From Tools to Results: The Intellectual Property Challenge in the Age of Services as Software

By Arturo Covarrubias and Mariano Wood

For more than two decades, much of the digital economy was organized around a relatively simple premise: software was an assistance tool that made human work more efficient. Companies bought licenses, contracted SaaS platforms, integrated APIs, and measured technological success in terms of productivity. In this framework, the lawyer’s role was focused on reviewing license agreements, service level agreements (SLAs), code ownership, and data protection. That framework remains important, but today it is being radically challenged.

Generative artificial intelligence is displacing software, which is no longer merely a tool to help work but is beginning to become an operational layer that executes, in part or entirely, the work itself. This transition implies a shift in the value chain. Sequoia Capital’s thesis on the concept of “services as software” shows that the economic opportunity no longer lies solely in selling software to service providers, but in directly selling the output that was previously produced by a combination of professionals, processes, and tools.

In simple terms: the goal is no longer to sell a platform to close the books, but to deliver the books already closed. The aim is no longer to offer a tool to review contracts, but the completed review. Sequoia frames this as a shift in focus — from the means to the end.

This change is financial and technological, but also profoundly legal. When a company contracts a service executed by AI, the relevant questions are no longer exhausted by the license or the privacy policy. It is now essential to determine who designed the system, what data or content it was trained on, what degree of autonomy it has, what traceability exists in the process, and who is liable for potential third-party infringements in the training chain or in the final output.

It is clear that intellectual property is at the center of this transformation. AI generates texts, images, code, designs, and music — often based on pre-existing materials: literary works, databases, or trade secrets. Therefore, the IP discussion is not limited to whether an AI-generated work is protectable under copyright or capable of generating royalties. The prior legal question is: did the system that generated those results have the right to operate with the materials it used, and what are the consequences for those who exploit those results?

In the Chilean market, the forthcoming Artificial Intelligence Law (Bulletin 16821-19, currently in legislative proceedings) must be read in this context. Its architecture, inspired by the risk-based model of the European AI Act, seeks to regulate the development and use of these technologies. However, the complex intersection between innovation and copyright has re-emerged forcefully in the debate over “data mining” (TDM). The proposed addition of Article 71 T to Law No. 17,336 seeks to authorize acts of reproduction, adaptation, distribution, or public communication of lawfully published works when carried out for the purposes of extraction, comparison, classification, or other statistical analysis of large volumes of data without prior authorization, provided they do not constitute “covert exploitation.”

This debate is not exclusive to any single jurisdiction: each country needs its own data mining rule to attract technology and foster local innovation. However, the problem lies in the breadth of a formula that risks becoming obsolete by the time it is published. Recent case law already compels us to distinguish between stages. In the United States, cases such as Bartz v. Anthropic or Thomson Reuters v. Ross Intelligence are defining when data use is “transformative” and when it becomes exploitation that unfairly competes with the original rights holder. In Europe, the dispute between GEMA and OpenAI has drawn a critical line: when a model memorizes and reproduces protected content, the issue ceases to be an invisible technical process and becomes legally relevant reproduction.

The lesson is clear: not every copy for AI should be prohibited, but neither should every copy for AI be free. Robust regulation must distinguish between non-expressive technical use and substitutive exploitation; between scientific research and commercial operation; between lawful access and illicit sources; between statistical analysis and outputs that reproduce or replace protected works.

The difficulty in finding an adequate solution lies not only in the pace of technology, but in the need to understand that these rules will determine the advancement of entire industries. The challenge is to regulate with sufficient precision to not stifle innovation, yet with the firmness needed to protect the most valuable asset of the knowledge economy: human creativity. This is where intellectual property becomes the primary driver of the new services as software.