What Have We Learned? Who Owns What You Know? Non-Competes, Stay-or-Pay, and the Asset That Walks Out the Door. Literally.

In September 1957, eight young scientists resigned from Shockley Semiconductor Laboratory in Mountain View, California. They had grown tired of William Shockley's management style (he was a Nobel laureate and, by most accounts, a very difficult boss) and wanted to build silicon transistors their own way. Shockley called them the "traitorous eight." With backing from Fairchild Camera and Instrument, they founded Fairchild Semiconductor, which in turn spawned dozens of spin-offs, including Intel. Much of what we now call Silicon Valley traces its family tree to that walkout.

Shockley could not have used a non-compete to keep them out of the business, because California would not enforce one. Since 1872, California law has voided contracts that restrain a person from "engaging in a lawful profession, trade, or business." Had the eight been working in Massachusetts where courts routinely enforced non-compete agreements, the story might have gone differently. Stanford law professor Ronald Gilson made exactly that argument in 1999: engineers moving freely between firms spread knowledge faster in Silicon Valley than along Boston's Route 128, and the difference in non-compete law helps explain why one region pulled ahead. (Economists still debate how much of Silicon Valley’s advantage can be attributed to non-compete law rather than to other factors, including  venture capital, the influence of Stanford University, and defense spending, so it is probably best understood as one important  ingredient rather than the whole recipe.)

This month we look at the part of the employer-worker compact that governs what happens when the relationship ends. Every company invests in what its people know, and every worker carries some of that knowledge with them when they leave. What belongs to the employer, what belongs to the worker, and how far a company may go to protect its interests are among the oldest questions in employment law.

An old argument

The earliest recorded non-compete case is older than the printing press itself. In Dyer's Case (1414), an English master sued a former apprentice, John Dyer, who had promised not to practice his trade in the same town for six months. The judge was so offended by the agreement that he declared (in the court's Law French) that if the master were present he would go to prison until he paid a fine to the King. The case reflected the early common law’s deep suspicion of agreements that prevented people from practicing their trades—a concern that was especially salient in a labor-scarce economy still shaped by the aftermath of the Black Death.

Over the next three centuries, however, the economy grew more complex and the courts’ approach evolved. In Mitchel v. Reynolds, an English court distinguished sweeping  restraints on trade from narrower restrictions that could be justified as reasonable. The case involved a baker who had agreed not to compete within a particular area as part of a business transaction, and the court upheld the restraint because it was limited and connected to a legitimate commercial interest. That reasoning became the foundation for the “reasonableness” approach that later developed in England and the United States.

By the nineteenth century, American courts were generally asking whether a restraint went no further than necessary to protect a legitimate business interest while also considering its impact on the worker and the public. For much of the next two centuries, that case-by-case approach became the prevailing model across most states.

The man with a code

This month's “person with a mission” never set out to help workers change jobs. David Dudley Field was a New York lawyer who spent much of his career on a single project: replacing the patchwork of judge-made common law related to civil law and procedure with written codes that lawyers and ordinary people alike could understand. His Code of Civil Procedure, adopted by New York in 1848, reshaped how American courts operate. Ironically, his proposed Civil Code, which covered contracts, property, and much else, never became law in New York; the state's legal establishment preferred its common law.

California took it instead. The new state was trying to merge Spanish, Mexican, and English legal traditions into a single system, and in 1872 its legislature adopted Field's draft. Section 833 of Field's proposed code, which voided contracts restraining anyone from exercising a lawful profession, was adopted verbatim and survives today as Business and Professions Code section 16600. Gilson called it "a serendipitous result" of the codification movement.

Field's story is a useful reminder that history is not always shaped by visionary individuals deliberately setting out to change the world. A rule that may have done more than almost any other to shape the modern knowledge economy did not emerge from a campaign for worker mobility. It began as a provision in a codification project that Field's own state rejected, was later adopted by a fast-moving frontier legislature, and only much later came to be recognized as an economic advantage.

