Agentic AI Is Shipping Code. Who Owns The Risk?
Andrea Scaduto · 5 November 2026
Read the originalAndrea Scaduto, chief executive of the security training company SecureFlag, argues that the developer who approves an AI agent's change owns it, so the approval cannot be a formality.
Andrea Scaduto is co-founder and chief executive of SecureFlag, which trains developers to write secure code. Before AI agents, a developer made a change, a reviewer checked it and the service team owned what reached production. Now a product manager can give the instructions, an agent writes the change and a developer approves it, and Scaduto asks who owns the decision to ship it if the change exposes customer data.
- Scaduto's answer is the person who approves the merge: once approved, agent-written code is the organisation's code, and a reviewer who cannot explain what a change does and why it is safe should not let it ship.
- Teams should list every AI agent in use, approved or not, with who uses it and what it may do, because an agent that proposes a change differs from one that can deploy code.
- Agents should hold only the permissions they need, with a person's approval kept for riskier actions such as merging to protected branches or changing authentication and sensitive data.
- Scaduto suggests measuring from the code: vulnerability rates in agent-written changes against those written by people, time to fix, and how often problems are caught before merge.
An engineering team using AI agents can decide, agent by agent, which actions need a person's approval, and mark agent-written changes so their effect on security can be measured. This is an opinion piece by the head of a company that sells developer security training, and it reports no team's results.
Keep a list of every AI agent in use, with who uses it and what it may change, and give each agent only the permissions it needs.
Label every code change an AI agent writes, then compare how often those changes contain security flaws, and how often reviewers catch problems before merge, with changes people wrote.
Derived by Working Surface from the article; more in the Playbook. Source line: If the person signing off can’t explain what a change does and why it’s safe, it shouldn’t be deployed.
10 November 2026: as AI agents do more of the building, teams are naming the one point where a person's judgement must sit: approving a merge or release, vouching for a summary or defining good output.