It’s crazy out there right now folks.
The LMS is dying (apparently), Roger, the latest automated AI workflow, is going to replace my content engine and make me trillions of dollars while I sleep (allegedly), and Ninja claw, Tigerclaw or Dragon claw will do life for me so I can just wander around asking “why am I here?” (how fabulous).
Jokes aside (I do love them, though), I’ve never seen such a typhoon of pressure on the everyday human.
So, I think we all need to…BREATHE.
Better? Great.
What we can focus on is: What actually matters to us?
I believe that part of the answer to that is our skills. I get it’s hard to build the “right” skills to work with AI. That’s why I focus on skill sprints of 3-6 months these days because so much is expiring, evolving and emerging all at once.
In this one, we’re exploring 3 high-value AI skills you need in 2026.
What if your AI skills from last year are already outdated?
I know that’s a scary question, but damn does it move fast.
One week I’m happy making my AI assistants smarter with internal knowledge files and now I’m running around extending their capabilities with MCP’s, plugins and all sorts of sorcery.
My point being, sitting still for too long on the technical side of AI tools is becoming somewhat dangerous. I know too many people chasing after tools, but I’m more interested in cross-platform skills that enable high value from the majority of tools.
I have 3 today which I believe will serve you well across the never-ending spectrum of AI applications.
What are AI skills?
An engineer will have a different view on this vs an L&D pro.
I break AI skills into two components:
- One component is “How do I use ‘x’ tool and it’s features”
- The second is “How do I work with AI tools to get the best outcome?”, which would involve a host of metacognitive skills and behaviours.
AI ‘fluency’ and ‘literacy’ are better ways for us to frame “AI skills”
These frameworks acknowledge that using a generative AI tool, most popular being an LLM, is about more than the technical skill of using the tool alone. Personally, I like the term ‘AI fluency’ because the technology moves so fast. Which means both the tech skills and behaviours you need to work with it are forever updating.
For the purpose of today’s conversation, we’ll focus specifically on the technical skills side of this because I have written a lot lately on the behavioural side.
Know what’s expiring, evolving and emerging (The 3 E’s)
I do a quarterly skills review.
It used to be every half year, but AI changed that plan.
Your skills are an ever-flowing organism (best analogy I could think of in this moment). That means they never stand still, and without attention, they’ll degrade.
If your goal is to build a talent stack (a combo of your skills, behaviours and attitudes) that’ll keep you employed for the long term, you’d be wise to perform regular maintenance on yours too.
I use a simple framework I picked up from a Gartner report over a decade ago.
All you do is look at your current skills and ask:
- What skills are expiring and no longer serve me and/or the world today?
- What skills do I need to evolve to meet the demands of today?
- What are the emerging skills I can get ahead of?
While I say simple, that doesn’t always equal easy.
You could even spin this as a bit of a life analogy too. We all need to know what to leave behind, what to double down on and what to keep a lookout for.
I’ve tagged each of the 3 AI skills I recommend you invest in with one of these categories to give you a sense of where it’s in your current skill cycle.
3 high-value AI skills you need in 2026
1️⃣ Prompting [Evolving]
No, it’s not dead, but it has evolved.
AI skills have such a short shelf life, and prompting was hailed to be the most important skill of the century. While I didn’t agree then (or now), it is the main way we interact with LLMs.
At its core, prompting is just inputting a query.
A lot of frameworks from 2023-2025 are no longer needed because models have become much more capable with memory and custom instructions. Prompting is also weird because no one template works and two people with the exact same prompt can get widely different outputs.
You still need to prompt, just less skilfully than before because of our next two skills.
Bonus: How To Learn The Meta-Skill Of AI Prompting
2️⃣ Context Layering & Context Engineering [Evolving/Emerging]
These sound the same but I’m framing them in two ways.
One for builders/engineers and the other for non-technical users.
For builders, context engineering is the science of high-signal curation. They’re not stuffing AI with every possible file. The goal is to build dynamic systems that “just-in-time” retrieve the exact tool, specific knowledge chunk, or agent-to-agent communication needed for the next step.
This is not too dissimilar from what end users like me, and you can now do.
As we start using more AI agents, knowing how to provide the right source, in the right format at the right time is critical. We already see these opportunities with memory, custom instructions, connecting our favourite apps and uploading multiple file sources into our chats.
Context engineering for the everyday human is about building the infrastructure around the AI, so it has everything it needs to make better decisions.
An art and science in itself.
We could say that prompting + context layering + skills/instructions = the 3 layers of a strong AI response.
Fyi, I have a guide and a video series on context layering/engineering coming in a few weeks.
3️⃣ Creating Skills for AI [emerging]
The best example of this is Claude Code and Cowork.
You can create a skills package which teaches the LLM how to perform a task. It’s a combination of context, instructions and guidelines that can repeatedly do the task with little oversight.
To make the best use of this, you’ll need to get comfortable with creating Markdown files. In these, you’ll unpack not only how to complete a task but why, and all the micro elements that make up the big choices. Then maintaining that skill just like you do with your own.
We’ll see this capability with more LLMs across this year. The great thing is you can take your .md files to any tool, so maybe we’ll all be building .md files now.
The real skill here is knowing how to break down a process for an LLM to understand with a combination of instructions, examples and evaluations.
Final thoughts
Ok, folks.
That’s my thinking out loud for the day.
Obvs, AI moves so fast, these are relevant skills right now, but maybe not forever.
Before you go… 👋
If you like my writing and think “Hey, I’d like to hear more of what this guy has to say” then you’re in luck.
You can join me every Tuesday morning for more tools, templates and insights for the modern L&D pro in my weekly newsletter.
