Source note

My coworker Iris isn't a person

AI Coding AgentSlack AgentCode IntelligenceHuman AI InteractionDeveloper Workflow

The article describes Iris, a Slack-based AI agent at CrewAI that handles small engineering tasks such as opening code PRs. Its main claim is practical: agents can reduce context switching for low-risk work, even when they are not reliable enough for complex engineering tasks.

  • Small software fixes can carry high workflow cost: the example fix needed 2 lines of logic but would have required 9 manual steps.
  • Context switching matters because acknowledging a small task can create an open mental loop that interrupts ongoing work.
  • For larger engineering work, autonomous agents can shift effort into review and supervision rather than saving time.
  • CrewAI provisioned Iris as an internal Slack agent built on CrewAI and connected to company tools.
  • Engineers tag @Iris in Slack with a task request, such as asking for a whitespace trim fix.
  • Iris can start crews, file Linear issues, run Claude Code, open GitHub PRs, and send emails.
  • In the example, Iris found the correct handler, added trimming logic, added a newline test, and posted a PR link back in the Slack thread.
  • The example task closed in about 3 minutes after the engineer tagged Iris.
  • The task avoided a 9-step manual workflow for a 2-line logic change.
  • Iris added a test for the newline case, so the PR covered the reported failure mode.
  • The article gives no benchmark, dataset, controlled comparison, success rate, or production reliability metric.
  • The strongest claim is that agent delegation helps with small, well-scoped engineering tasks by reducing context switching; it does not claim that Iris handles complex software work well.