TL;DR. A barbell is a deliberate mix of two extremes: a large, very safe position on one side and a small, high-risk, high-upside position on the other, with little in the medium-risk middle. In a career, the safe side is income, a cash buffer and durable skills; the bold side is a set of capped experiments. Of U.S. private establishments that opened in the year ending March 2021, 47.2 percent were still operating in March 2026 (BLS Business Employment Dynamics). In the Federal Reserve’s 2025 household survey, 58 percent of self-employed adults said their income varied month to month, against 28 percent of employees. Founders who start while keeping a day job had a 33.3 percent lower exit probability after going full time in one U.S. study, as summarised by later researchers. The CEO move: set a hard cap on what any single bet can cost you before you start it. The student move: treat each bold bet as a paid lesson, built to produce evidence cheaply.
This guide is a concrete spoke under The Antifragile Individual, which introduced the barbell as one habit among several. It builds on optionality as a career strategy, which explains why keeping doors open has value, and it borrows the stopping rules from kill criteria. Here the focus is sizing: what the bold end can cost and how to spot the middle.
This article reports research findings and offers a planning framework; it is general information, not personalised financial, legal or career advice.
What is a barbell strategy, in plain terms?
The term comes from finance and was popularised by Nassim Nicholas Taleb, whose book Antifragile: Things That Gain from Disorder was first published in hardcover on 27 November 2012, according to its publisher, Penguin Random House. The cleanest definition in Taleb’s own technical writing is in a 2014 paper with Donald Geman and Hélyette Geman, “Tail Risk Constraints and Maximum Entropy”. The authors define the barbell as “the mixing of two extreme properties in a portfolio”, maximal conservatism for one fraction and maximal or high risk for the rest. They also show that when an investor constrains only the left tail (the size of the worst losses) and stays honest about not knowing the rest of the distribution, the barbell “emerges naturally” as the best-shaped portfolio.
If someone holds 80 percent of a portfolio in securities that are treated as riskless, “the risk of losing more than 20% is zero independent from all possible models of returns”. The protection comes from construction, not forecasting: the worst case is fixed by how much is placed on the risky side.
Popular summaries often attach a 90 percent safe, 10 percent speculative split to the barbell. That split does not appear in the Geman, Geman and Taleb paper, which uses a general fraction and an 80 percent illustration. The worksheet below asks for a cap instead.
For an individual:
- The safe end keeps you solvent and employable if every experiment fails.
- The bold end is a few bets with open upside and closed downside: a side product, a skill tried in public, a pilot project at work.
- The middle is positions that feel moderate but can lose far more than they can win.
Why not simply aim for medium risk?
Because medium risk, in a career, usually means uncapped downside with capped upside. Three patterns recur.
The first is the all-in jump without a test: leaving a job for a venture that has never met a paying customer. The downside includes the income given up and the time out of the labour market, not only the money spent.
The second is the half-commitment that never resolves: an idea kept alive for years at low intensity, never tested hard enough to succeed or fail.
The third is concentration disguised as stability: one employer, one client or one narrow skill. Our guide to the half-life of skills shows why a skill valuable this year is not automatically a long-lived asset.
In the Geman, Geman and Taleb construction, the investor controls the worst-case loss directly. A medium-risk position lets every part of the portfolio lose something, so the total worst case is harder to know and to cap.
What do the base rates say about the bold end?
Bold bets fail often, and the distribution of outcomes is skewed. Both facts are measured in public data.
