The 6 Skills That Get More Valuable as AI Gets Better

"Learn AI" is the new "learn to code."

Everyone says it. Almost nobody can tell you what it actually means.

Greg Isenberg just dropped a 30-minute video making the opposite argument: stop trying to "learn AI" and start learning the skills that get MORE valuable as AI gets better, not less.

I watched the whole thing so you don't have to (you still should, it's embedded below). Six skills. None of them need a degree. All of them could be started this weekend.

Here's the breakdown, plus my honest take on which ones actually matter for a bootstrapped SaaS founder.

1. Setting up and managing AI agents (plus running local models)

Greg calls this "the grown-up version of prompt engineering."

Typing a good prompt into ChatGPT was the 2023 skill. The 2026 skill is designing a little AI employee: context, tools, permissions, memory, a goal, and a way to check its own work before it bothers you.

Why it's valuable: most companies are about to have the same problem. 10 AI tools, 50 workflows, a bunch of half-working automations, and nobody who can turn that mess into an operating system.

The person who can walk in and say "here's the support agent, here's the research agent, here's the sales follow-up agent, here's what each is allowed to do, and here's how we know it's working" becomes very hard to replace. Agencies are already building whole businesses around this (TwiLead does exactly this kind of work with business automations for lead-gen clients).

Greg's starter rep: build a daily briefing agent for yourself. Three sources (calendar, a notes folder, saved links). Its job: tell you what matters today, what decisions are waiting, what follow-ups you owe people. One rule: it must show sources and ask approval before sending anything.

That one boring project teaches you context, retrieval, tool use, permissions, and evals. Which is the shape of every serious agent inside a company.

→ The mistake: building an all-knowing mega-agent first. Build one small agent, make it save you 10 minutes a day, then stack.

I run 11 AI agents in my own business, so I'm biased. But he's right. I wrote about this shift in AI agents are the new SaaS if you want the founder angle.

2. Marketers who can build distribution (not just post)

People confuse distribution with posting.

Distribution is knowing where attention already lives, what people are anxious about, what language they use to describe the problem, and how to turn that into trust before you ask for money.

When anyone can ship a landing page or a SaaS in a weekend, the bottleneck moves to one question: can you make people care?

Greg's starter rep is a distribution map:

• Pick a niche (dentists using AI, solo consultants, Shopify operators)
• Write down the 20 places their attention goes: newsletters, creators, Reddit threads, Slack groups, podcasts, search terms, tools they already pay for
• Write one painful sentence they'd actually say out loud ("I know I should follow up with leads faster, but by the time I sit down, half of them are cold")
• Then write 20 hooks for the same idea: curiosity hooks, fear hooks, status hooks, money hooks

The mindset shift: stop asking "how do I promote this?" after the product is done. Start asking "what existing desire am I pointing this at?" before you build.

If you're doing this for a SaaS, I broke down the channels that actually work in SaaS distribution channels.

3. Robotics engineers who can build hardware, wire in AI, and source manufacturing

The one nobody on your timeline is doing.

Greg's framing stuck with me: the last decade rewarded people who moved pixels. The next decade rewards people who can move atoms too.

Robotics used to be a PhD thing. Now you have open-source robot learning projects (Hugging Face's LeRobot), cheap camera modules, low-cost arm ecosystems like the SO-100 and SO-101, better simulation, and small vision-language-action models you can train without an industrial setup.

The valuable person isn't the AI researcher. It's whoever can make the whole loop work: cheap arm on the desk, camera mounted, demonstrations collected, model fine-tuned, one useful task repeated reliably. Then look at a supplier listing on Alibaba and know if the thing is actually manufacturable.

Greg's starter rep: buy or assemble a low-cost arm, teach it ONE boring task (sorting three objects, pressing a button), and document every failure. The bad camera angle, the lighting that changed, the model that looked smart until the object moved 2 inches.

On the sourcing side: ask suppliers for a sample before bulk, motor specs, CAD files, lead times, minimum order quantities, and a short video of the part doing your exact task.

This skill is rare because it sits between worlds. Software people avoid hardware. Hardware people avoid distribution. The person who connects both builds things that feel like science fiction but sell like tools.

4. Curators who can yap on camera

The internet is drowning in information. The person who makes sense of it in public wins.

Curation has evolved past "here are five links in a newsletter." The curator of the agentic era watches the timeline and says "this matters because..." for their specific niche. New model demo, weird startup launch, pricing change: what should you learn, what should you ignore, what's hype?

