How we use AI, in our work and on this site.
We build on Anthropic’s Claude for a living, so it would be odd not to say how we use it ourselves. This page is the honest version: where AI helps us, where a person is always in charge, and what we will not do with your data.
In delivery
AI is a tool inside the work, the same way a compiler or a test suite is. It does not decide anything about your project, and it does not touch your data on terms you have not agreed.
A person is accountable for every deliverable
Code, models, pipelines and documents may be drafted with AI assistance. They ship only after the engineer responsible has reviewed and tested them, and that engineer’s name is on the work.
Your data is not used to train models
Not ours, not anyone else’s. Where a build needs a model to see your data, it sees it under your agreement, inside your environment where that is possible, and the pattern we prefer is structured outputs only.
What we build is evaluated before it ships
Every AI feature carries a graded evaluation set, guardrails and a kill switch. We hold our own use of AI to the same standard we sell.
We say when AI is the wrong answer
A large share of what arrives as an AI project is a form, a rule and a lookup. The audit says so when that is the case, and we do not build a model to satisfy a brief.
On this website
Three places where AI is at work on woodfrog.tech, each one disclosed.
Priya is an AI representative
The “Talk to Priya” chat is an AI system built by us on Claude. It says so if you ask, it commits only to what this site publishes, and a person reads every conversation and follows up. You verify your work email before it starts; the conversation clears from your browser after 30 days and we keep a copy so the follow-up starts from what you said.
How the chat works →The writingWritten from the work, checked by the people who did it
Pieces may be drafted with AI assistance. Each one is reviewed by the person who did the work it describes, and carries no number we cannot stand behind.
The writing →The picturesSome photographs are generated
Where a real photograph was not available, an image on this site may be generated. Such images illustrate; they are never evidence. Case study figures, customer stories and screenshots of real systems are never generated.
The case studies →Questions people ask about this
Will an AI make decisions about my project?
No. AI drafts, checks and monitors. People decide scope, design, what ships and what does not, and a named engineer is accountable for every deliverable.
Which models do you use?
Anthropic’s Claude models, chosen per task. We are an Anthropic Build Partner and a Registered member of the Claude Partner Network, and we build on the models directly rather than through a reseller.
Can you work without sending our data to a model provider?
Often, yes. Many builds run on structured outputs only, with no raw records leaving your environment. Where a model must see data, that happens under an agreement you have signed and nowhere else.
Is the chat on this site recording me?
It stores the conversation so a person can follow up, tells you that before you start, and clears it from your browser after 30 days. It does not use your conversation to train anything.
What if this policy changes?
This page is dated in its commit history and we change it in the open. Anything that would loosen it is a decision for the firm’s founders, not a quiet edit.
Hold us to it
If anything on this page does not match what you experience with us, write to hello@woodfrog.tech and a person will answer within one business day.