LESSON 10 OF 10 · RUNNING IT · 3 min read
Who owns what when agents write code: AppSec, platform and developer roles
Short answer
By The Pipeline Desk · Best Agentic AppSec · Published 2026-10-10
Why ownership gets blurred
A coding agent is a developer tool, so a platform team often rolls it out. Its actions are a security concern, so AppSec wants a say. Its output lands in a developer's pull request. Without a written split, each group assumes another one is watching.
A workable split
AppSec owns:
- the agent policy: what is allowed, steered and blocked;
- risk acceptance and exceptions;
- the choice and configuration of triage and fix agents;
- the metrics in lesson 8.
The platform or developer experience team owns:
- which coding agents and models are approved and how they are installed;
- managed settings, such as Claude Code managed policy settings, that enforce hooks across machines;
- the inventory of agents and MCP servers, shared with AppSec.
Developers own:
- every change they merge, including changes written by a coding agent or a fix agent;
- following instruction files and raising exceptions through the agreed process.
Fix agents and review
A fix agent's pull request needs an owner. Assign it to the code owner of the files it touches, not to AppSec. AppSec reviews the fix agent's configuration and acceptance rate, not every change.
Regulation raises the stakes
From 11 September 2026, the EU Cyber Resilience Act requires manufacturers of products with digital elements to report actively exploited vulnerabilities and severe incidents through the CRA Single Reporting Platform, with an early warning within 24 hours. Agent-written code is still your code under that duty. NIST's Secure Software Development Framework is a common way to show customers how you build software; mapping agent controls to it keeps the story consistent.
Where vendors fit
Some platforms place agent controls with security teams and others with developers. Legit Security, for example, describes VibeGuard as discovering assistants, models and MCP servers across the engineering organization with approve or block decisions, a security-team view. Whatever you buy, decide which group operates it day to day.
Write it down
One page is enough: the three groups, what each owns, how exceptions work, and who to call when an agent is blocked. Review it every quarter as agents and tools change.
A short checklist for the first meeting
- Name the owner of the agent inventory.
- Name the owner of the agent policy and the person who grants exceptions.
- Agree who reviews fix agent pull requests in each repository.
- Agree how a blocked developer gets help within the working day.
- Set the date of the first quarterly review.
That is the end of the course; lesson 1 has the overview, and the rankings show how the platforms compare.
Sources
Source: Legit Security: VibeGuard · Read 27 Sep 2026
Source: Claude Code docs: hooks · Read 27 Sep 2026
Source: NIST CSRC: Secure Software Development Framework · Read 27 Sep 2026
Source: European Commission: CRA reporting obligations · Read 27 Sep 2026
Read next
Lesson 1: What agentic AppSec covers
Agentic AppSec has two halves: controls on the coding agents that write code, and AI agents that triage and fix findings. What each half includes.
Buyer questions
Short answers to the questions buyers ask.
Agentic AppSec rankings
The top 10 platforms ranked on nine weighted criteria.