In short
- Being the best in the world at one skill is a valid strategy with very few winners. Being the only person in your market who can do three things at once is a strategy with many.
- A stack is not a list. Three skills become a stack only when they are used in the same decision, by the same person, without a handoff.
- The market evidence is real but has an edge: hybrid roles pay more and are harder to automate, and they are also mostly not entry-level.
- Build one spike first. Then add multipliers. Then add the binding skill that lets other people buy the result.
Where the idea comes from
The phrase most people know is talent stack, popularised by Scott Adams, the cartoonist behind Dilbert. In a 2017 Uncommon Knowledge interview at the Hoover Institution, Adams described his own career as having compiled a set of complementary skills while not being the best in the world at any of them — he draws, he writes jokes, and he has business training. Few people have all three. In that same conversation he makes a point that gets lost in most retellings: "The talent stack is a system." It is something you build on purpose, not a description of whatever you happen to know.
The popular version of the idea is usually stated as arithmetic: if you are in the top 25% at two things, you are one in sixteen. Be careful with that. The maths only holds if the two skills are statistically independent, and most skills are not — people who are good at writing tend to be good at structuring arguments, and people who are good at SQL tend to be good at spreadsheets. The honest version of the claim is weaker and still useful: rarity comes from distance between skills, not from the number of skills. Two adjacent skills barely multiply. Two distant skills that are still forced to meet in one job multiply a lot.
What the labour-market data actually shows
Adams gave the idea a name. Job-posting data gave it a price tag.
In January 2019, Burning Glass Technologies published The Hybrid Job Economy, based on a database of close to a billion job postings. Its findings on combined skills are specific:
- Fully one-quarter of all occupations in the US economy showed strong signs of hybridisation, and about one in eight postings was highly hybridised.
- Postings for the most hybridised jobs were projected to grow 21% over ten years, double the 10% projected for jobs overall.
- The report estimated 42% of all jobs could potentially be replaced by technology, compared with only 12% of hybrid jobs — because hybrids depend on judgement rather than rote steps.
- Adding a single skill to an existing role could raise salaries by up to 40%. The report's foreword cites marketing managers who know SQL earning 41% more than those who do not, and workers with project-management experience earning 21% more.
More recent posting data points the same way for AI specifically. Lightcast's July 2025 analysis of more than 1.3 billion job postings, Beyond the Buzz, found postings requesting AI skills advertised salaries 28% higher — and, importantly for stackers, that 51% of postings requiring AI skills sit outside IT and computer-science occupations. The demand is for AI skills attached to something else: marketing, HR, finance, operations.
Two honest caveats before you build a plan on those numbers. First, posting data describes advertised roles, not causation — it does not prove that learning SQL will raise your salary by 41%. Second, the underlying skills matter on their own: across 23 countries, Hanushek and colleagues found in Returns to Skills around the World that a one-standard-deviation increase in numeracy was associated with an 18% wage increase on average among prime-age workers. Stacking weak skills stacks nothing.
The MitHub Stack Test
Most "my talent stack" posts are a list of tools with a slash between them. Here is the test MitHub uses to tell a real stack from a list. A combination of skills is a stack only if it passes all three.
1. The one-decision test. Can you name a single decision that requires all the skills at once? Not a project, a decision. "Should we call this lead now or enrich it first?" needs data literacy, automation knowledge and sales judgement in the same breath. If each skill is used on a different day, you have a list.
2. The no-handoff test. Does the combination remove a handoff that currently exists in a company? Value shows up where the baton is being dropped. A marketer who can query the database removes a ticket to the data team. An operator who can read an API response removes a ticket to engineering. Every removed handoff is measurable in days.
3. The one-artifact test. Can you point to one thing you built that could not have been built without every skill in the stack? If two of your three skills are invisible in the artifact, the market cannot see them, and it will not pay for them. This is where stacking meets proof of work.
The MitHub Stack Canvas
Three roles, not three skills. Fill it in for yourself.
| Layer | What it is | How you know you have it |
|---|---|---|
| Spike | The one skill you are genuinely strong at — the reason someone lets you near the problem | Someone has paid for it, or asked you to do it twice |
| Multipliers (1–2) | Distant skills that make the spike apply to more situations | They change what you can attempt, not just how fast you work |
| Binding skill | Almost always writing and explaining | A non-expert can read your summary and make a decision from it |
| The seam | The decision where all of them are used at once | You can name it in one sentence |
The binding skill is the one people skip and the one that pays. A stack that nobody can understand is a hobby. Being able to explain a system to the person who funds it is what moves you along MitHub's capability ladder from Doer to Director to Designer to Owner — the director vs doer distinction is really a distinction about which layer of the stack you operate from.
