▶ Watch ↗AI Engineer World's Fair 202619:28
Conference affiliation: Snyk
Eli Cohen co-founded Helios and served as its CEO, building tools that used distributed tracing to help developers understand, troubleshoot, and test applications. His work subsequently moved into application security: the behavior of a running system can explain both why software fails and which vulnerabilities need attention.
Cohen moved from engineering into product and startup leadership. In a published interview about Helios, he describes a formative problem at a fintech company: a service change passed its individual tests but disrupted transactions after deployment into a larger microservices system. Engineers had to reconstruct interactions across services to diagnose the failure. Cohen and his co-founder saw an opportunity to use OpenTelemetry traces to automate parts of that investigation.
Helios made those traces useful for troubleshooting, reproducing failures, and generating tests. Cohen’s 2022 Postman integration guide explains why inspecting an HTTP response alone can miss important application behavior. A service may acknowledge a request while downstream work continues through message brokers such as Kafka or RabbitMQ. Tracing those operations lets developers follow what happened beyond the initial response and investigate failures across service boundaries.
Helios moved toward application security before its acquisition. The visibility that helped developers diagnose failures could also help security teams prioritize vulnerabilities in running applications. This extended the company’s existing technology into a new customer problem.
Snyk acquired Helios in January 2024, bringing its team into Snyk’s research and development organization. Helios contributed runtime intelligence collected through OpenTelemetry and eBPF to Snyk’s application security posture management work, connecting development-time findings with information about executing applications.
Cohen’s later work addresses the pressure AI places on application security. In “AI Hackers Are Faster Than Your Pen Test”, recorded at the 2026 AI Engineer World’s Fair, he argues that increased code production and attackers’ use of AI make occasional penetration tests insufficient. His proposed response combines existing scanners with continuous offensive security: testing application behavior as software changes and validating whether suspected vulnerabilities can actually be exploited.
Three ideas explain this approach:
His security work also extends to AI applications through agent red teaming. These systems add behavior that security teams must examine alongside conventional application vulnerabilities. Across observability and offensive testing, Cohen’s recurring concern is understanding how an application actually behaves well enough to turn a finding into a useful investigation or repair.
▶ Watch ↗AI Engineer World's Fair 202619:28
Loading more talks and highlights…