Artificial intelligence is shrinking the amount of time, labour, and code required for software companies to experiment, but it is also raising questions around what work then qualifies for research and development (R&D) tax credits.
Programs like the federal Scientific Research and Experimental Development (SR&ED) tax credit and Québec’s CDAE-IA are designed to offset some of the financial risk associated with scientific and technological innovation. To qualify for SR&ED, companies need to demonstrate that their work sought a scientific or technological advancement, involved technological uncertainty that couldn’t be resolved through standard practice, and was carried out through systematic investigation or search.
“A team doing work that can be completely handled by AI was never doing SR&ED in the first place.”
Mat Rutishauser, Boast
Software development accounted for 42.6 percent of total SR&ED investment tax credits allowed in the last tax year, the largest share among all other fields of science.
As AI absorbs more of the coding and implementation work in this industry, companies have to make a clearer case for where eligible R&D takes place.
Boast is working directly on that problem. In addition to using AI and tax expertise to help businesses in Canada and the United States claim more capital from their R&D, Boast has a close view into how AI-assisted development is reshaping SR&ED claims and how companies can build a defensible case to these programs.
The tool doesn’t kill the claim
According to Mat Rutishauser, Director of SR&ED West at Boast, AI-assisted development has not changed what programs like SR&ED are designed to reward.
“Coding, whether done by a human or an AI, has always been considered support work in the context of SR&ED,” Rutishauser said. “The ‘core’ of SR&ED is rooted in the development of a new algorithm, advancing a system architecture, and designing a better model—the coding involved is simply the implementation of that SR&ED core, regardless of whether the coding was done by a human or an AI.”
What has changed, Rutishauser said, is the amount of implementation work developers may need to perform themselves. AI has moved from a novelty to a standard tool on many software teams. For some, this takes the form of vibe coding, where developers rely on natural-language prompts to generate and refine code. These new tools and methods are allowing companies to automate work that previously required more people.
“Claude Code can now do the work of 50 junior developers,” Rutishauser said. “Where you used to spend a lot of money on the support work, … filling in the gaps, putting the bricks together, and just building the scaffolding, that work is now largely being done by AI.”
But using AI does not automatically disqualify work from SR&ED. According to Boast, eligibility still depends on whether the work involved technological uncertainty, technological advancement, and systematic investigation.
Prompt engineering, for example, which involves designing inputs for AI tools that will produce optimal outputs, is generally not eligible for SR&ED on its own, but it can form part of an eligible claim when it contributes to a larger investigation into genuine technological uncertainty, according to Boast.
SR&ED’s eligibility criteria are technology-neutral by design, which means companies need to demonstrate that AI-assisted projects still satisfy the same underlying tests applied to other forms of R&D.
The lead investigator still matters
As AI takes on a larger share of the development grunt work, Boast expects the human contribution to qualifying R&D to become more concentrated among senior technical staff.
“You still need that senior architect, that lead developer, that CTO, the person who can actually see the overall vision,” Rutishauser said.
Rutishauser expects that the strongest claims to increasingly centre on the work performed by people directing the investigation, defining the technological uncertainty, evaluating possible approaches, stepping in where an AI system fails, interpreting results, and deciding what to test next. In other words, human judgement will be central to establishing that a project involves experimentation.
“A team doing work that can be completely handled by AI was never doing SR&ED in the first place,” Rutishauser added. “The flip-side is also true: a team doing SR&ED cannot be fully replaced by AI.”
Track the decision trail
Boast is working with technology and finance leaders to identify steps to make claims involving AI-assisted work easier to substantiate. Joshua Martin, Senior Director of Business Development at Boast, believes technology is moving too quickly for companies to wait until year-end to reconstruct their SR&ED claims. For Martin, the fundamental question behind any SR&ED claim is: “How are you going above what’s uncertain?”
“When what’s uncertain is evolving every quarter, work that was innovative in Q1 is no longer innovative in Q4,” Martin said. “How we position SR&ED claims and how companies should take advantage of the program is something that really should be done proactively, strategically, quarter by quarter.”
Documentation is also fundamental to substantiating an SR&ED claim, but traditional development metrics provide an incomplete picture of AI-assisted work. Boast’s position is that CTOs should consider lines of code to be a dead metric. The company is advising leaders to instead document the decision trail surrounding the work. This includes recording the alternatives the AI proposed, what was rejected and why, what failed in testing, and instances where a developer had to intervene, debug, or redirect an AI-assisted approach.
Boast also pointed to the financial value of getting the claim right. With SR&ED-eligible expenditures capped at $6 million, the company says a well-documented claim is worth up to $2.1 million in refundable cash, but a poorly documented one is worth an audit.
Boast’s platform combines AI with technical and tax expertise to surface eligible work and build the record needed to support a claim. The company has helped over 2,000 companies in the United States and Canada claim more than $900 million in R&D capital.
As AI changes the shape of R&D, Boast is betting that the companies that clearly document the work that remains genuinely experimental will be in the best position to capture the value of that work.
PRESENTED BY
Reach out to Boast today to learn more about claiming the R&D tax credits your team has earned for tackling true innovation.
Feature image courtesy Unsplash. Photo by Mohammad Rahmani.