From engineers to sandwich makers

Non-competes were designed to protect trade secrets and client relationships, the kind of knowledge a senior engineer or salesperson inherently has, by dint of doing the work. Over the past few decades they spread far beyond that group. A national survey by economists Evan Starr, J.J. Prescott, and Norman Bishara found that about 18% of American workers were bound by a non-compete and 38% had signed one at some point. Only about 10% had negotiated the terms, and roughly a third first saw the agreement after they had already accepted the job.

The case that made the practice visible to the public involved, of all things, low-tech  sandwiches! In 2016, attorneys general in New York and Illinois challenged Jimmy John's for requiring shop employees to agree that for two years after leaving they would not work within two miles of any Jimmy John's store for a business earning more than 10% of its revenue from sandwiches. The company agreed to stop and paid $100,000 to settle the Illinois suit. It is hard to imagine that a sandwich assembler leaves his job with an industry-stopping trade secret.

The evidence on what non-competes  do to workers is fairly consistent. When Oregon banned non-competes for hourly workers in 2008, wages rose 2 to 3 percent on average, with larger gains for women, and the job-to-job mobility increased.. But employers raise a legitimate concern on the other side of the compact: companies may be less willing to invest in training or share valuable knowledge if employees can immediately take that investment to a competitor. 

But what is good for an individual employer is not necessarily what is best for business—or the economy—as a whole. When workers move, they take skills, ideas, and experience with them, spreading knowledge across firms, helping talent find its most productive use, and sometimes creating new businesses altogether. The tension, then, is between protecting a company’s legitimate investment and preserving the mobility that helps a dynamic economy work. The policy question is whether restricting a person’s ability to work is the right tool for protecting that narrower employer interest.

Washington swings, the states split

In April 2024 the Federal Trade Commission under Chair Lina Khan issued a rule banning most non-competes nationwide. The agency estimated that about 30 million workers, nearly one in five, were covered by them, and projected higher earnings, more than 8,500 new businesses a year, and up to $194 billion in lower health care costs over a decade. A federal court in Texas blocked the rule in August 2024, and in September 2025 the FTC under Chairman Andrew Ferguson dropped its appeal, leaving the nationwide rule unenforceable

The agency did not abandon the issue. It shifted from a universal rule to a case-by-case enforcement against agreements it considers anticompetitive. In November 2025 the FTC finalized a consent order requiring Gateway Services, the country's largest pet cremation company, to stop enforcing nationwide non-competes against the nearly 1,800 workers. In April 2026 it reached a similar agreement with Rollins, the parent of Orkin, covering more than 18,000 pest-control technicians and customer-service workers who had been barred for two years from working within 75 miles of any of Rollins' 700-plus locations, without extra pay for signing. Readers of last month's post will recognize the pattern: one company at a time, while the national rule stays unsettled. The FTC finalized that order in June.

Readers of our blog last month may recognize a pattern: rather than rewriting the rules for everyone at once, the federal government is proceeding one employer at a time. The nationwide ban is gone, but the argument over how far employers may restrict workers after they leave is very much alive. 

That leaves the states to draw their own lines - and those lines now look very different. California, Minnesota, North Dakota, Oklahoma broadly prohibit employee non-competes, while states including Montana and Wyoming impose substantial restrictions with important exceptions. Virginia expanded its prohibition in 2025 to include employees  entitled to overtime under federal law, regardless of earnings, and a growing number of states have created special provisions for physicians and other healthcare workers. Florida has moved in the opposite direction (no surprise). Its CHOICE Act, in effect since July 2025, allows non-competes of up to four years for workers earning more than twice the annual mean wage in the relevant county.. For qualifying agreements, the law gives employers unusually strong enforcement tools, including a presumption that a worker had access to confidential information or customer relationships when the worker acknowledged that access in writing. The result is a remarkable geographic divide. A worker in Miami and a worker in San Francisco can perform essentially the same job, possess the same skills, and leave with the same accumulated experience - yet face fundamentally different rules about what they may do with that expertise next. 

Stay-or-pay: the non-compete's newer-ish cousin

As non-competes drew scrutiny, a different tool spread: the training repayment agreement, or "stay-or-pay" clause. The worker receives training (a nurse residency, a commercial driver's license, a sales certification) and agrees to repay a predetermined amount if they leave before a specified date. The Consumer Financial Protection Bureau warned in 2023 that these arrangements can function much like non-competes: workers may remain in jobs they want to leave because quitting would trigger a substantial repayment obligation. 

