Weekly team update
MondaysInput
This week's call notes, project Slack channels, closed Jira tickets
Output
A cited status update posted to Slack
For teams on Claude & ChatGPT
Agentic automations
you can trust.
Describe a recurring job once. An AI agent like Claude or ChatGPT then does that job for you, every time. No flowchart to build.
With Pomegranate:
- You choose which sources the agent reads
- You see the sources behind every result
- You review each result before your team uses it
Weekly launch update
Every Mon · 9:00
✓
✓
✓
✓
✓✕
#team· 9:00Launch on track. API frozen, assets shipped. GA moves Thu → Mon after QA.
SourcesORION-212#launchQA recap
ORION-212#launchQA recapSame setup, any recurring job
- Competition watch· Monthly
- Customer account review· Before each QBR
- Social media planning· Fridays
- Update documentation· On each release
- Brief before every sales call· Before each call
- Release notes from closed tickets· On each release
- Weekly KPI summary· Mondays
- Customer feedback digest· Weekly
- Pipeline review before forecast· Thursdays
- Brand-voice check on drafts· On each draft
- New-hire onboarding pack· On each hire
- Investor update draft· Monthly
Why agentic automations
Workflows follow steps. Agents follow intent.
Legacy workflow · Zapier, n8n, Make…
You write every step.
- ✕Takes days to build every branch, and more to maintain.
- ✕Breaks when the input changes: a new word, a new file format.
- ✕Can't read, judge or write. AI is limited to a single step.
Agentic automation · Claude, ChatGPT…
You describe the job. The agent does it.
“Every Monday, tell the team what moved on the launch, and why.”
- ✓Set up in one sentence.
- ✓Handles messy, changing inputs.
- ✓Reads, decides and writes, like a teammate.
The catch
An agent is only as good as the sources it reads.
Left alone, the agent searches all your tools. It misses files, reads outdated versions, and opens private threads. And you can't see which sources it used.
That's why agentic automations need Pomegranate.
Pomegranate gives the agent only the sources you chose, in their latest version, with private content left out. Every result lists the sources behind it.
How it works
You choose what the agent reads. The agent does the rest.
Your tools fill a Pomegranate folder on their own. The agent reads that folder, and nothing else.
1Your sources
GmailDriveSlackNotionJira
GitHub
Teams
Salesforce1
Define the input.
Pick the sources this automation may read.
Connect Gmail, Slack, Drive and the rest to one folder. Rules keep DMs and private drafts out, and only the newest version of a doc counts.
2Pomegranate + your agent
app.pomegranate.expert
Launch
4 read · 3 left out
Launch· Notes
Weekly launch update- Launch plan v7DriveTeam
- API freeze is doneORION-212Team
- GA moves Thu → Mon after QA#launchTeam
- QA recap: 2 blockers closedGmailTeam
- Launch plan v6DriveReplaced by v7
- Sam ↔ LeaSlack DMBlocked: no DMs
- Pricing draftNotionOnly me
RulesNewest version winsNo DMsPrivate stays private
The agent reads the Team notes. Nothing else.
scheduleeventon demand
2
See what the agent used.
The folder is the full list of what the agent reads.
One line per note, with the tool it came from. Anything left out says why. Every answer links to the notes behind it, or says “I don't know”.
3Output
SlackEmailDocsOutputs folderGreenYellowRed
3
Check the output.
Review the result before anyone relies on it.
Each result is saved next to the notes the agent used. Drafts are flagged red, yellow or green against your latest notes, so you know what to fix.
Examples
Start with one automation. Probably one you already tried.
Brief before every customer call
Before each meetingInput
The account's folder: past calls, emails, CRM notes
Output
A one-page brief, every point linked to its source
Customer feedback digest
WeeklyInput
Support tickets, call transcripts, survey answers
Output
Ranked themes, each with the quotes behind it
Positioning check on every draft
On each draftInput
Your current positioning and brand-voice folders
Output
Red / yellow / green flags before the draft ships
Competition watch
MonthlyInput
Tracked competitor pages and field notes
Output
What changed since last month, and where it came from
Have an automation that “kind of works”?
Bring it. We'll put the sources it reads into a Pomegranate folder, then run the new version next to the old one so you can compare.
Bring it →Compare
Three ways to automate a job. Only one you can check.
Workflow tools are predictable but rigid. An agent on its own is flexible, but you can't see what the agent read. With Pomegranate, you get flexibility and can check every source.
How long to set up?
Workflow tool
Weak: Days
You draw every step and every branch.
AI agent on its own
Good: Minutes
Describe the job in one sentence.
AI agent + Pomegranate
Good: Minutes
Describe the job in one sentence.
What if the input changes?
Workflow tool
Weak: It breaks
A new word or file format stops the flow.
AI agent on its own
Good: It adapts
The agent reads the input like a person would.
AI agent + Pomegranate
Good: It adapts
The agent reads the input like a person would.
What does the automation read?
Workflow tool
Partly: Fixed fields
Only what each step was told to fetch.
AI agent on its own
Weak: Unclear
Whatever its search happens to find.
AI agent + Pomegranate
Good: What you chose
Only the notes in the folder you picked.
Could the automation use old or private data?
Workflow tool
Partly: If you miss a filter
You add a filter to each step yourself.
AI agent on its own
Weak: Yes
Old versions and private threads can slip in.
AI agent + Pomegranate
Good: No
Newest version only. DMs and private notes stay out.
Can you check which sources were used?
Workflow tool
Good: Yes
Each step leaves a log.
AI agent on its own
Weak: Hardly
You can't tell where an answer came from.
AI agent + Pomegranate
Good: Yes
Every answer links to the notes behind it.
Fair's fair
When a legacy workflow tool is the right answer.
If the steps never change and run thousands of times a day (copy a field, sync two tools, route a webhook), use Zapier, n8n or Make. They're cheap and predictable. Agentic automations are for work that needs reading and judgment. Many teams run both.
Rule of thumb · fixed steps → workflow tools · read, decide, write → agentic automations
Sign up
Control the input. Trust the output.
Create your account and set up your first automation in Claude or ChatGPT.
- Connect Gmail, Slack, Drive, Notion and more
- Choose what each automation may read
- Check every answer against its sources