Outreach Agent

Why It Matters

The Outreach Agent is a conversational AI that lives on every outreach page. Ask a question and get exact numbers and a next action.

It activates automatically under Outreach, alongside the Hiring Agent and Security Agent.

What You Can Do

Check campaign health: Ask how a campaign is performing. The agent returns open rate, reply rate, and bounce count against your account baseline. On a campaign page, it summarizes status the moment you open the chat.

Diagnose problems: “What’s failing?” surfaces high bounce rates, low reply rates, and subject-line problems with concrete fixes: warm up your sending domain, check SMTP reputation, A/B test subjects.

Resolve stopped deliveries: Ask which deliveries need review. The agent distinguishes an ambiguous SMTP outcome, exhausted safe retries, and known authentication, sender, content, or policy failures. It shows the attempt’s RFC Message-ID, SMTP evidence, allowed decisions, and remediation, then previews the exact effect before asking for confirmation. It never treats an error alone as proof that a message was not sent.

Recover expired approvals: If the reviewed recipient, sender identity, subject, or body changed before SMTP started, the agent explains that no delivery attempt began and asks you to inspect and approve the current message again.

Approve drafts: Approve one draft or one bounded page of up to 25 pending drafts. The agent shows the full recipient, sender, subject, and body for every item and waits for your confirmation. If more remain, next_cursor opens the next exact page after you finish the current one.

Control campaign state: Pause, activate, or complete campaigns from the chat. Reversible actions happen immediately; completing a campaign requires confirmation.

Create campaigns: Spin up a new campaign in draft status, optionally from a template. Add prospects and activate it when you’re ready.

Tune campaign config: Change drafting and sending settings from the chat: tone, language, length limit, banned words, signature, target volume, follow-up count, and auto-response behavior.

Add prospects: Give the agent a name, email, and optional context, one prospect or several at once. It adds them and kicks off AI research and drafting in one step.

Respond to replies: Draft and send a response to a prospect who replied. The agent previews the exact email and sends only after you confirm.

Suppress addresses: Add an email to the account-wide suppression list to block sending across campaigns while the address remains suppressed.

Search docs: Ask any platform question and the agent searches Kit’s documentation inline.

Silver Medalist Check

Before launching a campaign, ask the agent to compare prospect email addresses with past hiring applications. It returns matches for candidates rejected without an offer, with the role, rejection date, and reason.

This requires access to both Outreach and Hiring, plus the relevant connection scopes when using MCP. Results cover only job postings available to the acting member. The tool returns up to 10 recent matches, so the list may not be exhaustive.

Read the earlier application before referring to it in a message. An email match does not mean that the application was highly rated.

Example Queries

What you type	What happens
“What’s failing?”	Diagnoses every active campaign: flags bounce spikes, low reply rates, and subject-line problems with one fix per issue
“Approve all pending drafts for YC Launch”	Previews the first 25 complete drafts and approves only that unchanged page after confirmation; 6 remain, and next_cursor opens their next page
“Run the silver medalist check”	Cross-references prospects against past applicants; flags matches with role and rejection date
“Add [email protected] to Series A Founders, she just raised on TechCrunch”	Creates the prospect with context and triggers research + drafting immediately
“Which deliveries need review?”	Lists unresolved delivery_unknown, deferred, and failed messages with attempt evidence, RFC Message-ID, and safe next actions
“I checked Sent and provider logs. The provider confirms the message was not accepted”	Previews the audited resolution and one retry, then waits for your confirmation

The agent leads with numbers (“31 drafts pending, 68.8% reply rate”) and suggests one next action per turn.

What the Agent Can’t Do Yet

Action	Where to do it
Hand-edit the copy of a specific draft	Campaign > Messages
Import prospects from a CSV file	Campaign > Prospects > Import
Edit or remove an existing prospect	Campaign > Prospects
Set up an A/B experiment	Outreach > Experiments
Configure SMTP or deliverability settings	Outreach settings

The agent tells you where to go and offers related help it can provide.

Key Behaviors

• Check the comparison basis: the agent uses Kit metrics and should flag a small sample or missing baseline.
• Confirms before destructive actions: bulk approvals and campaign completion require explicit confirmation.
• Evidence before resend: a delivery_unknown message is never retried without human confirmation. Check Sent and provider evidence; absence from the folder alone does not prove that the email was not sent.
• Stale confirmations fail closed: a delivery-resolution confirmation is bound to the exact attempt and evidence, plus the recipient, sender, subject, and body for a retry. Any change requires a fresh inspection and preview.
• Page-context aware: open the chat on a campaign page and it summarizes that campaign immediately.

Quick Checklist

• Navigate to any page under Outreach
• Open the AI chat
• Try “What’s failing?” to diagnose active campaigns
• Run the silver medalist check before launching a new campaign
• Use “Approve all pending drafts” to review and confirm one page of up to 25, then continue with next_cursor

Next Steps

• Creating Campaigns: set up campaigns the agent can manage
• AI Research and Knowledge Base: improve the drafts the agent approves
• AI Providers & Spend Controls: manage account keys, spend caps, usage, and provider failures
• Handling Replies: manage reply threads and follow-ups