| Measure | Figure as reported | Period and population | Source |
|---|---|---|---|
| Establishments still operating 1 year after opening | 78.9% | Opened in year to March 2025, observed March 2026; U.S. total private | BLS Business Employment Dynamics, Table 7 |
| Establishments still operating 5 years after opening | 47.2% | Opened in year to March 2021, observed March 2026 | BLS Business Employment Dynamics, Table 7 |
| Establishments still operating 10 years after opening | 34.2% | Opened in year to March 2016, observed March 2026 | BLS Business Employment Dynamics, Table 7 |
| Venture-backed startups terminated at a loss | About 55% | First early-stage round 1985-2009, all U.S. VC investments | Kerr, Nanda and Rhodes-Kropf (2014) |
| Venture-backed startups returning more than 5 times investment | 6%, accounting for about 50% of gross return | Same | Kerr, Nanda and Rhodes-Kropf (2014) |
| Self-employed adults whose income varied month to month | 58% (employees: 28%) | U.S. adults, 2025 | Federal Reserve, Economic Well-Being of U.S. Households in 2025 |
| Self-employed adults who struggled to pay bills because income varied | 22% (employees: 10%) | U.S. adults, prior 12 months, 2025 | Federal Reserve, Economic Well-Being of U.S. Households in 2025 |
Sources: BLS Business Employment Dynamics, Table 7, survival of private sector establishments by opening year (data to March 2026); Kerr, Nanda and Rhodes-Kropf, NBER Working Paper 20358; Federal Reserve Board, Report on the Economic Well-Being of U.S. Households in 2025 (May 2026). Establishment survival counts business locations, not founders, and includes all kinds of new establishments.
Failure is the normal outcome of a bold bet: even among venture-backed startups chosen by professionals, more than half lost money. The winners are rare and large. Kerr and co-authors read this skew as evidence that even knowledgeable investors cannot tell in advance which bets will pay, the condition under which a barbell beats prediction.
Self-employed adults were about twice as likely as employees to report variable income (58 percent against 28 percent) and to report struggling with bills because of it (22 percent against 10 percent). The Fed notes the association remained after controlling for income. That is the case for keeping a salaried base while a bet is unproven.
Exit rates by establishment age across all cohorts are tabulated in the kill criteria guide linked above; this article uses three single cohorts as inputs.
What does the research on keeping your day job show?
Researchers call starting a business while keeping a wage job “hybrid entrepreneurship”. It is the closest real-world analogue of a career barbell.
It is common. Folta, Delmar and Wennberg (2010, Management Science) followed men aged 25 to 50 who started a job in Sweden’s high-technology manufacturing and knowledge-intensive service sectors in 1994. By 2001, 3.23 percent of the sample were self-employed and 2.55 percent were hybrids, so hybrids made up “over 44%” of all self-employment activity (2.55 divided by 5.78 is 44.1 percent, CEOtudent calculation). The authors describe small-scale hybrid entry as “a real option to invest heavily if early returns are promising and to exit if they are not”.
It is a path, not just a hobby. In the same study, being a hybrid was associated with a far higher likelihood of moving into full self-employment than remaining a plain wage worker. The coefficient of 3.64 implies an odds ratio that the authors describe as “over thirty-eight times” (the exponential of 3.64 is 38.1). Intensity mattered: hybrids whose side income was 20 percent or more of their salary were 143 times more likely to transition than hybrids whose side income was below 5 percent of salary, while those between 5 and 20 percent were twenty-eight times more likely.
It is associated with better survival after the jump, with a caveat. Raffiee and Feng (2014, Academy of Management Journal) report in their abstract that hybrid entrepreneurs who later quit their day job “have much higher rates of survival” than people who enter full-time self-employment directly from paid work, and that the advantage is stronger for people with prior entrepreneurial experience, which the authors read as a learning effect. The full text was not accessible for this article. Two later summaries put a number on it: Kritskaya Lindelid and Nair (2024) describe the U.S. study as examining 2,198 self-employment stints and finding “a 33.3% decrease in exit probability”, and the University of Wisconsin-Madison alumni magazine reported that founders who waited to quit were “33 percent less likely to fail”. Both are summaries of the paper, not the paper itself, and are labelled that way here.
The evidence is not unanimous. Kritskaya Lindelid and Nair (2024) studied a matched sample of 9,550 Norwegian employees who became majority owners of incorporated firms between 2005 and 2016, half of whom used staged entry. They found staged entry “was associated with a shorter first stint” of self-employment, an effect that became insignificant for the second stint, and that staged entrants were more likely to return to self-employment in the same firm. For a barbell, a shorter failed stint may simply be the capped loss the structure was designed to produce.
It is the pattern among millions of workers. The BLS Current Population Survey counted 8.8 million multiple jobholders in 2025, 5.4 percent of all employed people, up from 8.4 million in 2024. Of them, 5.0 million held a full-time primary job and a part-time secondary job, 57.3 percent of the total (CEOtudent calculation). BLS notes 2025 figures are 11-month averages because October 2025 data were not collected. Multiple jobholding is not entrepreneurship, but it shows how many people already run a two-part working life.
