AI agents can generate code, refactor it, and create pull requests—but the PR is only the midpoint of software delivery. Building, testing, approvals, security, deployments, rollbacks, and incident response still span fragmented tools and workflows.
So how do we give agents safe, reliable control over the rest of the delivery lifecycle?
This talk explores why the command line is emerging as a control plane for agentic software delivery. With capabilities exposed through standardized interfaces such as the Model Context Protocol (MCP), agents can interact with testing, security, deployments, and operational tooling through a consistent workflow.
We'll demonstrate an incident resolved from a single terminal: an agent identifies a feature-flag-related error spike, requests just-in-time access, rolls back the change, verifies recovery, and documents the incident.
You'll leave with practical patterns for building agent-native delivery workflows and safely connecting AI agents to your software delivery platform.
About the speaker:
Mohit Suman works as a Staff Product Manager at Harness, Developer Experience, shaping how engineers and AI agents interact with the platform. With a hands-on software engineering background, he brings deep technical fluency to product decisions. His focus is on building developer tools that are equally legible to humans and to the coding agents now working alongside them.
