Agents: Say Who to Chase and When. It Writes Every Email Itself.

Last updated: September 2026

In July I wrote about drip programs and stage blasts. They still run, on the same engines. What changed is how you set one up. You used to build a cadence step by step. Now you tell FUNDesk what you want in a sentence, and it builds the agent.

"Every Tuesday at a random time, email my Open Approvals a nudge to take the approval." That's a complete setup. If it can't tell which stage you mean, it asks one question. Then it shows you the plan, and every setting on that plan shows the words it came from, so you can see it read you right and change anything it didn't.

The test run comes first

Nothing sends on the first run. The agent writes five drafts and you read them. If they sound like you, promote the agent and it starts working the stage. If they don't, fix the goal or the stage and run the test again. I put this in front of every agent because the failure mode of automated outreach is a message you'd never have sent going to a merchant you needed.

Two kinds

A scheduled agent runs the whole stage on a clock: every Tuesday, every weekday morning, or once. That last one is the old blast. The same thing, once, to a whole stage, paced to protect your sending domain.

A sequence agent puts each lead on its own cadence from the day it enters the stage. Email or text. That's the old drip, set up in a sentence.

One stage per agent. Your own deals, or as an admin, everyone's or one rep's.

One email per lead, from that lead's file

There's no template and no mail merge. Three leads in the same run get three different emails, because the writer starts from each lead's own file: the business, where it sits in the pipeline, what's missing, and the goal you gave it. Every message carries that goal, whether it's "take the approval", "send the statements", or "finish the application".

You can teach it your voice. On the agent's page, pick the emails that got replies, and the writer matches them from then on.

When it stops

An agent stops for a lead on a reply, when the lead leaves the stage, or after six touches. A merchant who answered on Tuesday doesn't get a scheduled nudge on Wednesday. The agent moves on to the next lead.

The same guards as every other email

An agent runs through the same engines as a hand-approved campaign, so it can't skip a check. Consent and do not contact are checked before every send, and a held lead shows up in the run with its reason. Sends go out about a minute apart, 200 a day per workspace, and you can raise that to 1,000. A member's agent sees that member's deals. Running one across the whole team is admin only.

Runs are reviewed by default. Once you trust an agent, you can let it send on its own.

What you see

Each agent shows leads reached, replies, holds and failures by reason, and the cost per email. That last number is there so you know what an agent costs to run before you spin up six of them.

You can set one up from the Agents page or from Agent Chat: "Every Monday morning, check in with my App Out leads" builds one. The Campaigns page now opens Agents. The product page is at fundesk.ai/product/agents.