A follow-up to “The AI Didn’t Sign Your IP Agreement”
6 min read
DX’s Q2 2026 State of AI Impact in Engineering report, based on data from more than 500 teams, found that the measured share of AI-generated code increased from 34% in the first quarter to 52% in the second.1
The exact percentage will vary by study, organization, and method of measurement. The direction still deserves attention. AI is contributing a rapidly growing share of the software entering production codebases.
That figure caused me to revisit a question I recently explored in The AI Didn’t Sign Your IP Agreement.2
If AI is writing half the code, what actually represents the intellectual property of a software company?
My original article focused on ownership, accountability, and the need to understand how AI-generated software enters a company’s products.
Those issues remain important. The growing percentage of AI-generated code introduces a broader question about competitive advantage.
Having led engineering organizations responsible for complex proprietary software, and having worked through the integration of acquired platforms, I rarely saw source code alone define the value of a product.
Much of the advantage came from understanding how customers worked and why the system had evolved around their needs. Years of interpretation and operational learning were embedded throughout the product. Individual features could often be recreated. Reproducing the accumulated reasoning behind the complete system was far more difficult.
AI makes that distinction increasingly important.
Private code can still become easier to reproduce
Closed-source licensing continues to provide meaningful protection. Competitors cannot inspect a company’s actual source code, copy it, or redistribute it. Contracts can restrict access, while trade-secret protections help preserve confidential systems and methods.
Qualifying human-authored expression may remain protected, including creative modifications and the selection, coordination, or arrangement of material across the software.
A privately licensed software company does not become an open-source provider because its engineers used AI to generate portions of the product.
AI changes the economics of reproducing software.
As models improve, common capabilities can be recreated from product requirements, observed behavior, documentation, and customer expectations. A competitor may never obtain the original source code. Increasingly, it may not need to.
Complex products still reflect years of decisions and adaptation. They contain integrations shaped by real operating constraints, exceptions discovered through use, and lessons earned through failure. AI can produce implementation quickly without automatically recovering the reasoning that shaped the system.
That reasoning may become more valuable than the individual code used to express it.
The percentage provides only part of the ownership picture
A statement that AI generated 52% of the code sounds significant. It says little about the legal status or commercial value of the complete product.
Commercial software usually combines many forms of contribution. Human-authored code may sit beside AI-assisted work, licensed components, configuration, documentation, tests, and deployment systems.
The percentage alone cannot reveal the level of human direction or how extensively generated output was reviewed and changed. It also does not indicate how much value resides in the product’s data, workflow, architecture, or operating model.
A high percentage of AI-generated code may be operationally important while remaining legally incomplete as a measure of ownership.
Copyright also represents only one layer of protection. Closed software providers depend on agreements, restricted access, confidential business knowledge, and other intellectual property rights to protect the broader product.
AI-generated code should therefore be evaluated as part of the complete ownership position rather than treated as a substitute for it.
Provenance (origin) becomes part of the asset
As AI adoption expands, companies will need stronger evidence of how software entered the codebase.
That evidence will matter during an acquisition, investment review, enterprise sale, or intellectual property dispute. The organization should be able to connect approved AI tools to the terms governing their use. It should also understand what company or customer context was provided to a model.
Review and testing records can show how generated contributions moved into production. Together, these practices create a chain of custody for AI-assisted software.
A buyer will likely care less about a declaration that AI was used than about whether the organization can reconstruct that chain.
The company should be able to explain where significant contributions originated and how sensitive information was protected. It should also be able to demonstrate that generated code passed through a defined AI-enabled software development lifecycle before release.
These concerns resemble software supply chain governance more closely than a traditional debate about authorship.
Traceability becomes part of the company’s intellectual property posture because it increases confidence in the software as an asset. A company with clear records and disciplined controls will be easier to evaluate than one that allowed informal adoption and cannot explain how portions of its product were created.
The deeper issue is decision ownership
My original thinking placed too much weight on whether engineers would continue to understand every part of the codebase. That concern is relevant to current engineering organizations. It may become less central as AI capabilities develop.
AI may eventually handle most implementation and maintenance. It may monitor production, diagnose incidents, test remedies, and support future development. People may no longer need to understand every implementation detail in the same way they do today.
The enduring challenge is preserving ownership of the decisions surrounding the software.
Someone still determines which customer problems deserve investment and how the product should respond. Leaders decide which risks the company will accept and where human approval remains necessary. The organization also carries the consequences when software harms a customer, violates an obligation, or produces an unacceptable result.
AI can execute more of the work without assuming responsibility for those consequences.
The company and its leaders still carry the legal, fiduciary, and organizational accountability attached to the product.
That distinction matters more than whether a person personally wrote each block of code.
Defensibility moves beyond implementation
A company’s strategic advantage increasingly depends on what surrounds the source code.
Customer knowledge shapes what gets built. Proprietary data can make the product more useful and difficult to reproduce. Product judgment directs investment toward the right problems, while disciplined operations turn those choices into a reliable business capability.
Companies that possess those strengths will remain difficult to copy even when competitors have access to the same models.
Companies whose advantage rests primarily on the quantity of code accumulated over time may face more pressure. AI lowers the cost of recreating common functionality and reduces some of the effort required to build a credible alternative.
Complex software will not suddenly become easy to replicate. Years of customer interaction and operating experience remain embedded in the complete system.
Routine implementation may provide a weaker moat than it once did.
Rethinking my original conclusion
My original article asked leaders to take the ownership of AI-generated code seriously. I continue to believe they should.
The rapid growth of AI-assisted development expands the conversation. As AI-assisted development grows, confidence in the asset will depend on clear provenance and decision accountability, while competitive defensibility will increasingly come from the knowledge surrounding the product.
The most important intellectual property may extend beyond any individual block of source code. It may reside in the accumulated understanding that enabled the company to build the right product, adapt it to real customer needs, and operate it successfully.
Closed-source products can still be protected through contracts, access controls, trade secrets, and copyright in qualifying human-authored contributions.
It changes where leaders should look for lasting advantage.
In your organization, how much of the competitive advantage still resides in the code, and how much resides in the knowledge and decisions surrounding it?
About the Author
Phil Clark is a technology executive, advisor, and author of Profitable Engineering (profitableengineering.com). His work focuses on helping leaders connect software delivery, operating discipline, and AI adoption to measurable business value.
References
- DX. “State of AI Impact in Engineering: Q2 Report.” July 2026. https://newsletter.getdx.com/p/the-state-of-ai-impact-in-engineering
- Clark, Phil. “The AI Didn’t Sign Your IP Agreement.” July 11, 2026. https://rethinkyourunderstanding.com/the-ai-didnt-sign-your-ip-agreement/