All jobs

What Did Customers Ask Us to Build This Week? A SaaS Customer-Success Team's Friday Digest From Shared Slack Channels

A SaaS customer-success team's agent reads every shared Slack channel with customers each Friday and lists the week's feature requests in their own words.

Run this on RushOpens the console with this prompt ready to send. Connect Slack there first.

What the agent does

This is for a B2B software company whose customer-success team talks to its larger customers in Slack Connect channels, one shared channel per account: #ext-northwind, #ext-harbor-logistics, #ext-clearwater-schools. Customers ask for things there all week, in passing and in threads: an export in another format, a permission for a role that does not exist yet, a filter on a report. The product team wants those requests every week with the customer's exact words, and the customer-success manager who heard them is in a renewal call when the planning meeting starts. The agent collects them every Friday afternoon.

The run names one connection, slack, signed in as the customer-success lead, who is a member of every shared channel. Rush's Slack sign-in asks for channels:read, channels:history, groups:read, groups:history and users:read, among others, and the agent uses the slack CLI as that person. It calls conversations.list with types set to public_channel,private_channel and exclude_archived set to true, following next_cursor until the list ends, and keeps channels whose is_ext_shared is true. For each one where is_member is true it reads the week with conversations.history, opens every thread with conversations.replies, and calls users.info to put a name on each author.

The prompt names the company's own people (the customer-success and product team) so the agent can tell a request from a customer apart from a reply by staff. A request is a customer message asking for something the product does not do, or does in a way that blocks them; a bug report, a billing question or a how-do-I question is not, and the agent leaves those out. For each request it records the account (from the channel), the person, the date, the request quoted word for word, and whether a staff member replied with a commitment, such as "on the roadmap" or "we'll look at it", quoted as well.

It then groups requests that ask for the same thing in different words, with a short plain heading for each group ("Export reports as XLSX"), and lists the accounts behind each group. It writes requests.md and posts nothing: the digest goes to the product team through the customer-success lead, who reads it first. Shared channels the lead is not a member of are listed by name at the end, so a missing account is visible rather than silently absent.

Start it with one API call

POST https://api.phoenix.so/o/{org}/sessions with this body:

{
  "agent": "claude",
  "input": "Collect this week's customer feature requests from our Slack Connect channels, for messages from 2026-11-09 00:00 to 2026-11-13 17:00 America/New_York. Use the slack connection. Call conversations.list with types public_channel,private_channel and exclude_archived true, following next_cursor, and keep channels where is_ext_shared is true. For each one where is_member is true, read the window with conversations.history and every thread with conversations.replies; call users.info to name authors. Our staff (not customers): Dana Ortiz, Sam Patel, Lee Chen, Priya Rao, Tom Burke. A feature request is a customer message asking for something the product does not do or that blocks them; skip bug reports, billing and how-to questions. For each: account (from the channel name), person, date, the request quoted exactly, and any staff reply that commits to something, quoted. Group requests asking for the same thing under a short heading and list the accounts in each group, largest group first. Write requests.md. List at the end any is_ext_shared channel where is_member is false. Do not post, react or message anyone.",
  "connections": [
    "slack"
  ]
}
  1. Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
  2. Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
  3. Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.

What comes back

The session returns requests.md as an artifact, and the answer names the largest groups. The customer-success lead reads it on Friday afternoon and forwards it to the product team, usually after merging a group the agent kept apart or splitting one it joined too eagerly. Every quote carries its channel and date, so a product manager who wants the context opens the thread in Slack.

What the lead checks: that a quoted request really is one (a customer musing "it would be nice if" in a thread about something else is a judgment call), that every commitment a staff member made is one the company means to keep, and the list of shared channels the lead could not read. Nothing in the digest is counted toward anything until a person decides it is a real request.

Run it on every Friday afternoon

The lead saves the agent once with POST /o/{org}/agents, naming the slack connection and these instructions, and adds a console automation with a schedule trigger (0 16 * * 5, timezone set to the team's) that runs it. A schedule runs the saved prompt as written, so the saved prompt asks for "messages since Monday 00:00" in place of the fixed dates above. Setting notify_each_run to true brings each Friday's result to the lead on its own instead of folding it into one daily message.

A customer-success manager who wants the same digest for one account before a quarterly review can start the request above from the console with that account's channel named in the prompt and a longer window.

The use case this job follows, with its automation and trigger: Produce a report or file, no repo.

Why run it on Rush?

The requests are already written down in Slack; the work is reading every shared channel, every thread, and separating what customers asked for from everything else said there. The Slack sign-in stays in Rush's connection as the lead's own account, so the agent reads exactly the channels the lead can read, and no bot has to be invited into each customer's channel.

Each run has its own computer and keeps a record of every Slack call it made and what its model calls cost. Because each Friday is its own session, the company keeps a dated history of what each account asked for, which is the evidence a product manager needs when a customer says they asked for something months ago.

Frequently asked questions

Does the agent see Slack Connect channels it was not invited to?

No. It reads as the lead's own Slack account, so it sees the shared channels the lead is in. Slack's conversations.list returns is_member for each channel, and the digest lists every shared channel the lead is not a member of so nobody mistakes silence for no requests.

Will customers see anything from the agent?

No. The prompt forbids posting, reacting and messaging, and the run only calls read methods. The digest leaves the run as a file the lead reads, and the transcript records every Slack call.

How does it tell a customer from our own staff?

The prompt lists the staff names, and anyone else in a shared channel is treated as the customer. Keep that list current when the team changes, or a new hire's reply may be quoted as a customer request.

Can it file the requests in our tracker?

This page keeps the agent to a digest, so a person decides what becomes a ticket. A second step can take the approved groups and create issues in a tracker Rush lists, such as Linear, in a run that names that connection.

More Slack jobs

More jobs like this: Produce a report or file, no repo

Keep reading