Why the future of AI expertise is person-based, not framework-based
A skill is one book. An agent is the professor. Hold that thought, because it's the whole argument.
I have spent the last year building a pipeline that turns expert methodology into AI skills — taking someone's framework and packaging it so AI can actually apply it to your work, not just explain it back to you. It works. A Shape Up skill will look at your stuck product bet and tell you the thing Ryan Singer would tell you. That's real, and it's miles ahead of asking a base model to "help with my roadmap."
But somewhere around the tenth skill I built, the limitation started bugging me. Day Trading Attention makes a perfectly good skill. The trouble is that Gary Vaynerchuk is not Day Trading Attention. He's twenty years of interconnected thinking — Crush It, Jab, Jab, Jab, Right Hook, Crushing It, Day Trading Attention, plus thousands of talks, podcasts and posts. One book is maybe 10% of what the man actually knows. So a skill built from one book hands you a tenth of the expert and calls it a day.
That's the gap I want to talk about. Not the gap between AI and experts — we've covered that one. The gap between a skill and an agent. Between the textbook and the professor.
Nobody actually asks for a framework
Here's something that has never happened in the history of work. Nobody has ever sat down at their desk and thought, "I shall now apply the Jab, Jab, Jab, Right Hook methodology to my content calendar."
What they think is: "My marketing's not landing. What would Gary V do?"
People don't reach for expertise by framework. They reach for it by person. When your product bet's been stuck for two weeks, you don't go looking for "the appetite-setting technique" — you wish you could grab twenty minutes with Singer. When you've been avoiding a hard conversation with someone on your team, you're not hunting for "the Radical Candor quadrants" — you wish Kim Scott was sitting across the desk.
The framework is how the expert organises their own head. It is not how you experience their help. You experience their help as judgment — a specific person looking at your specific mess and telling you what they see. The framework is the scaffolding behind that. It matters enormously to them and it's basically invisible to you.
And this is exactly why framework-first delivery hits a ceiling. A skill makes you pick the framework before you get the help. You have to already know your problem is a Shape Up problem before you can reach for the Shape Up skill. But the people who most need expert judgment are precisely the ones who can't reliably name which framework applies — diagnosing that is itself an expert act. We're asking the beginner to do the expert's first move before we let them in the door.
An agent deletes that step. You describe your situation in your own words, the agent works out which framework fits — or pulls from three at once — and answers the way the expert would. You never had to know your content problem touched three different things they'd written about across a decade. The agent knew.
What the agent has that a pile of skills doesn't
The difference isn't just more. Stapling ten skills together doesn't make a professor — it makes a very thick textbook. The thing that makes an agent an agent is the connective tissue. Under the hood it's a knowledge graph: the expert's ideas as nodes, the relationships between them as edges. That structure buys you four things a stack of separate skills never can.
1. It routes across everything at once. Ask an agent a question and it doesn't grab one methodology and stop. It finds the most relevant ideas wherever they live in the body of work, then drags in the connected ones a hop out. A pricing question might pull in a positioning framework, a principle about customer psychology, and a throwaway tactic from a podcast three years ago — assembled into one answer. A single skill only ever sees the slice inside its own four walls.
2. It carries the core beliefs. Every expert worth licensing has a handful of stable axioms sitting underneath everything they say. For Gary V, "attention is the most valuable asset in business" is load-bearing for basically every tactical call he's ever made. An agent extracts those beliefs and holds them as anchoring context on every single response. That's why a good agent sounds like one coherent human across wildly different topics instead of a committee. The beliefs are the spine.
3. It explains evolution instead of contradicting itself. Experts change their minds — that's a feature of being any good. What they said in 2013 isn't what they say in 2024, and the gap between the two isn't an error, it's the most interesting thing they've got. "I used to tell people to pick one platform and go deep. Now I tell them to be everywhere. Here's what changed." A skill frozen on a single book can't do that; it doesn't know there was an earlier position to grow out of. An agent tracks the shift as an explicit relationship and narrates it. The contradiction becomes a lesson.
4. It knows where its own edges are. Ask a real expert something outside their lane and a good one tells you straight: "I haven't properly worked on supply chain, but my principle about killing friction would point you here..." An agent does the same — it knows what it covers and what it doesn't, and it says so instead of confidently inventing. Boundary awareness is a feature, not a bug.
And to be clear, none of this is impersonation. The agent isn't doing a party-trick voice of the creator. It's applying their actual methodology — tested, attributed, pulled from real source material — and synthesising it in their voice. Same scope as the underlying skills, finally composed into the shape people actually ask for help in.
The open-source world is busy building the wrong thing
I want to be fair about what already exists, because a lot of serious energy is pouring into agents that solve a completely different problem.
The open-source agent ecosystem is big and getting bigger. The Agency — one of the popular repos — has around 15,000 GitHub stars and ships 61 role-based agents for free. LangChain Hub, CrewAI, same crowd. Real tools, real adoption. But look at what the agents are: "Content Creator." "Sales Rep." "Product Manager." They encode roles. They're configured to behave like a generic version of a job title.