California's AB 692, effective January 1, 2026, largely bans non-competes, with narrow exceptions for signing bonuses, relocation help, and tuition for credentials that are transferable to other employers, and gives workers the right to sue for $5,000 or actual damages. New York's Trapped at Work Act, signed in December 2025, takes a similar approach, though its effective date was later pushed back a year.

From a human capital perspective, the stay-or-pay clause is revealing. Companies describe training as an investment in their people. Accounting treats it as an expense, recorded in the year it is spent, and it never appears on the balance sheet as an asset. A training repayment agreement is one way to turn that invisible investment into something that can be recorded: a debt owed by the worker. In effect, the company books its human capital investment as the employee's liability. If training really is an investment, the more durable way to protect it is to make staying worthwhile, through career paths, pay progression, and managers people want to work for.

What other countries are doing

Other countries have taken the knowledge question seriously and answered it with price tags and time limits.

Germany allows post-employment non-competes for up to two years, but only if the employer pays the worker at least 50 percent of their total compensation for every month the restriction applies. An agreement without that payment is unenforceable. The logic is simple: if keeping someone off the market is worth that much to the company, the company should pay for it.

Ontario banned most non-competes in 2021, keeping exceptions for senior executives and for owners who sell a business. Employers there still protect confidential information and client relationships through non-solicitation and confidentiality agreements.

The United Kingdom, where approximately 5 million employees have non-compete clauses in their contracts, published a working paper in November 2025 considering options ranging from a three-month cap to an outright ban. The consultation closed in February 2026, we’ll see what emerges.

Each approach has costs. Employers in Germany pay for restrictions they may not have needed, and employers in Ontario rely more heavily on litigation over trade secrets. Their shared principle is that restricting a worker's future livelihood should either be rare or be paid for.

AI and the knowledge that no longer needs to leave

AI adds two new twists to the question of who owns what you know.

The first is visible inside in the AI industry itself. Most leading AI labs are based in California, where non-competes are generally unenforceable , so talent moves quickly and companies rely on other tools - trade secret laws, compensation, retention packages and aggressive recruiting - to protect and acquire scarce expertise. When xAI sued OpenAI for allegedly obtaining its trade secrets through former employees, a federal judge dismissed the claims against OpenAI in June 2026, emphasizng that passive receipt or mere possession of trade secrets is not enough, by itself, to constitute misappropriation. Big technology companies have also embraced the "reverse acquihire." In July 2025, Google agreed to pay about $2.4 billion to license certain Windsurf technology while  its CEO, co-founder, and a group of senior researchers, without acquiring the company itself. The transaction makes the point vividly: in a knowledge business, acquiring the people who hold critical expertise can sometime matter as mich as acquiring the company that employs them. (Hint: that’s why HR due diligence of a professional services firm is so critical.) 

The second twist reaches far beyond the AI industry. For centuries, the employer's problem was that knowledge left with the worker. AI makes it increasingly possible for at least some of that knowledge to remain behind. Companies are asking employees to document workflows, write prompts, codify their decision rules, create prompts, train models, and review the outputs of AI systems that can reproduce parts of their work. Some of this is simply better knowledge management. But some of it is also the conversion of individual expertise into an organization asset that can continue producing value after the employee leaves. Employees have noticed, and some are responding by holding back what they know. That changes the incentives around knowledge sharing. Emerging research suggests that when workers perceive AI as a threat to their professional relevance or job security, some become more likely to withhold knowledge rather than contribute it freely. That is precisely the opposite behavior a learning organization needs.

The employer-worker compact has not caught up. Non-compete law asks what a worker may take with them. AI raises the mirror-image question: what knowledge may the company keep?

Few legal rules, and few company policies, address whether workers should have any claim on the value created when their expertise in encoded into software, how that contribution should be recognized, or what obligations employers have when knowledge capture also makes the worker less necessary. 

What is knowledge worth, and to whom?

