Durable AI skills: decomposition, verification, taste

September 6, 2026 · 5 min · Free, with full transcript

If the model you use today is replaced in six months, what is left of the AI skill you built? The position here is blunt: prompt tricks tied to a specific model, like telling GPT-4 it is an expert, are quirks vendors patch out on purpose, and tool certifications age faster than the tools. Anthropic's Claude interface barely resembles its version 18 months earlier. You walk away with three transferable skills and a way to practice them: problem decomposition, turning "increase retention" into ordered executable steps; verification literacy, treating every output as a claim, as the sanctioned 2025 law firms citing invented cases learned; and taste, the domain judgment a model cannot supply.

Listen to the podcast

4 min

Chapters

Key takeaways

  • Stop optimizing for model quirks like role prompts, because vendors train those workarounds out within months.
  • Break fuzzy goals into ordered steps a machine can execute, such as pull 30 day cancellations, cluster reasons, draft win back emails per cluster.
  • Treat every AI output as a claim, ask for sources and periods, and spot check one number by hand.
  • Write the first draft yourself sometimes so you keep a reference point for what bad work feels like.
  • For one recurring AI task, write the human steps before opening the tool and one sentence describing how you will know the output is wrong.
Read the full transcript

Host:Welcome to the Leaders Insights Podcast. Today's episode, AI skills that stay relevant as tools change. Last week, OpenAI shipped yet another model, and half my LinkedIn feed is people panicking that the prompt writing tricks they spent 2025 learning are already dead. Are they right?

Expert:Mostly yes. If you built your whole edge on knowing that GPT-4 liked being told you are an expert, congratulations. You optimized for a quirk that vendors patch out on purpose. The models are being trained to need less handholding. Your clever incantations have a shelf life measured in months.

Host:That's brutal for the people selling $200 prompt courses.

Expert:It is, and I don't feel bad about it. There's a real statistic here. The article points to how tool-specific certifications age faster than the tools themselves. Anthropics' Claude interface in September 2026 barely resembles what it was 18 months ago. If you memorized menus, you learned furniture, not carpentry.

Host:Okay, so what's the carpentry? Give me a skill that survives the next model drop.

Expert:Problem decomposition. The unglamorous ability to take a fuzzy business goal and break it into steps a machine can actually execute. Increase retention is not a prompt. Pull the customers who canceled within 30 days. Cluster them by why they left. Draft three win-back emails per cluster. That's a plan. The model changes. That thinking doesn't.

Host:That sounds like project management with extra steps. Why does it matter more now?

Expert:Because the models got good enough to punish vagueness in a new way. They'll confidently do the wrong thing beautifully. A 2026 report from GitHub, and yes, they sell the coding assistant so they can integrate accordingly. Found developers accept AI suggestions faster than they verify them. Speed without decomposition just means you ship the mistake sooner.

Host:You mentioned verification. Isn't checking the AI's work exactly the thing everyone's too lazy to do?

Expert:That's the second durable skill, and it's the one people skip. Call it verification literacy, knowing how to test an answer instead of vibing it. If the model gives you a market size, you ask which sources over what period, and you spot check one number by hand. The professionals keeping their edge treat every output as a claim, not a fact. The tool improves. Your skepticism should not.

Host:Give me a case where skipping that blew up in someone's face.

Expert:Lawyers, multiple firms in 2025 filed briefs citing cases the model invented, hallucinated, and got sanctioned by actual judges. The tool wasn't the problem. Nobody ran the one check that mattered. Does this case exist? A paralegal with verification literacy would have caught it in 90 seconds.

Host:So decomposition, verification, what's the third thing or are we not doing three?

Expert:We're not doing a tidy list. But there is a third that's underrated, taste. Knowing what good looks like in your field. The model can generate 40 versions of a landing page. It cannot tell you which one your specific customer trusts. That judgment comes from domain experience, and it's the least automatable thing you own. People with taste direct the tool. People without it get directed by it.

Host:Isn't taste just a fancy word for gatekeeping? How does a junior person build it?

Expert:By doing the boring work anyway. You still write the first draft yourself sometimes, so you know what a bad one feels like. Juniors who let the model do everything from day one never develop the reference point. It's like learning to cook only by ordering takeout. You can't tell when the sauce is off.

Host:Let's get concrete. Someone's listening mid-career worried their AI skills evaporate every quarter. What do they actually do Monday morning?

Expert:Take one recurring task you already use AI for. Before you touch the tool, write out the steps a competent human would take in order. That's your decomposition. Then define in one sentence how you'll know the output is wrong. That's your verification test. Do that for a week and you've built something no model update can delete.

Host:So stop collecting prompts. Start writing your own checklist.

Expert:Exactly. The prompt is disposable. The thinking behind it is the asset. Guard that and every new tool just makes you faster instead of obsolete.

Host:That's the note to end on. Skills over shortcuts. That's a wrap. Fresh AI briefings drop daily at MBA-training.com.

Read and go deeper