How much can a bold bet cost? A CEOtudent downside calculation
The barbell’s core claim is that structure changes the worst case more than the average. The table combines the BLS five-year exit rate with the hybrid survival estimate and two labelled illustrative assumptions, expressed in months of income so that it works in any currency.
| Path | Probability the venture has closed within 5 years | Cost if it fails (months of income) | Expected loss (months of income) | Worst case (months of income) |
|---|---|---|---|---|
| A. Quit the job and start directly | 52.8% | 12.0 (illustrative assumption) | 6.3 | 12.0 |
| B. Start as a hybrid, quit only after evidence | 35.2% (rough scaling) | 12.0 (illustrative assumption) | 4.2 | 12.0 |
| C. Capped hybrid bet, killed before quitting | 52.8% (no survival benefit assumed) | 1.2 (illustrative assumption: cap of 10% of a year’s income) | 0.63 | 1.2 |
CEOtudent analysis of BLS Business Employment Dynamics Table 7 and the Raffiee and Feng (2014) estimate as summarised by Kritskaya Lindelid and Nair (2024). 52.8 percent is 100 minus the 47.2 percent five-year survival of the cohort opened in the year to March 2021. Path B scales it down by 33.3 percent (0.528 times 0.667 is 0.352), a rough illustration because the study measured founder exit, not establishment survival. The 12-month and 1.2-month costs are illustrative assumptions.
- Staging helps the average. Under these assumptions, starting as a hybrid and quitting only after evidence cuts the expected loss from about 6.3 to 4.2 months of income, a difference of about 2.1 months.
- Capping helps the worst case far more. Path C assumes no survival benefit and the same 52.8 percent failure rate as Path A, yet its expected loss is about 0.63 months and its worst case 1.2 months, one tenth of Path A. The gain comes entirely from the cap.
- The paths are sequential, not rival. A barbell runs C first, as often as needed, and moves to B only when a bet produces evidence such as side income that is a meaningful share of salary. Path A is the middle in disguise: most of its risk is the loss of the safe end.
Whatever the failure rate, the worst case of a capped bet is the cap. For the cash a full jump requires, see how much runway you need to quit your job.
Why should the bold side be designed as a learning bet?
Kerr, Nanda and Rhodes-Kropf argue that entrepreneurship is experimentation: success probabilities are “low, extremely skewed and unknowable until an investment is made”. Their worked example shows why a cheap test can turn a bad bet into a good one.
| Step | Figure in the example | Result |
|---|---|---|
| All-in bet | Cost 110; 1% chance of a 10,000 payoff | Expected value -10: not worth pursuing |
| Run a test first | Test has a 10% chance of a positive signal; after a positive signal the project has a 10% chance of working | Overall chance of success unchanged at 1% |
| After a positive signal | 10% chance of 10,000, minus the 110 cost | Expected value 890 |
| Value of the test | 10% chance of reaching a project worth 890 | Worth running if it costs less than 89 |
Source: Kerr, Nanda and Rhodes-Kropf (2014), NBER Working Paper 20358, figures as printed; recomputed in calc.py.
This is the student lens in numbers. The test does not raise the odds of success; it lets you abandon failing paths before paying their full cost. A bold bet is judged first by what it teaches per unit of money and time.
So each bet needs a question (“Will 20 people pay for this?”), the cheapest way to answer it, and a pre-set decision for each answer. See thinking in bets for judging decisions by process, and the explore-exploit guide for sizing an exploration budget.
The U.S. Census Bureau counted 501,609 seasonally adjusted business applications in October 2026 and projected that 27,867 new businesses with payroll tax liabilities would form from them within four quarters, about 5.6 per 100 applications (CEOtudent calculation). Many applications never intend to hire, so this is not a failure rate, but filing is the start of a test, not its result.
How strong does the safe side need to be?
The safe end is what makes boldness affordable, and many households lack it. In the Federal Reserve’s 2025 survey, 55 percent of adults said they had set aside emergency savings to cover three months of expenses, which means 45 percent had not. Thirty percent said they could not cover three months of expenses by any means, including borrowing or selling assets.