A century ago, IRC4HR's founders argued that workers were assets rather than costs. Today, we are still debating this fundamental premise – what is the asset that matters most? What people know and can do, especially in a service driven economy, is still ignored in terms of its material impact and sorely missing from disclosures on financial statements. The SEC has required public companies to describe their human capital since 2020, but the disclosures remain largely narrative and hard to compare. Without a way to measure knowledge, companies protect it through contracts: non-competes, stay-or-pay clauses, and confidentiality agreements. These tools treat knowledge as something to be kept from leaving rather than something to be built, measured, and renewed.

That creates a measurement problem. If a company tracks training spend but not the capability it produces, it cannot tell whether a non-compete is protecting something valuable or simply suppressing turnover. If it counts headcount but not where critical knowledge sits, it cannot see the risk when a key person leaves or when a system replaces them. And if AI tools now hold part of the company's know-how, the usual measures of productivity, such as revenue per employee, will rise without telling leaders whether the company has become more capable or simply more dependent on what its former employees knew.

It is also a governance question. Boards and leadership teams should understand not only where critical knowledge resides, but how dependent the organization is on it, how that knowledge is protected, and whether those protections create more value than risk. They should be able to answer: How material are people and workforce capability to our business model and enterprise value? Where does our most important knowledge sit—in people, systems, or vendors? How concentrated is it? What would we lose if our most knowledgeable employees left? How many workers are bound by non-competes, repayment agreements, or other restrictions, and are those tools proportionate to the interests being protected? As employees help train AI systems, what value is being transferred from people to technology, how is that contribution recognized, and who governs the tradeoff? These are no longer just employment-law questions. They are questions of human capital, operational resilience, materiality, reputation, and enterprise value.

Your Sphere of Influence

What you can act on	Practical solutions
Inventory	Count how many employees have signed non-competes, non-solicits, and repayment agreements, by role and pay level. Many companies cannot produce this number without a manual review of contracts.
Fit for purpose	For each restrictive covenant, identify the specific interest it protects (trade secrets, client relationships, training investment). Where a narrower tool, such as a confidentiality or non-solicitation agreement, would do the job, use it.
Jurisdiction	Map where your employees and contractors work against the state and national rules (California, Florida, New York, Germany, Ontario take very different approaches). A single template agreement is now a compliance risk.
Stay-or-pay review	Review training repayment terms against California's AB 692 and prepare for New York's Trapped at Work Act which takes effect on December 19, 2026, even outside those states. Consider whether retention incentives would achieve the same goal.
Measurement	Track what training produces (capabilities, internal mobility, time to proficiency), not only what it costs. Identify roles where knowledge is concentrated in a few people and measure the business value and risk associated with that concentration
Knowledge capture and AI	Set clear, transparent terms for how employee expertise is captured in AI tools, how it is credited, what value the organization receives, and how contributors participate in or are recognized for that value. Explain the purpose before asking people to document their work.
Line of sight	Help people see how their knowledge contributes to the company's value, and create incentives for them to share and develop it. Knowledge sharing is less likely when employees believe doing so diminishes their own value or job security.
Governance	Give boards and senior leaders visibility into how material people, workforce capability, and critical knowledge are to the business model and enterprise value. Report on knowledge concentration, restrictive-covenant use, AI knowledge capture, and the risks created when essential capability resides in a small number of people, systems, or vendors.

Conclusion

For six centuries, the law has tried to balance an employer's interest in what it taught its workers against a worker's right to earn a living with what they know. The answer has swung from the hostility of Dyer's Case to the expansive protections now available to employers in states such as Florida and the pendulum is unlikely to stop moving. 

AI changes the stakes. Companies can increasingly capture, codify, and retain parts of their employees’ knowledge even after those employees leave. That makes the old question - what may a worker take with them? - only half of the problem. The other half is what the company can keep, how much value that knowledge creates, and what is owed to the people who helped create it.

The organizations that manage this well will do more than protect knowledge. They will measure it, renew it, govern it, and give people a reason to keep sharing it.

Next month

Who is watching the work? In 1914, Henry Ford's Sociological Department sent investigators into workers' homes to decide who qualified for the five-dollar day. Today, monitoring software tracks keystrokes, and AI tools score productivity in real time. We look at a century of workplace surveillance and what it has done to trust.

---

This is the October 2026 installment of "What Have We Learned?", a monthly series for HR leaders on the history of the employer-worker compact.