How we use LLM tools

Our support is handled by people. We use LLM tools for product prototypes, website content, and quality assurance work under human supervision.

Blue industrial controller modules with connected network cables and colorful wiring

TL;DR: Our support team handles every customer request without LLM tools. In development, we use LLMs moderately to explore prototypes, then people modify, review, and validate any resulting code before delivery. For this website, LLMs help with content drafts, but people review every result; documentation is written by people and checked with LLMs only for quality assurance. We use LLMs more extensively in quality assurance, with a person starting and monitoring every use. Our automated development checks and tests use deterministic tools, not LLMs.

Support ​

Customer support is handled entirely by people. Only our team processes the support inbox, and customer requests are never sent to LLM tools. We do not use LLMs to write replies to customers.

Development ​

In product development, we use LLM tools moderately and only to explore prototypes. We do not deliver code exactly as generated by an LLM. People heavily modify any code developed from a prototype, then fully review and validate it before delivery.

We also use automated tools to check code quality and run integration and end-to-end tests. These workflows use deterministic tools rather than LLMs, so the same input produces the same output.

Website development ​

We use LLM tools to help draft website content with search visibility and web standards in mind. A person reviews every LLM output before it is used on this website. No automated workflow uses LLM tools to generate website content.

The documentation on this website is written by people. We use LLM tools only for quality assurance of that documentation.

Quality assurance ​

We use LLM tools extensively to help with security audits and code reviews, generate test data, test scenarios, and sample files, and explore possible use cases. We also use them to check how SFTPPlus MFT aligns with relevant standards and best practices. A person starts and monitors every use, reviews the findings, and decides what action to take. We do not run autonomous LLM agents because their nondeterministic output cannot be relied on for unattended work.

Our automated tests and other quality assurance automation use deterministic tools, not LLMs. For more about our release tests and ways to validate your own deployment, see SFTPPlus MFT quality assurance.

Banner photo by Raymond Sime on Unsplash.