Case Study

Simple email to production-ready software changes

How North America Diving Dogs lets staff ask questions and request software changes by email while an AI agent handles the planning, audit trail, preview, and review loop.

North America Diving Dogs

  • Sports & Event Software
  • Custom Portal
  • Email Intake
  • Live since July 2026

North America Diving Dogs runs a national events and member platform used by competitors, facilities, and staff across the country. Foundation operates a dedicated AI agent that lets their team request software help in plain language, without learning GitHub, ticket systems, or developer workflows.

Visit northamericadivingdogs.com
01

Email becomes structured software work

NADD staff can ask a question or request a change in a normal email thread. The agent turns that conversation into a tracked, auditable workflow.

  • The agent monitors approved NADD request threads and separates feature requests, bug reports, support questions, and follow-ups.
  • It deduplicates repeat messages so an active request continues in the same workflow instead of spawning parallel work.
  • No-change questions can be answered directly, so every email does not automatically become engineering work.
  • Requests that need code are tracked in the NADD portal with status, context, plan, PR, preview, testing, and completion details.
  • Every active flow keeps durable state, so the process survives handoffs, restarts, and customer reply delays.
For the technically curious
  • A short detector runs on a schedule to classify trusted request threads, resume existing flows, and enqueue only the work that needs deeper handling.
  • A worker processes at most one runnable request at a time, keeping the workflow controlled instead of letting multiple customer-facing actions race each other.
  • The workflow state tracks completed, no-change, waiting-on-plan, waiting-on-testing, and review-ready requests with no top-level locks in the healthy path.
  • The NADD portal acts as the customer-facing audit trail rather than burying request status in developer-only tools.
02

Plans get approved before code changes

The agent does the developer homework first, then asks for approval before changing software.

  • It inspects the NADD codebase and supporting data before proposing a fix or feature path.
  • When production context is needed, the workflow uses read-only checks instead of risky live edits.
  • The customer gets a plain-English plan that explains what will change before implementation starts.
  • If the customer asks a follow-up question, the agent continues the same request instead of losing the thread.
  • Only signed-off plans move into implementation.
For the technically curious
  • The worker uses per-request state to decide whether a flow is ready for research, plan confirmation, implementation, testing, or human review.
  • Read-only production SQL helpers are available for investigation, keeping discovery separate from mutation.
  • Plan confirmation is an explicit workflow state, not an informal comment lost in an email chain.
  • Support-only answers resolve without PRs, which keeps the system from creating code churn when the right answer is an explanation.
03

Preview-first implementation and human review

Approved changes get their own branch, PR, preview environment, customer testing loop, and developer review before production.

  • The agent creates an isolated worktree for the request instead of mixing customer changes together.
  • It implements the signed-off change and opens a pull request with the context reviewers need.
  • A preview environment is generated so NADD can test the change before production.
  • Customer approval moves the request to developer review; it does not auto-merge to production.
  • After review and merge, the agent records completion and follows up so the customer knows the request is done.
For the technically curious
  • Per-request branches and worktrees keep implementation isolated and easier to review or abandon.
  • Preview URLs are tied back to the request flow so customer testing happens against the exact proposed change.
  • The workflow can wait safely on customer testing, plan confirmation, or human merge approval without blocking unrelated requests.
  • Developer PR review remains the final production gate, preserving human accountability for software shipped to users.

Built on Foundation AI

If your team runs custom software, this is the kind of agentic development workflow Foundation builds: normal business requests come in through the channels your staff already use, then move through audit trails, approval gates, preview testing, and developer review before production.

Why this approach holds up

Meets the client in the channel they already use

NADD staff do not need to learn a developer tool. They can start with a normal request, and the agent handles the structure around it.

Turns agent work into an audit trail

Requests, plans, previews, approvals, PRs, and completion status are tracked so the workflow is visible instead of magical.

Keeps humans in the right approval points

The agent can research, plan, code, and coordinate testing, but customer sign-off and developer review remain explicit gates.

Foundation AI

Have custom software your team keeps improving through emails, screenshots, and side conversations? Foundation can turn that into an agentic development workflow with plans, previews, audit trails, and human review built in.