What is vibe coding, and why does the question come up?
Vibe coding means building software by describing what you want, in plain language, to an AI tool that generates the code, often without reading every line. Many SME teams use it to move fast. The question follows: can this work earn SR&ED?
What the CRA examines: the work, not the tool
The CRA does not look at how the code was written. It looks at whether the work meets two requirements: pursuing a scientific or technological advancement and relying on a systematic investigation, by experiment or analysis. The starting point is technological uncertainty: it is unknown or uncertain that a result can be achieved, because of an insufficiency in the available knowledge.
Another point often missed: in guide T4088, computer programming appears as support work. It counts when it directly supports eligible experimental development. Writing code is therefore not, in itself, SR&ED.
When vibe coding is routine
Most projects built on "feel" are routine development: a website, a dashboard, an internal tool, a standard integration. For these, the solution already exists. The AI reproduces it faster, but the uncertainty was never there.
The CRA excludes acquiring knowledge or know-how that already exists. It also states, in its 2015 policy (archived), that novelty, added features or improved functionality alone do not establish a technological advancement. An app that is new on the market is therefore not eligible for that reason alone.
When a project built with AI could qualify
The tool does not rule the project out. What matters is the nature of the problem. A project could deserve a closer look if, for example:
- You are trying to reach a performance, accuracy or reliability that no known approach achieves, and you cannot tell in advance whether it is possible.
- You must make systems work together whose technical compatibility is unknown, with no documented solution.
- Your team forms hypotheses, tests them through experiments, measures the results and draws conclusions, with the AI as one tool among others in that process.
In our view, the uncertainty has to be about your technical problem, not about whether the tool can produce code that works. These are cases to examine, never guarantees.
What evidence the CRA expects
The CRA says documents generated while the work is being carried out are the most common, and on the software side it lists software revision or commit logs among the evidence. It also mentions photographs, videos and artifacts.
A useful practice, which is not a CRA requirement: keep dated notes on the problem, hypotheses, failed trials and decisions, including what the AI tool produced and why you kept or rejected it. That thread shows an investigation took place.
What the CRA has not said yet
To our knowledge, the CRA has not published any guidance specific to AI coding tools. Without an official position, different interpretations are circulating. Take nothing for granted, and in particular:
- Do not assume that generated code counts as eligible work.
- Do not presume that an AI tool subscription is an eligible expenditure: we did not find a CRA position on it. Have it validated.
If you want an answer from the CRA before incurring costs, it has offered an optional pre-claim approval since April 1, 2026: a determination within eight weeks, before incurring SR&ED costs.
Having your project reviewed
If your team had to solve a real technical problem, with or without AI, a review is worth it, ideally before your year-end, while code logs and notes still exist. At Hubvention, we help you see whether your work qualifies and structure the file. Book my diagnostic →