# Gremlin turns chaos tests into AI-generated fixes for reliability risks > Source: > Published: 2026-10-08 10:06:28+00:00 # Gremlin turns chaos tests into AI-generated fixes for reliability risks **Gremlin's Foresight AI extends co-founder Kolton Andrus's fault-injection work at Amazon and Netflix into a product that proposes fixes and retests them.** By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff) ยท Published Primary source: [PR Newswire](https://www.prnewswire.com/news-releases/gremlin-launches-foresight-ai-to-proactively-fix-reliability-risks-302898275.html) ## Why it matters Gremlin is moving chaos engineering from controlled experiments toward AI-assisted remediation. Foresight AI proposes tests and changes for teams to approve, and the test-and-verify loop could reduce remediation work. Buyers will still need evidence on accuracy and safety controls before trusting it in production. Gremlin launched [Foresight AI](https://gremlin.ai/?ref=runtimewire) on October 7th, turning the controlled failure tests that built its business into a workflow for finding reliability risks, proposing fixes and testing whether those fixes worked. Founder [Kolton Andrus (@KoltonAndrus)](https://x.com/KoltonAndrus?ref=runtimewire), who built fault-injection tools at Amazon and Netflix, is betting that engineering teams will need to automate more of that work as AI speeds up software development. The [general-availability announcement](https://www.prnewswire.com/news-releases/gremlin-launches-foresight-ai-to-proactively-fix-reliability-risks-302898275.html?ref=runtimewire) describes a product that identifies weaknesses, recommends or delivers configuration patches and infrastructure-as-code changes, then reruns the test that exposed the risk. Gremlin says the product completed a beta, but the announcement provides no beta customer count, duration, incident-prevention results or other independent performance measures. Gremlin's claim that Foresight AI has already proven it can identify and address risks before incidents remains unverified by those measures. ### From breaking systems to fixing them Before founding Gremlin in 2016, Andrus was responsible for uptime of Amazon's retail site and later built a second generation of fault-injection tooling at Netflix, after the company popularized the practice with Chaos Monkey. Gremlin's [account of its founding](https://www.gremlin.com/about?ref=runtimewire) traces the company's start to that work: deliberately testing how systems fail so weaknesses can be found before customers encounter them. Foresight AI expands the practice beyond running an experiment and handing the result back to an engineer. Gremlin says its [proprietary Failure Atlas](https://www.prnewswire.com/news-releases/gremlin-launches-foresight-ai-to-proactively-fix-reliability-risks-302898275.html?ref=runtimewire) draws on more than a decade of cause-and-effect data about system failures and recovery. The product uses that data to surface risks and suggest remediation, then checks the proposed change against the test that revealed the weakness. Gremlin says this lets engineering teams schedule reliability work before an outage forces it. In an August 6th [explanation of the product](https://www.gremlin.com/blog/what-your-ai-sre-cant-see-and-what-you-can-do-about-it?ref=runtimewire), Gremlin contrasts AI tools that respond to an incident already underway with Foresight AI's focus on failure modes teams can test in advance. Incident response tools help manage failures that get through; Gremlin says Foresight AI can help reduce foreseeable failures before they happen. Gremlin's product documentation describes an operational workflow, not just a chat interface. Its [Reliability Intelligence documentation](https://www.gremlin.com/docs/platform-reliability-intelligence?ref=runtimewire) describes diagnoses based on the test, service and health-check context, with recommendations tied to observed test results. Gremlin says use of an external large language model is optional and requires customer permission. Infrastructure buyers can assess what system context the product uses and when that context can be shared with an AI service. ### The hard part is proving the fix Verification is central to Foresight AI's pitch. A recommendation can be plausible and still leave the underlying failure mode untouched; rerunning the originating test gives engineers a way to check the result under the same condition. Gremlin says Foresight AI can produce configuration patches or infrastructure-as-code changes and repeat tests as systems evolve. Its [current product materials](https://www.gremlin.com/?ref=runtimewire) also describe risk prioritization and reliability scores across services, putting individual fixes into a broader picture of organizational resilience. The announcement leaves performance measures unanswered: it does not quantify the beta's accuracy or say how often suggested fixes were accepted without modification. Gremlin's [homepage](https://www.gremlin.com/?ref=runtimewire) says Foresight AI proposes tests and changes for the team to approve. The human approval step is explicit, while the announcement does not report how often teams accept proposed changes. For teams running mission-critical systems, buyers will still need evidence of accuracy and a clear account of safety controls, including how to stop a test or roll back a change. Public company materials identify Andrus as CEO and co-founder in the October 7th announcement and the company's current [About page](https://www.gremlin.com/about?ref=runtimewire). In February 2022, Gremlin announced that [Josh Leslie had become CEO](https://www.prnewswire.com/news-releases/chaos-engineering-leader-gremlin-crosses-10m-of-arr-hires-industry-veteran-josh-leslie-as-ceo-301478655.html?ref=runtimewire), with Andrus moving to CTO. That release reported more than $10 million in annual recurring revenue for 2021 and said Gremlin had more than 100 enterprise customers at the time; those are historical company-reported figures, not current operating metrics. Gremlin's financing history provides context for the business Andrus is extending. [TechCrunch reported](https://techcrunch.com/2018/09/28/gremlin-series-b/?ref=runtimewire) that Gremlin raised a $7.5 million Series A led by Index Ventures in 2017, followed by an $18 million Series B led by Redpoint Ventures in 2018. Amplify Partners general partner Mike Dauber, whose firm backed Gremlin, described Foresight AI in the [launch release](https://www.prnewswire.com/news-releases/gremlin-launches-foresight-ai-to-proactively-fix-reliability-risks-302898275.html?ref=runtimewire) as a trainer that does the repetitive work of strengthening systems. No new financing was announced with the product. The commercial bet is that companies shipping software faster will pay to make resilience a repeatable engineering process. Foresight AI is Gremlin's proposed way to do that: find a weakness, produce a candidate fix and test the result. Customers will need evidence that those steps work reliably across production systems without creating new risks.