Summary of "Can’t Code Without AI? Get Ready To Pay"

Summary of the video’s main points

The speaker argues that AI coding tools are becoming more expensive (or providing less value for the same cost). This is presented as a predictable outcome of “cheap compute” subsidized by investor money during an earlier growth phase. Now that customer acquisition is “over,” companies will raise prices, and the speaker expects this to continue.
The speaker warns that if you rely on AI because you “can’t code,” you should expect significant cost increases. They extend the same logic to organizations: if companies expect AI to replace developers (“AI do the coding, bro”), they may still end up paying more—especially as token/model pricing rises—making that approach less cost-effective.
The speaker frames AI as a way of creating customer dependence. People become accustomed to fast, effortless answers from frontier models, so slower or manual alternatives feel painful—leading users to pay to avoid friction.
They address the idea of running models locally but argue it’s not a full solution. As model quality improves, local usage demands more hardware (e.g., RAM/SSD). The speaker also claims local running is becoming less feasible due to rising hardware costs.
The speaker contrasts “learning the old way” with “asking the AI.” They argue that real understanding requires effort, such as:
- reading technical materials
- using documentation/man pages
- testing
- implementing concepts manually (They illustrate this by describing how to learn system calls in C using Linux/kernel resources.)
Their proposed solution is independence for both individuals and teams:
- Individuals: build deep skills (“get cracked”) so AI companies can’t easily replace them.
- Companies: hire and develop real engineers instead of treating automated code generation as a cheap substitute.
From a business perspective, they note that companies adopted heavy AI usage because tokens were previously “practically free.” As costs rise, companies may reconsider whether widespread AI use is worth it—particularly if it effectively doubles engineering expenses.
Overall message: AI coding assistance may still be useful, but overreliance will lead to higher bills and reduced leverage. Long-term security comes from developing durable technical competence.