A bet cannot be sized until the safe side can absorb its worst case. For variable incomes the buffer needs to be larger, as set out in emergency fund math for the self-employed.
The safe side has four parts, and only one of them is money:
- Income floor: income that covers essential costs without help from any bet.
- Cash buffer: months of essential expenses held in cash.
- Long-half-life skills: abilities that keep value across tools and employers.
- Employability: a network and record that would find comparable work if the floor disappeared.
The Career Barbell Allocation Worksheet
This CEOtudent editorial framework is not a validated instrument and prescribes no numbers. Each row asks for a floor or cap set in advance.
| Side | Field | What to write | Rule of thumb in the framework | Evidence it draws on |
|---|---|---|---|---|
| Safe | Income floor | Which income covers essential costs if every bet fails? | Keep at least one source that does not depend on any bet | SHED 2025 income-variability gap (58% vs 28%) |
| Safe | Cash buffer | Months of essential expenses held in cash | Build before sizing any bet; size larger if income is variable | SHED 2025: 55% have a three-month fund |
| Safe | Long-half-life skills | Two or three skills that stay valuable across tools | Spend part of the safe-side time maintaining them | Skill half-life guide |
| Safe | Re-employment path | Who would hire you within three months, and for what? | Name real people and roles, not categories | Optionality scorecard |
| Bold | Bet and question | One sentence: what the bet is and what it will prove | Every bet answers one testable question | Kerr et al. experiment logic |
| Bold | Money cap | Maximum cash the bet can consume, ever | Set as a share of annual income the safe side can absorb without debt | Geman, Geman and Taleb: cap the left tail |
| Bold | Time cap | Hours per week and an end date for the first test | Write the hours and the date before starting | Kill criteria guide |
| Bold | Scale signal | What evidence would justify more time or quitting | Tie it to income, users or paid work, not enthusiasm | Folta et al.: intensity predicted transition |
| Bold | Kill signal | What result ends the bet at the cap | Pre-commit; do not raise the cap mid-bet | BED survival base rates |
| Middle | Exposure check | List any position whose worst case you cannot state | Shrink it, cap it, or move it to one side | Barbell definition |
CEOtudent editorial framework built on the sources named in the right-hand column.
A worked illustration, every input an illustrative assumption: a salaried analyst works 45 hours a week and gives a side product 5 hours, about 11.1 percent of working time. The money cap is 10 percent of a year’s income (Path C’s 1.2 months), the first test ends after 90 days, and the scale signal is side income reaching 5 percent of salary, the lower bound of Folta and co-authors’ higher-intensity band. These are not recommendations.
The middle-risk trap test
Use this CEOtudent editorial framework once a quarter. Answer each question yes or no.
- Is there any single position (employer, client, skill, investment, project) whose loss would remove more than half of your income, with no plan to replace it?
- Is any bold bet running without a written money cap?
- Is any bold bet running without an end date for its current test?
- Have you raised the cap on a bet after it missed its last target?
- Is a bet more than a year old without having answered the question it was started to answer?
- Would leaving your job today depend on a bet that has not yet produced income or paid users?
- Are you running more than three bold bets at once, so that none gets enough time to produce evidence?
- Has your safe-side skill maintenance stopped because the bold side absorbs all spare time?
Reading the result. Zero or one “yes” suggests a working barbell. Two or three suggests drift; fix caps and end dates first. Four or more: rebuild the safe side before adding any bet. The thresholds are editorial judgement.
How do the CEO lens and the student lens work together?
The CEO lens is allocation: decide what the portfolio can lose and still survive, then fund bets inside that limit. The student lens is content: Folta and co-authors frame hybrid entry as a way to learn about a venture’s upside and “their own fit”, and Raffiee and Feng attribute the survival advantage to learning. Without caps, the student pays tuition after the lesson is learned; without questions, the CEO funds bets that cannot be judged.
What this does not mean
- It does not mean never quitting. Hybrids moved into self-employment far more often than wage workers; the barbell changes when, not whether.
- It does not mean the hybrid effect is settled. The U.S. survival advantage comes from one study whose full text was not opened here; the Norwegian study found staged entrants had shorter first stints. People who choose hybrid entry differ from those who jump, so selection may explain part of the gap.
