. His objection is to the pressure to get far ahead of everyone else, particularly when the knowledge being acquired may age quickly.
The essay offers analogies and reasoning, not labor-market data or a measured timeline for product improvements. Its prediction that techniques will become defaults within months is Miller’s assessment, not an independently established rate of change.
How a Useful Technique Could Lose Its Standalone Value
Miller’s case rests on a feedback loop: users discover ways to make AI systems more useful, and product makers have a reason to incorporate the successful approaches.
A person might initially need to remember a particular instruction or manually arrange a sequence of steps. If a product later handles that sequence, the benefit becomes available to people who never learned the original technique.
The important change is who has to carry the complexity. An expert user may initially supply the missing structure. Once the product supplies it, the same knowledge can become less valuable as a separate service or credential.
Miller does not identify a particular product release that proves this cycle. His essay presents it as a general expectation. Still, the mechanism is useful to examine because it separates two things that are often bundled together under AI skills:
- Getting a useful result today: A technique may improve a current workflow and be worth learning immediately.
- Holding a durable advantage: The same technique may provide little differentiation once it becomes widely available or automatic.
Those outcomes can coexist. A shortcut does not need to remain valuable for years to justify learning it. It needs to save enough time or improve enough work during the period when it is useful.
My view is that this is the strongest reading of Miller’s argument. Treating every temporary technique as worthless would ignore its immediate payoff. Treating every temporary technique as a career moat would ignore how easily software can distribute it.
The Compiler Analogy Explains His Concern
Miller opens with a programming habit from his early experience learning C: the belief that pre-increment, ++i, was faster than post-increment, i++.
The distinction concerns the value an expression produces. Post-increment produces the previous value before incrementing the variable; pre-increment produces the incremented value. For an ordinary integer, if the expression’s result is unused, both leave the variable with the same increased value.
The old performance advice assumed that post-increment would require preserving a copy of the original value. Miller’s point is that an optimizing compiler could recognize when that value was unnecessary and avoid the extra work. The programmer’s ritual could survive after its supposed benefit had disappeared.
In Miller’s analogy, prompting habits face a similar risk. People may continue supplying elaborate instructions because those instructions once seemed necessary, even after changes to the underlying product have reduced their usefulness.
The practical lesson is about maintaining advice. A technique learned for one model or product version carries assumptions about how that system behaves. When the system changes, the advice deserves another look.
I would take that as an argument for retiring stale instructions, not for refusing to learn instructions in the first place. A useful habit should remain connected to a reason. If the reason disappears, continuing the habit can add complexity without an obvious benefit.
Productization Changes the Value of Different Skills
The most consequential distinction is between knowing how to steer a particular interface and knowing what successful work looks like.
Miller concentrates on the first category. A broader implication, derived from his argument, is that people should separate product-specific techniques from the judgment needed to evaluate their results.
| Kind of knowledge | What happens if the product absorbs the technique? | Practical implication |
|---|---|---|
| A particular prompt formula | Its wording may become unnecessary or less distinctive | Learn it when it helps, but revisit it after product changes |
| A manual sequence for a recurring task | Some steps may become automatic | Keep the task requirements clear so the automated result can be checked |
| Domain knowledge and evaluation criteria | The product may simplify execution, but someone still needs to determine whether the result meets the task | Invest in defining quality, recognizing errors, and understanding consequences |
This table is an analysis of Miller’s premise, not a finding reported in his essay. It shows why the question is more useful when framed around which part of the work is changing.
Consider a hypothetical software task. Remembering a prompt that encourages an assistant to review its output is one kind of knowledge. Understanding the required behavior, identifying a missing case, and deciding whether a proposed change is acceptable are different kinds.
A product could make the review step automatic. That would remove some prompting work. It would not, by itself, answer whether the review covered the right requirements.
That distinction also keeps the argument from becoming too broad. Productization can reduce the effort needed to perform a task without eliminating the value of understanding the task. The more consequential the output, the more important it is to know what would count as an unacceptable result.
Keeping Up Is Different From Chasing Every Trick
Miller explicitly preserves the case for staying current:
“I'm not saying it's worthless to keep up with the state of the art.”
His essay contrasts that practical need with an anecdote about an unnamed CEO warning an audience to embrace AI immediately or risk being left behind. Miller rejects the assumption that being early is necessarily more valuable than getting the eventual implementation right.
Frequently Asked Questions
3 questions
1Is Miller saying people should stop learning AI tools?
No. He explicitly says people need to keep up enough to do their current jobs well. His criticism concerns treating short-lived techniques as durable advantages.
