Current status: Further research required
Description
Currently, user guides are generated using Guidde - a browser-based generative AI platform that enables users to instantly create, edit, and share video documentation, such as how-to guides and tutorials. This requires users to manually record a new video whenever an updated user guide is required - for example, if a new feature has been released.
In future, regressions tests may be used to automatically update user guides using AI whenever a new feature is developed.
Providing the scenarios and the code to test them to an LLM provides wider context on the expected software behaviour. Furthermore, having the code live in the codebase makes it much easier to build an agentic workflow which could follow the steps in the attached image.
Further research is required to conduct an initial feasibility spike to evaluate initial results. Beyond that, it would require significant development to ensure:
- tests are complete and up to date
- develop production-quality user guide generation workflow
- make infrastructure changes to integrate the workflow within the release process
It is worth noting that the by-product of opting for this approach is very-high quality and coverage of tests that would likely improve the general quality of the demonstrators through bug-discovery and better code quality.
Current status: Further research required
Description
Currently, user guides are generated using Guidde - a browser-based generative AI platform that enables users to instantly create, edit, and share video documentation, such as how-to guides and tutorials. This requires users to manually record a new video whenever an updated user guide is required - for example, if a new feature has been released.
In future, regressions tests may be used to automatically update user guides using AI whenever a new feature is developed.
Providing the scenarios and the code to test them to an LLM provides wider context on the expected software behaviour. Furthermore, having the code live in the codebase makes it much easier to build an agentic workflow which could follow the steps in the attached image.
Further research is required to conduct an initial feasibility spike to evaluate initial results. Beyond that, it would require significant development to ensure:
It is worth noting that the by-product of opting for this approach is very-high quality and coverage of tests that would likely improve the general quality of the demonstrators through bug-discovery and better code quality.