In the first quarter of 2026, a series of AI-related events ignited vigorous debates about the future of software. Autonomous workflow agents, popularized through tools such as Openclaw (formerly known as Clawdbot or Moltbot) showed the ability to generate production-ready bespoke software from user’s intent alone. At the same time, the stock market reacted strongly to the accelerating capabilities of agentic coding systems. Some SaaS company stocks were in sharp declines, fueling a broader debate about whether SaaS companies will be disrupted as their controls of customer workflows may be taken away by AI agents.
The rapid proliferation of agentic coding, combined with growing skepticism about the durability of SaaS software moats, should prompt a fundamental rethinking of how companies approach intellectual property (IP).
Agentic coding systems bring new focus to the idea-expression dichotomy in copyright law. Expression is the protectable component in copyright law, not ideas. But high-level product concepts (such as linking application behavior to user events, dynamically generating media outputs, or coordinating multiple software components) are ideas. And so, if they are protectable at all, it is through patent law.
By contrast, the specific source code that implements those ideas is an expressive work traditionally protected by copyright. In agentic coding, however, that expression may be generated largely or even entirely by an autonomous system, with no meaningful human authorship of the resulting code.
For example, an engineer may provide a short, natural-language description as an AI prompt that describes a consumer-facing application combining certain user interactions, business logics, and media outputs. At present (or in the near future), an agentic system could produce a functioning application end-to-end based on the engineer’s description alone and without any human input in architectural guidance, implementation details, or iterative debugging. This type of process is widely described as “vibe coding,” in which the human describes their intent for the functionality rather than actually writing the code for execution.
The surge in popularity of vibe coding reveals a broader and inevitable trend: the expressive layer of software is increasingly generated by a machine and not authored by a human. As agentic coding matures, companies will increasingly encounter scenarios where the expressive components of their products originate from AI systems rather than human developers, while humans retain some of the idea creation. This shift alters both how IP is created and how its value should be assessed.
As vibe coding becomes commonplace, strong copyright protection may no longer be an automatic byproduct of software development. Instead, copyrightable code creation may become a deliberate business choice that can be weighed against cost, speed, and strategic value.
In the pre-AI era, copyright arose naturally from manual code writing. Developers necessarily authored source code. This made the marginal cost of producing copyrightable code effectively zero.
Agentic coding has changed this calculus. For reasons such as rapid iteration and reduced labor costs, companies may choose to rely heavily on AI systems for code generation. Under current U.S. Copyright Office policies, software produced entirely by non-human systems without meaningful human authorship may not qualify for copyright protection. Obtaining copyright may therefore require additional human involvement, introducing an explicit cost.
As such, the marginal cost of protectable copyright in software produced by agentic code may no longer be zero.
This shift forces companies to consider new IP tradeoffs. Key questions include:
When specific code can be easily reimplemented through different code expression by agentic systems to avoid copyrighted materials, copyright may offer diminishing strategic value compared to the pre-AI era. In such cases, companies may prioritize other IP protections, including patents and trade secrets, or non-IP strategies such as speed-to-market advantages.
Despite these cost considerations, the benefits of copyright or IP protection remain significant and are often underappreciated in the software world, even if a company does not enforce its IP through litigation.
Historically, copyright has served as a foundational mechanism for controlling downstream use of software. Open-source licenses, end user license agreements, and terms of service often rely on underlying copyright rights for enforceability, as an alternative to the contractual rights. This is because, in many real-world scenarios, there can be no contractual relationship between the rights holder and a downstream user. For example, in cases involving redistribution, integration of open-source components without awareness of the governing license, or unauthorized access to software, end users may not have agreed to any of the license or contract terms associated with the software. In those situations, copyright provides an alternative enforcement pathway when parties are not in a contractual relationship. While businesses often conceptualize downstream control as a contractual issue, copyright has long been the quiet enabler that makes those controls viable.
Hence, the possession of enforceable IP is still important to enable many common business strategies in the software world.
If agentic coding weakens copyright as a practical enforcement tool, two other forms of IP protection may assume greater importance: patents and trade secrets.
Assuming copyrightable expression continues to diminish as agentic AI use for code generation grows, patents may become the primary anchor for a company’s protectable IP.
Patents have their own challenges but also significant upsides. A human needs to conceive the functionality they seek to protect. Patents must also meet eligibility requirements and often involve a higher cost structure to procure. If successful, however, patent protection is powerful. It includes the right to exclude others from making, using, selling, offering to sell, or importing the patented functionality, even if independently developed.
Like copyrights, patents can be licensed and enforced in the marketplace. Patents may assume new roles that fill the potential voids resulting from weakening copyright protection. Similar to how copyright controls downstream software use, patents can support licensing structures, enable enforcement where copyright is unavailable, and help preserve community norms in open or hybrid ecosystems by discouraging uses that violate software licenses.
Patents also help undergird the substantial investments being made in AI technology today. Filing and receiving patents may serve as legally enforceable proof that a company has outexecuted its competitors by being the first to bring new features to market. Such patents can target methods, system architectures, workflows, and algorithms that are part of the engineers’ ideas, even when some implementations have been completed by AI agents.
In addition to patents, trade secret protection provides another path for companies to protect code generated by agentic AI deemed to be valuable. Trade secrets, unlike copyright and patent, require no formal registration process. Rather, a trade secret is confidential information that provides competitive advantage because it is not generally known or readily ascertainable. The information must have economic value and be subject to reasonable efforts to maintain its secrecy. Importantly, unlike copyright and patent, trade secrets need not be human-created. Further, unlike copyright and patent, there is no time limitation on trade secret protection. It could remain for decades if handled properly.
Code generated through agentic AI may qualify as a trade secret if it is kept confidential and provides economic value to the company. Most companies must wrestle with one challenge, however: What are the reasonable measures to protect the code generated through agentic AI? Using public AI models to help generate such code may result in inadvertently exposing the trade secret. In contrast, company-specific code (as well as personalization, such as memory.md and system prompts) can be kept as trade secrets and may provide competitive edge over other companies.
Companies should understand, however, that trade secrets are not solutions to every situation. Trade secrets do not protect against independent development or reverse engineering. AI models may be able to quickly reverse engineer code initially thought to be protected by trade secrets and may simply re-implement the same solution using different code. Coding agent personalization currently tends to focus on increasing productivity and accuracy, not necessarily generating trade secrets that are protectable in the long term. Trade secrets, even if kept secret, tend to have a shorter actual advantage in the software world compared to the secret ingredients of, say, Coca-Cola, because software often can be independently developed to achieve the same functionality. If a company is considering a long-term differentiation, patents often provide better solutions than trade secrets.
Agentic coding is poised to exponentially increase the volume of software code being created, driving down the cost of producing functional software. In that environment, many bespoke features or narrowly tailored workflows may not justify the time and expense required to secure traditional IP protection. At the same time, the erosion of automatic IP protection under copyright will force companies to be far more deliberate. Decisions about whether to pursue copyright, patents, or alternative forms of protection will increasingly be an explicit and ROI-driven business choice rather than a passive consequence of building software.
In an era of vibe coding and autonomous software agents, IP is no longer incidental. It is strategic.