# Argus & Iris

Author: Julio C. Othon
Canonical URL: https://juliocothon.com/projects/argus-and-iris
Updated: 2026-10-09
Status: In use
Period: September 2026 to present

Two Slack-based agents for prospect research, outreach preparation, and campaign monitoring at Nettle.

## The problem

Outbound involves substantial work before anyone speaks to a prospect: finding relevant accounts, identifying the right people, researching them, checking previous contact, and choosing an outreach campaign. Once outreach starts, someone also needs to track progress and keep things moving. At Nettle, much of that work was manual and fell to me as Chief of Staff. With other workstreams competing for my attention, I simply didn’t have the capacity to do it consistently.

## The approach

I built two agents and interact with both in Slack. Argus finds the people worth contacting. Iris prepares approved prospects for outreach and monitors their campaigns.

They share a lead book in PostgreSQL containing research, sources, decisions, and outreach history. Approval recorded there is the handoff between them, so their work doesn’t depend on passing messages back and forth.

## How it works

I give Argus a company, person, market, or conference list. He checks existing knowledge, researches the account on the public web, uses Apollo to find and enrich likely contacts, and retrieves their LinkedIn profiles through Evaboot’s targeted Sales Navigator searches. A separate reviewer receives the evidence without Argus’s research conversation and checks each proposal against the targeting criteria. Approved prospects move to Iris; rejected proposals retain their reasons and can be revisited.

Iris prepares each record, checks La Growth Machine for previous outreach and Attio for existing relationships, and flags conflicts. She proposes an audience from our active campaigns for me to approve, correct, or veto in Slack. Afterward, she monitors campaign progress, reports the funnel, and brings back manual steps with personalized drafts for me to review and paste into La Growth Machine. Re-targeting also requires approval.

## Key decisions

- Separate research from outreach, with independent review before the handoff.
- Keep sources, dates, and decisions in one shared record.
- Narrow the shortlist before paid enrichment, with explicit spending limits.
- Automate research and review while keeping outreach approvals and exceptions with a human.

## Current use

Argus and Iris are in use at Nettle, taking research requests through to reviewed prospects and campaign intake. La Growth Machine runs the outreach sequences. I set direction, approve audiences, resolve exceptions, and handle manual campaign steps, with Iris preparing the material and tracking what needs attention.

## What I've learned

Making the agents dependable took much longer than getting them to complete a task, including a full rebuild. Reliable operation required traceable evidence, explicit approvals, and visible failures. I also learned that a stricter reviewer isn’t always a better one, and that agents can interpret ordinary language while the system underneath enforces the boundaries.