Greg says the quiet part out loud: you don't need to be super smart to get 50,000 followers in a niche. You don't need net-new ideas. You need taste and consistency.

Why is "yapping" (his word) working right now? Because the algorithms are drowning in AI slop and are actively prioritizing raw, authentic, face-to-camera takes. Nothing is more un-AI than a real person saying "here's what I found, here's my take."

Greg's starter rep: a 7-day curation sprint. Pick one lane. Every day, find three things and make one short video with the same structure:

→ I saw this. Most people think it means this. I think it actually means this. Here's the move.

That structure forces a take. A take is the difference between curation and forwarding links.

Build a swipe file as you go: great hooks, great analogies, comments that reveal what people are genuinely confused about. Your outputs are only as good as your inputs.

5. The builder-distributor (the one-person startup)

This is the one that matters most if you're reading this blog.

For decades there was a clean split: one person builds, one person sells. Your Wozniak and your Jobs. AI is compressing that split into one person who can prototype the product, write the landing page, record the demo, DM the first 100 users, edit the clips, and iterate. No handoff, no waiting.

When Sam Altman talks about the one-person billion-dollar startup, this is the person he's describing.

Greg points at Peter (OpenClaw) as the live example: clearly a great builder, but look at his X account. He's also doing marketing, support, and storytelling in public, all at once.

Greg's starter rep is a 48-hour loop:

• Pick one tiny problem you personally understand
• Build the smallest version with AI (a script, a form, an ugly web app)
• Create 10 pieces of distribution before you feel ready: one demo video, three clips, three posts, two DMs to people with the problem, one landing page

You're training yourself to stop separating the product from the market. AI made building fast, which means the marketing reps can start way earlier. You don't spend 6 months wondering if people want it. You spend a weekend earning a real reaction.

Most people do half the loop. They build in private forever, or they talk in public forever. The builder-distributor cycles between both.

6. IRL community builders

The old-school one, and honestly my favorite.

Greg's logic: AI makes content, software, and advice abundant. Scarcity moves to belonging, trust, and context. Who actually knows you? Who answers your text at 10pm? Who introduces you to a customer?

Real rooms become more valuable as more work moves to agents and feeds.

And the money is real. Jason Lemkin built SaaStr into a massive business around one event. But Greg's bet (and mine) is that people don't want mega-conferences anymore. They want small, bespoke rooms.

Greg's starter rep: host 6 to 8 people around one sharp question, at a dinner or a breakfast or a walk. Something like "what are you automating in your company right now?" Then send a recap after: best quotes, inside jokes, one follow-up everyone should do. The recap turns the room into a network.

I'll add my own number here. I once paid $13,000 to join a mastermind while making $15K to 20K a month. Stupid decision, right? Six months later I was at $75K a month. The room did that, not a course. It's the whole reason I built a community for SaaS founders myself.

Over time the room becomes a media asset, a recruiting asset, a deal-flow asset, and a life asset.

The pattern behind all six

Look at the list again:

• Agent operators
• Distribution marketers
• Robotics tinkerers
• Curators
• Builder-distributors
• IRL community builders

Not one of them is "learn AI." Every single one is a human skill that AI happens to amplify.

Greg's bigger point: the future favors the person who combines a few of these. The agent person builds tools for the community. The curator turns the best conversations into content. The builder-distributor launches products from it.

Pick one. Do the rep this weekend. That's the whole assignment.

FAQ

What are the 6 skills Greg Isenberg says to learn instead of "learning AI"?

Setting up and managing AI agents (including local models), distribution marketing, robotics engineering with AI and manufacturing sourcing, curation and short-form video, the builder-distributor combo (shipping product AND getting attention), and IRL community building. All six get more valuable as AI gets better.

Why is "learn AI" bad advice?

Because "AI" isn't a skill, it's a layer. Prompting gets commoditized fast. The durable skills are the ones AI amplifies: judgment about which agents to build, taste in curation, trust in communities, and the ability to make people care about what you ship.

What's the fastest skill to start this weekend?

The builder-distributor loop. Pick one tiny problem you understand, build the smallest version with AI in 48 hours, then create 10 pieces of distribution for it (demo video, clips, posts, DMs, landing page) before you feel ready.

Do you need to be technical to benefit from any of these?

No. Curation, distribution, and IRL community building need zero code. Agent setup is mostly configuration and judgment. Even Greg's robotics path starts with a cheap arm and documenting failures, not a PhD.