Chainguard has introduced Chainguard Checks, a continuous verification engine designed to run software tests and security checks directly within AI-assisted development workflows. The platform brings continuous integration (CI) closer to where developers and coding agents write and modify code, returning structured test results while work is still in progress. It also uses isolated microvirtual machines, short-lived credentials, network access policies, and signed attestations to strengthen software supply chain security. The launch addresses a growing infrastructure challenge for agentic software development: keeping verification fast and secure as AI agents generate and revise code more frequently.
Chainguard Moves CI Verification Into AI Coding Workflows
Why traditional continuous integration is under pressure
AI coding agents can generate, revise, and test software through repeated cycles of automated work. But as these systems become more autonomous, traditional continuous integration pipelines can struggle to keep pace. CI systems typically validate code after developers submit changes, creating queues and repeating environment setup tasks as development activity increases.
Chainguard is targeting this problem with Chainguard Checks, a continuous verification engine designed to make testing a callable tool within agentic development workflows. Instead of waiting for a completed change to enter a conventional CI pipeline, an AI agent can run checks during development, receive structured failure information, and continue iterating.
The product is designed to help engineering teams identify problems earlier while reducing repeated setup work and strengthening controls around untrusted code. Chainguard is positioning verification as an integral part of AI-assisted development rather than a final checkpoint before code review.
Verification becomes part of the agent loop
Chainguard Checks is designed to work alongside the company’s newly introduced Chainguard Workspaces environment. During a check, an agent session can suspend to stop consuming compute resources, then resume with structured failure logs available in its context.
This creates a feedback loop in which an agent can interpret test failures, adjust its code, and run verification again. The workflow is intended to reduce delays between generating a change and learning whether it works.
A central component is the use of warm snapshots. Chainguard says each CI run can start from a snapshot with dependencies already installed. When a project’s lockfile changes, the snapshots are rebuilt, allowing environment preparation to happen in the background instead of being repeated for every commit.
The approach targets a familiar source of development overhead: repeatedly provisioning environments and downloading dependencies. In agent-driven workflows, where a task may involve numerous iterations, reducing this repeated work could improve responsiveness and resource efficiency.
Chainguard has not disclosed independently measured improvements in execution time, queue length, or compute costs, so the scale of those benefits remains to be established.
Isolated execution addresses software supply chain risks
Speed is only one part of the problem. CI runners frequently execute code from repositories, dependencies, and automation scripts that may not be fully trusted. If those environments also hold persistent credentials, a compromised workflow can create opportunities for credential theft or unauthorized access.
Chainguard Checks is designed to limit that exposure through isolated microvirtual machines. The company says each workflow step runs in its own hardware-isolated environment built with Chainguard OS Packages, which it describes as zero-CVE packages, including FIPS-compliant toolchains.
The platform also avoids placing long-lived credentials directly in runners. Instead, a host-side proxy can add short-lived credentials to outbound requests, while tools that require credentials on disk receive encrypted, short-lived substitutes.
Declarative network egress policies provide another layer of control. Teams can restrict connections to approved destinations and configure policies to deny unauthorized requests or destroy a microVM when required.
These measures are intended to reduce the risk of malicious dependencies, prompt-injected agents, and other untrusted code reaching sensitive resources. They do not eliminate all software supply chain threats, but they move important security controls into the execution environment rather than relying exclusively on the behavior of an AI agent.
Signed attestations aim to reduce redundant testing
Each verification run produces a signed attestation linking the execution environment, test result, and commit. Downstream merge gates can verify the attestation and, when configured requirements are satisfied, merge the code without running the same CI test suite again.
This feature targets redundant computation in conventional development pipelines. If a passing result can be trusted and verified downstream, teams may be able to avoid repeating identical checks solely because code has moved to another stage.
The design also creates a record that can support software provenance and verification policies. However, skipping a repeated test is appropriate only when the attestation is trustworthy, corresponds to the exact code and relevant environment, and satisfies the receiving pipeline’s policies. Changes to dependencies, configuration, or execution conditions may still require fresh testing.
Signed evidence therefore complements rather than replaces sound testing practices and carefully configured merge controls.
Enterprise adoption and the road ahead
Australian payments provider Cuscal is participating as an early design partner. Brad McCoy, the company’s senior manager of cloud platforms, said secure, efficient development workflows are becoming a priority as the organization expands its digital payments infrastructure.
RedMonk analyst and co-founder James Governor also highlighted the pressure agentic development places on existing CI infrastructure, including potential downtime, lost productivity, and delays in shipping code.
Their comments illustrate the problem Chainguard is addressing, but they do not establish independently measured product performance.
The broader market opportunity sits at the intersection of AI development tools, DevSecOps, software supply chain security, and cloud infrastructure. As coding agents take on more development tasks, organizations will need systems that can validate code continuously without exposing credentials or creating unsustainable testing overhead.
Chainguard Checks offers one approach: combine fast verification, isolated execution, restricted network access, and signed evidence within the agent’s development loop. Its longer-term value will depend on how reliably these mechanisms integrate with existing repositories, testing frameworks, and enterprise release policies.
Market Landscape
AI-assisted software development is increasing the frequency of code changes and placing new demands on continuous integration infrastructure. Traditional CI pipelines remain essential, but repeated environment setup, queueing, and post-commit validation can become friction points when agents generate many iterations.
Chainguard Checks reflects a broader shift toward continuous verification and shift-left security—testing and enforcing controls earlier in the development lifecycle. Its combination of isolated microVMs, credential protection, network restrictions, and signed attestations also addresses software supply chain risks associated with executing untrusted code.
The competitive landscape includes established CI/CD platforms such as GitHub Actions, GitLab CI/CD, and Buildkite, alongside emerging AI-native development environments. These products differ in architecture and capabilities, so their performance should not be assumed to be directly comparable.
For engineering leaders, the key evaluation criteria are verification latency, test reproducibility, security boundaries, attestation integrity, compatibility with existing pipelines, and total compute cost. Chainguard’s approach is notable for tying verification evidence directly to the code and allowing downstream systems to reuse passing results when policy permits.
Top Insights
- Chainguard Checks moves CI verification into AI coding workflows, enabling agents to receive test feedback and iterate before submitting changes for review.
- Warm snapshots preserve installed dependencies, reducing repeated environment setup across development iterations and targeting unnecessary compute overhead.
- Hardware-isolated microVMs, short-lived credential handling, and configurable egress restrictions are designed to reduce risks from untrusted code.
- Signed attestations link environments, commits, and test results, allowing eligible downstream merge gates to verify prior checks without rerunning identical suites.
- Cuscal is an early design partner, but independent benchmarks demonstrating performance improvements and cost reductions have not been disclosed.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
