Is AI automation and agent-building still worth learning, "or we should learn something else. I mean I have heard many guys talking about very automation in 2027 is not worth." A viewer asked that under "Matt Pocock: The Most Valuable Skill to Learn in the Age of AI" on the Nate Herk channel (1 like), and in the comment corpus we scan daily (5,927 comments as of this morning), we found the same question in three more forms over the past two months. The question bundles two different investments, and the split between them decides where learning time goes.
Two bets in one question
One bet is on specific tools: learn Claude Code, learn prompting, learn the automation platform of the month. The other is on judgment: describing work clearly, checking what AI produces, and knowing your own field well enough to recognize a wrong answer. Tools and interfaces change quickly. The judgment is what transfers. The custom-GPT retirement shows why.
What the custom-GPT retirement teaches
OpenAI has scheduled custom GPTs, its no-code reusable chatbots, to retire on Dec 11, 2026, per its retirement FAQ. A GPT bundles instructions, reference files, and connected tools, and OpenAI is migrating them to a plugin system, with ChatGPT Projects suggested for instruction-plus-files work. Whatever happens to the interface, the part that transferred for the people who built them was the setup logic: what to include, what to connect, what to check. Product-specific knowledge depreciates fastest; the structure moves to whatever comes next.
The three that transfer
Describing work clearly. A useful prompt is usually a description: what happened, who's reading it, what should happen next, how you'd say it. The skill predates AI and works as well in a brief for a colleague as in a chat window.
Checking the output. AI answers can sound right and be wrong. Ask for a summary of a contract and the tool may state a deadline that isn't in the document; opening the document catches it in a minute. The habit (open the source, search the claim, drop what you can't confirm) works across tools, and it matters more as tasks get bigger.
Knowing your field. You can only check output you understand. Depth in your own work is what lets you catch the subtle wrongness a machine can produce; a source, a checklist, or a trusted reviewer can help, but someone has to know what good looks like.
How you practice changes the outcome. In a 2024 study of about 1,000 students (Bastani and colleagues), those who practiced with unrestricted GPT-4 access scored about 17% worse on the exam they took without it, while the guardrailed tutor group didn't show the same penalty. The skill is more likely to build when AI checks your work than when it does the work for you.
For the 10th grader: school and AI skills overlap
A 10th grader asked the Dan Martell channel whether he should "focus on my study and grades or should I focus on building skills and study how do money works?" (2 likes). The framing treats school subjects and "AI skills" as competitors. School builds the checking layer: writing clearly, reading carefully, enough math to spot a nonsense number. A course promising AI income shortcuts may teach familiarity with one product; a way to test any course is whether it teaches you to check the output or only where the buttons are. The grades matter, and so does practicing AI on schoolwork that has a correct answer to check against.
For the video editor: what AI changes and what it doesn't
A video editor asked whether learning Premiere or DaVinci is a waste of time now that AI edits video (1 like). AI video tools are taking on more mechanical work: cuts, retiming, caption generation. The parts of editing more likely to remain valuable are the decisions: what to shoot, what the story is, and what reads as good pacing. An editor whose skill is tied to one product's menus faces the same risk as any tool-specific learner. The practical move: learn the craft's decisions with the current AI tools in the loop, and expect to re-learn the mechanics as the software changes.
For the overwhelmed beginner: skip the tool list
Another comment on the same video listed what a 35-year-old non-technical viewer thought he needed to learn: Claude Code, TTS tools, video generation tools, editing skills, prompting. That list names products, but it doesn't yet tell you what to practice. A beginner can start with one task and one tool. Pick a repeated job from your own week that produces word-shaped output (summarizing, drafting, reformatting), run it in a free ChatGPT or Claude account, and check the results yourself. Judgment has something to build on from there. Tool lists can wait until a task demands one.
The "not worth it in 2027" claim
The tool market is difficult to predict a year out, and anyone selling certainty about it is guessing. The trackable part: "AI automation" as a job title may get absorbed into better tools, while the judgment layer (what to hand over, what to check, when an answer looks suspicious) tends to gain value as AI takes on bigger tasks. One practical limit: learning time usually isn't what holds learners back. Practicing on your own work is, so give the one-task exercise above a week before adding anything new.
FAQ
What AI skill should I learn first?
The first move is choosing the task, not the skill. Pick something you repeat weekly that produces word-shaped output, in a place where you'd notice a wrong answer: summarizing notes or drafting a reply passes; picking a business strategy doesn't. Skills attach to the task from there.
Is AI automation still a good career?
The tool work may consolidate as platforms get easier; the demand that created it (handing repeated work to machines) hasn't gone away. The durable version of the career is the judgment side: deciding what to automate, checking what it produces, and owning the outcome when something breaks.
Will AI make my current skills worthless?
It depends which layer of the skill you own. Mechanics that follow rules (formatting, template writing, routine cuts) are the most exposed. Decision layers (taste, diagnosis, judgment about what's worth doing) hold up better, and AI tools may help you practice them faster when you work against feedback.
Do I need to learn to code to use AI?
No. For most everyday work, describing a task and checking the result need no code. Coding helps if you want to build custom agents or automations, and AI tools can make parts of learning it easier, though generated code still needs checking like any output.
Want to go further?
Two posts pair with this one: How to Fact-Check AI Output covers the verification habit in detail, and AI Tools Compared: Which One Should You Start With? narrows the first-tool choice. Pick one repeated task from your week, run it through a free account, and check one output against its source.