- It does not mean the safe side is permanent. A safe job, client or skill can become concentrated risk.
- It does not resolve employment contracts. Many employment agreements restrict outside work; Folta and co-authors point to non-compete covenants as a factor in hybrid activity.
- It does not replace financial advice. Taxes, debts, dependants and health cover are not modelled here.
Frequently asked questions
What is the barbell strategy for a career?
A very safe base (income, cash buffer, durable skills) plus a few capped, high-upside bets, with little in positions whose worst case cannot be stated.
Is the barbell always 90 percent safe and 10 percent risky?
No. Geman, Geman and Taleb use a general fraction and an 80 percent illustration; the 90/10 split was not found there.
Is it better to keep your day job when starting a business?
For survival after quitting, U.S. evidence says yes (a 33.3 percent lower exit probability, as summarised by later researchers); a Norwegian study found shorter first stints, so the evidence is mixed.
How many people already run a side job?
The BLS counted 8.8 million multiple jobholders in the United States in 2025, 5.4 percent of all employed people, an 11-month average.
How big should a bold bet be?
Big enough to answer its question, and no bigger than a loss the safe side can absorb without debt. The worksheet asks for a money cap and a time cap in writing before the bet starts.
When should a bet move from the bold side to the main career?
When it produces evidence tied to income, users or paid work. In Folta and co-authors’ data, hybrids earning 20 percent or more of salary from the side venture were 143 times more likely to transition than those earning under 5 percent.
Sources
- Geman, D., Geman, H. and Taleb, N. N. (2014). Tail Risk Constraints and Maximum Entropy. arXiv:1412.7647 (q-fin.RM), version 1, 24 December 2014.
- Penguin Random House. Antifragile: Things That Gain from Disorder by Nassim Nicholas Taleb, publisher product page (format and publication dates).
- U.S. Bureau of Labor Statistics. Business Employment Dynamics, Table 7: Survival of private sector establishments by opening year, total private, data to March 2026.
- Kerr, W. R., Nanda, R. and Rhodes-Kropf, M. (2014). Entrepreneurship as Experimentation. NBER Working Paper 20358, July 2014; published in the Journal of Economic Perspectives 28(3), 25-48 (publication details via OpenAlex).
- Board of Governors of the Federal Reserve System (2026). Economic Well-Being of U.S. Households in 2025. May 2026.
- Folta, T. B., Delmar, F. and Wennberg, K. (2010). Hybrid Entrepreneurship. Management Science 56(2); IFN Working Paper No. 825 (author version).
- Raffiee, J. and Feng, J. (2014). “Should I Quit My Day Job? A Hybrid Path to Entrepreneurship”. Academy of Management Journal 57(4), 936-963 (abstract only, via OpenAlex).
- Kritskaya Lindelid, L. and Nair, S. (2024). Trading wage jobs for dreams: the interplay between entry modes into self-employment and the duration of subsequent self-employment stints. International Journal of Entrepreneurial Behavior and Research 30(11), 120-139 (CC BY 4.0).
- On Wisconsin Magazine, University of Wisconsin-Madison alumni magazine. Wait for It (summary of the Raffiee and Feng study).
- U.S. Bureau of Labor Statistics. Current Population Survey, Table 36: Multiple jobholders by selected characteristics, 2024 and 2025 annual averages (last modified 20 February 2026).
- U.S. Census Bureau. Business Formation Statistics, release CB26-184, 12 November 2026 (October 2026 data).
Note on methods: the downside calculation, the Career Barbell Allocation Worksheet and the middle-risk trap test are CEOtudent analyses or CEOtudent editorial frameworks; the 12-month, 1.2-month, 45-hour and 5-hour inputs are illustrative assumptions, and derived figures are recomputed in a script. The Raffiee and Feng survival figure comes from two labelled summaries because the full paper could not be opened. Deliberately not used: the 90/10 barbell split (not found in the primary text consulted), a “6,300 entrepreneurs over 14 years” sample description seen only in press coverage, and cross-cohort survival-by-age tables already covered in our kill criteria guide.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:















