The agentic AppSec market in August 2026

“Agentic AppSec” is not one product category. In August 2026 it describes several systems that use automation or models at different points in a security workflow.

Four lanes

Detection systems find code, dependency, secret, and configuration risks. Their value rests on coverage, precision, policy, and integration.

Triage and remediation agents add repository context, prioritize findings, propose changes, and sometimes open pull requests.

Autonomous offensive systems test reachable targets and attempt to prove exploitability. They operate under a materially different authorization and safety model.

Evidence systems preserve enough context for another person or system to review what was found, how it was found, and what happened next.

Verglos' current position

The public Verglos package ships a local JavaScript and TypeScript scanner. It produces findings and local report artifacts without requiring source upload. The repository and npm package are the current evidence for that claim.

The wider product direction is a bounded evidence agent: local verification and signed artifacts. In the current 2.0.0-alpha.1 boundary, hunt and attest are command shells, not completed workflows. This note does not count roadmap modules as market capability.

A useful comparison discipline

When evaluating an “agent,” ask what code leaves the environment, what it can execute, which targets it may contact, whether a human approves changes, what artifact remains, and whether the result is reproducible. Two products can both say “verify” while carrying completely different authority and risk.

Verglos is intentionally narrower than a cloud application-security platform or autonomous pentester. Its near-term test is whether local scanning and evidence can become trustworthy enough to hand from a builder to a reviewer. That is a product hypothesis, not a market victory claim.