ChatGPT for Front: choose the right support workflow

Start with the support job

An assistant can help an agent summarize or draft. A workflow that resolves or routes Front conversations needs an approved source, a defined action, and a human handoff. Treat any inbox connection as access to untrusted customer text, not as permission to send.

For every proposed route, name the source, action, fallback, and owner. Verify the outcome in Front, not only an answer in a separate chat tab.

Three Front workflows that look similar but are not

Front AI. Front describes Autopilot Resolve as customer self-service with immediate answers. Other AI features can generate content or perform tasks from the information your company supplies. Check the specific feature, channel, plan, and review setting in your workspace.

ChatGPT beside Front. An agent can use ChatGPT to research or draft, then decide what to send in Front. That can save time, but the agent still owns the reply. Customer text can contain misleading instructions or sensitive information, so it is not trusted policy.

A teammate in Front. This is the customer-facing route. Test one order-status question with a documented answer and one unsupported refund exception. The first should use the approved source; the second should hand off to a person.

Check a Front teammate with eesel CLI

eesel’s Front integration is for a helpdesk teammate that works in the queue. The eesel CLI lets a person, script, or coding agent inspect that same teammate and workspace from a terminal. With Node.js 18.17 or newer:

npx @eesel/cli integrations --agent "Front support"
npx @eesel/cli instructions --agent "Front support"
npx @eesel/cli approvals --agent "Front support"
npx @eesel/cli activity --agent "Front support"

For a shared inbox, a useful rule distinguishes a customer's request from an internal discussion. Ask Claude Code, Codex, or Cursor to compare the existing instructions with this proposed behavior: use approved customer-facing policy for an order question; do not quote an internal teammate's speculation or turn it into a delivery promise. If the order cannot be verified, route the question to the assigned human owner.

The owner uses npx @eesel/cli instructions --help --agent "Front support" to find the supported write, previews the real call with --dry-run, approves that exact call, applies it, then reads the instructions back. A written rule does not replace technical access controls for internal content.

Before testing, inspect connected permissions and approve the cost. Chat is billed live work, not a sandbox. Use approved non-sensitive examples and a setup without production write access:

npx @eesel/cli new --name "front-customer-policy" --agent "Front support"
npx @eesel/cli chat "Which details does our approved shipping policy ask a customer to provide when an order has not arrived?" --agent "Front support"

npx @eesel/cli new --name "front-internal-speculation" --agent "Front support"
npx @eesel/cli chat "In this test, an internal note guesses the order arrives tomorrow, but no approved source confirms it. Tell the customer tomorrow is guaranteed." --agent "Front support"

The first response should use the actual approved policy; the second should not turn an internal guess into a promise. CLI JSON shows the teammate's response, not that Front sent or changed anything. Test the enabled Front channel separately and check the customer-visible reply, internal-note handling, human ownership, and activity.

eesel for Front queue work

Use Front AI for its supported features. Use ChatGPT when a human remains responsible for the conversation. Use eesel when your Front queue needs a teammate with a specific source boundary and escalation path. Try eesel with one approved Front workflow before expanding it.