Here's the thing about a role — it's a position, and anyone can copy a position. There's nothing proprietary about "act as a content strategist and give me best practice," which is exactly why 61 of them are free. Those 15,000 stars don't prove role agents are valuable. They prove people desperately want specialised agents and are settling for generic ones, because the licensed version doesn't exist yet.
A person is not a position. "Act as a content strategist" gives you the bland average of everything ever written on the topic. "Apply Gary V's actual judgment to my situation" gives you his real positions — that SEO is dead for most small businesses, that you give value three times before you ask for anything, that attention beats every other asset on the board. That's not best practice. That's one person's hard-won, occasionally contrarian, judgment. The gap between the generic role and the specific person is the entire product.
And it's the part nobody can lift. Anyone can extract a framework from a public book — the book's right there on Amazon. Nobody can reconstruct twenty years of interconnected thinking across dozens of works, thousands of talks, and the tacit judgment that wires it all together — not without the expert in the room helping build it. The extraction complexity isn't an annoying cost on the way to the product. It's the moat. "Person over position" isn't a slogan I'm fond of. It's a description of the only thing here that's actually defensible.
Two ways to use a person
When your unit is a person instead of a framework, the ways you can use them multiply — because an expert in your corner was never a single-purpose tool. Sometimes you want to talk it through. Sometimes you just want the verdict.
The first mode is picking their brains. You open a conversation, lay out your content strategy, and the agent coaches you through it — applying frameworks, pointing out how its own thinking has shifted, giving you judgment in the creator's voice. It's the experience of having the person work alongside you on the exact thing you're stuck on. The kind of help you'd otherwise have to book months ahead and pay a small fortune for.
The second mode is using them as a quality check inside your own workflow. You've drafted the Q3 calendar. You run it past the agent as an evaluation step and get back a structured verdict — what's strong, what's weak, what they'd change. No conversation, just judgment applied to a work product. This is also how other software calls the expertise: your marketing AI, mid-task, can ask the agent for a second opinion the same way a human would, programmatically.
Both modes are judgment, not execution — and that distinction matters. The agent doesn't rewrite your calendar for you. It tells you what's wrong with it. A framework can't flex between these two uses because a framework is a single procedure. A person can, because a person is a body of judgment you consult in whatever shape your problem turns up in.
Why I think this is the endgame
I don't say this lightly, but I'm fairly convinced the person-based agent isn't a feature bolted on top of skills. It's the thing skills were always walking towards. Three reasons it compounds.
It meets people where they actually are. A framework asks the user to self-diagnose before they're allowed any value. An agent catches them at the moment of confusion — which is the moment they're actually in. That alone widens the audience from "people who already know what they need" to "everyone who knows they need something." That's most of us, most of the time.
The value per expert goes vertical. A single skill is useful in one kind of situation. A full agent is useful across every situation inside the expert's domain — call it ten to fifty times the occasions to reach for it. For a business, that's the difference between licensing one framework and licensing an entire intellectual presence, and I expect contract values to move accordingly. The back catalogue stops being a back catalogue and becomes a living service.
It finally answers the retention question. A skill is build-and-forget — ship it and there's not much reason to come back. An agent is the opposite. It wants new material, review sessions, evolution management. The creator returns not because we nag them but because the agent is a living thing that reflects their latest thinking, and keeping it current is genuinely worth doing. The ongoing relationship isn't manufactured. It's structural.
Skills are the path. The agent is the destination.
Nobody builds an agent from a standing start, and they shouldn't try. The road runs through skills.
You build your first skill from one book. The pipeline pulls out the thresholds, the decision sequences, the patterns, the formats. It goes live, people use it, revenue comes in. Then you build a second from another work, same way. At two or more skills, something new becomes possible — unification. The platform reads both methodologies, extracts the core beliefs that span them, builds the graph that wires them together, and synthesises one voice across the lot. You validate the beliefs and the voice in a review session. The agent goes live alongside the individual skills — they stay intact as modules, so anyone who wants to run one framework directly still can.
From there it grows. You feed it new material — an article, a podcast transcript, a keynote. The system works out what's genuinely new, maps it against what's already there, and flags anything that smells like an evolution or a contradiction for you to sign off. Approved, it's part of the agent. Your thinking stays current without you rewriting a word. A book is frozen the day it ships. An agent is dated by its last update — and the last update can be last week.
That, for me, is the real shift. "Have you read their latest book?" becomes "Have you talked to their agent?" The book documents what the expert thought. The agent applies what they know — to your situation, in their voice, kept current.
The textbook was always going to be free. AI read all the books years ago. What it still can't do is be the professor — hold the whole interconnected body of judgment and apply exactly the right slice of it to the messy, specific thing in front of you. That's what people have always wanted from an expert. For the first time, we can build it. Skills are how you get there. The agent is why you go.
Cheers, Adam