When the Robot Writes the Code, Who Gets the Credit? How AI-Assisted Development Impacts Research Tax Credit Eligibility

When the printing press arrived in the 15th century, scribes feared obsolescence. The machine could replicate in hours what took them weeks. Yet operating a printing press still required skill, judgment and expertise. The same was true with the invention of the typewriter and later, the computer. The craft evolved. It didn’t disappear.

Software development is having its printing press moment. However, the big question for technology companies and research tax credit (RTC) professionals is: if artificial intelligence (AI) is writing the code, does the engineer’s activity still qualify for the RTC? The short answer is “yes,” but the underlying reasoning matters, as does the documentation that supports it.

The Regulatory Void

One threshold issue is the lack of formal guidance. Currently, there are no regulations, case law or IRS guidance that address AI-assisted development and the RTC. That means taxpayers must apply the existing four-part test framework or the additional three criteria under the High Threshold of Innovation (HTI) for Internal Use Software (IUS) business components.

Importantly, the existing four-part test framework provides a taxpayer friendly position that, when paired with proper documentation, can be defensible during the exam.

Walking Through the Four-Part Test

Permitted purpose: This factor remains straightforward. Developers are still seeking to improve function, performance, reliability or quality. Whether hand coding every line or describing the objective and using an AI tool to generate a draft (sometimes referred to as descriptive or prompt‑based coding), the intent is the same. AI is a methodology change, not a change in a business component’s purpose.

Technological in nature: This is often where questions arise. If a developer is typing prompts in plain English, does that activity remain technological? The argument can be made that yes, it’s still technological. The developer’s role is evolving, not disappearing. With AI-assisted tools, developers operate at a higher level of abstraction, focusing on system architecture, logic, scalability, latency and system performance. They continue to apply principles of computer science, with the human remaining the primary decision-maker and AI serving as a tool.

Elimination of uncertainty: Some may argue AI’s massive knowledge base eliminates uncertainty. In practice, uncertainty still exists. Depending on the AI tool, uncertainty may arise from the underlying learning model or the data used to train it or essentially, whether the tool is the right one for the job. Additionally, AI reframes uncertainty rather than removing it. Instead of asking, “What does the correct code logic look like?” engineers are evaluating questions such as, “Which AI-generated architecture had the best performance and why?” and “Did the AI-generated code function as expected when integrated into the development environment?” In many cases, AI still generates outputs that require significant refinement before reaching the final solution.

Process of experimentation: This test may actually be strengthened by AI-assisted development. Developers can rapidly generate multiple alternatives, but each option still requires systematic testing, validation and refinement. From an RTC perspective, the focus shifts from manual hand-coding improvements to the cognitive work involved in designing tests, assessing results and making critical decisions. The iterative loop —prompt, select, refine, test and identify errors — remains central to experimentation.

A Note on Internal Use Software

With respect to whether AI-powered software is subject to the more stringent HTI test for IUS, there is good news: software developed for interaction with third parties, such as customer-facing chatbots, AI-driven mobile applications or client portals with intelligent features, generally falls outside the IUS definition under the finalized regulations. Emphasis should remain on what is being developed and how the product will be used, rather than how it was developed. If the software is designed to be commercially sold, leased, licensed or provided to customers in exchange for consideration, the standard four-part test applies. This distinction is important because the HTI test can be difficult to meet. Many AI development projects involve outward-facing functionality, which means the more favorable framework likely governs.

Documentation: What Needs to Change

As the nature of the work evolves, the documentation must evolve with it. Traditional metrics such as lines of code become less meaningful when AI can generate thousands of lines in seconds. Documentation should instead capture the thinking process of the engineer and the experimentation performed.

Practical steps:

• Document what AI-generated alternatives were considered: This includes documenting why certain options were rejected. JIRA histories and Confluence pages often contain this detail.
• Review code churn rates: High churn within a feature branch may suggest active iteration rather than passive acceptance of AI output.
• Evaluate test-to-production code ratios: A rising ratio could be a signal that the experimentation effort was complex.
• Consider the infrastructure costs: Cloud computing expenses tied to AI development, including GPU rentals and cloud-hosted development environments, may qualify as computer rental expenses under Section 41. These costs are frequently missed and can materially increase qualified research expenses (QREs).

Weaver Can Help

AI is a transformative tool and is reshaping how development work is performed. At its core, however, this remains human-led activity. The developer’s role has shifted toward systems architecture and decision-making. The role has elevated, not disappeared.

Because the IRS has not yet issued guidance, overly aggressive positions may draw scrutiny during exam. Documentation and thoughtful positioning remain critical. When guidance does emerge, it will matter. Just as the printing press didn’t eliminate human craft, AI doesn’t eliminate human ingenuity. It changes how that craft is expressed.

With AI rapidly changing development workflows, organizations are looking for confidence that their processes continue to meet RTC requirements. Our tax professionals can help you interpret the rules, refine documentation and identify risks or opportunities. Contact us to learn more.

©2026