A worked example (hypothetical)
Imagine Valeria, a marketing coordinator at a services company. Her spike is campaign copy. On its own it competes with everyone else who writes campaign copy, and increasingly with a language model.
She adds two distant multipliers over eight months: enough SQL and spreadsheet modelling to pull and clean her own lead data, and enough workflow automation to move records between tools without asking IT. Her binding skill — writing a clear one-page summary — she already had.
Her seam: "Which of the leads we generated last quarter actually turned into revenue, and what should we stop paying for?" Answering that needs the data skill to trace it, the automation skill to keep it updated, the copy skill to know what the campaigns promised, and the writing skill to make the finance lead act on it. This is exactly the follow the money discipline: start at the payment and trace each sale backwards to its source.
Valeria's title has not changed. Her position in the company has: she is now the person who answers a question the CEO cares about. That is what a stack buys.
This pattern is not theoretical at the top end. MitHub's pioneers have built AI voice campaigns that ran across 28 live branches of a multi-location lending business, including a 10-branch pilot with 13,159 AI calls. Work like that is not one skill — it is telephony, CRM data, workflow logic, prompt design, compliance awareness and branch-level reporting, held by a small number of people who can hold all of it in one head.
What is worth stacking right now
Stacks are specific to a market, so treat this as a method rather than a shopping list. The pattern that pays in the AI economy is: one domain you understand deeply + one systems skill + one data skill + writing.
- Domain — lending, clinics, logistics, legal intake, tax. Knowing where the money actually leaks in a specific industry is the least copyable layer.
- Systems — workflow automation, APIs, how an agent is wired to a tool. See how to learn AI automation.
- Data — pulling it, cleaning it, knowing when a number is lying to you.
- Writing — the binding layer.
For a fuller breakdown of the technical side of that pattern, see GTM engineer skills and the skills that make you more valuable.
The honest limits
Stacking is usually a second move. Burning Glass found only 16% of highly hybridised jobs were entry-level, against 58% of jobs overall. If you have no spike yet, stacking is a way of avoiding the harder work of becoming good at one thing.
Breadth without a spike is fragile. Four skills at 40% do not combine into one skill at 160%. They combine into someone who needs supervision on four fronts.
Skills decay at different speeds. Tool skills decay fastest, domain knowledge slowest, judgement slowest of all. Weight your stack accordingly, and see how skills increase income for the difference between a skill that raises your rate and a skill that just adds a line to your CV.
An unproven stack is worth nothing. The market cannot audit what you claim to know. It can audit what you built.
Build your stack in 90 days
- Days 1–7. Name your spike honestly. What have people paid you for, or asked you to repeat? If you cannot name one, work through how to find what you're good at before anything else.
- Days 8–14. Find the seam. Sit with the decision your team keeps getting wrong, or making too slowly. Write it as one question.
- Days 15–60. Add one multiplier — the one that seam needs. One, not three. Learn it against a real problem, not a course.
- Days 61–80. Build the one artifact that requires the whole stack. A working system, not a deck.
- Days 81–90. Write it up as situation, path, result and evidence, and put it somewhere a stranger can read it. That write-up is the stack made visible; how to write a case study for your portfolio covers the format.
Then repeat with the next multiplier. That is the loop MitHub teaches in the Faculty of Revenue Reverse Engineering: learn, build, create value, prove, earn, improve.
Key takeaways
- Skill stacking is combining a small number of distant skills so the combination is rare — Adams called it a talent stack and described it as a system you build.
- Hybrid roles that mix skill sets grow faster, pay more and are harder to automate, according to Burning Glass posting data; AI skills in particular command a premium and mostly sit outside IT.
- A stack is real only if the skills meet in one decision, remove a handoff, and show up in one artifact.
- Structure it as spike, multipliers, binding skill, and a named seam.
- Build the spike first. Hybrid roles are rarely entry-level, and an unproven stack has no market value.