Responsible AI

Where do we use AI in client work?

To write and review code, to plan and carry out well-defined engineering tasks, to summarize meetings, and to synthesize what your team tells us in AI Enablement sessions.

• Engineering. Our engineers use AI coding assistants, including Claude Code and GitHub Copilot, to prototype, write and test code. Our own agent pipeline drafts a plan for a well-defined request, then carries it out on a branch in your repository and opens a draft pull request for review. It only reaches a repository you have connected through our GitHub app, which you can remove at any time.
• Meetings. When we record a client meeting with Fireflies.ai, we use Claude to summarize the transcript, list the action items and flag concerns for our team.
• AI Enablement. In AI Enablement for Product & Engineering, AI groups the anonymous notes from a Fear Mapping session and drafts the session narrative and workshop recap your facilitator then edits.
• Project administration. AI sorts pasted statement-of-work text into milestones and routes an incoming request to the right project. It picks which of your pasted lines is which, and doesn't write new ones.

The AI agents we build for clients are governed by that project's agreement, which sets out their data, review and security requirements.

Is your data used to train AI models?

No. We don't use your data to train models, and Anthropic, which provides the Claude models behind most of our AI features, doesn't train on data sent through its API by default.

We also use TypeSafe for classification in our internal tools, Fireflies.ai for meeting transcription, and GitHub Copilot alongside Claude Code for engineering work. We haven't verified each of those vendors' training terms, so we configure tools so client code isn't used for training where the vendor allows it, and we state the no-training default above only for Anthropic's API.

Who reviews what AI produces?

An engineer or a facilitator, every time. AI output doesn't ship in your product, or reach you as our work, until a person has reviewed it.

• Code. AI-generated code is peer-reviewed by an engineer before it ships. Our agent pipeline can open a draft pull request, and it cannot merge one. The option that lets a simple request skip the human-reviewed plan is off unless a project turns it on.
• Session write-ups. AI drafts Fear Map and workshop write-ups, and the facilitator reviews and edits them before anything is shared with you. Notes the AI can't place with confidence stay ungrouped for the facilitator to sort by hand.
• Meeting summaries written with Claude are for our team, to help us follow up. In the client portal you can see the transcript and Fireflies.ai's own summary for the meetings a member of our team has chosen to share with you.

What data does AI see?

Only what the task needs, sent only to the providers named on this page and in our Privacy Policy.

• Fear Map and workshop synthesis receives anonymized response text and totals, never a participant reference or anything that identifies a person.
• Meeting summaries receive the transcript, which includes attendees' names and what was said.
• AI provider keys are held on our servers, and are never sent to a browser.
• If your agreement restricts which tools may touch your data, or where that data may go, we follow the agreement.

Will you tell us when AI is involved?

Yes. At the start of an engagement we will tell you which AI tools we plan to use on your project and what they will see, and if you want a tool kept away from your data, we will work without it.

Can AI features be switched off?

Yes. Every AI capability in our client portal and internal tools is registered in one place, and an administrator can switch any of them off from a settings page, without a code change.

Our software refuses to run an AI feature that hasn't been registered, so none can slip in unreviewed. Where a feature is off, the manual route still works: a facilitator can write the narrative by hand, and pasted milestones can be sorted by hand.

How is AI work secured?

AI work runs under the same controls as the rest of our delivery, described in our Security Policy. To ask about how we use AI on your project, use our contact page or call (888) 369-1330.