Why Your Agent Disagrees With Itself (And What To Do About It)
AI Engineer World's Fair 2026 · 25:38
Observability and cloud security
Datadog provides monitoring and security software for developers, IT operations teams and business users. Its platform brings applications, infrastructure, data, models and security into one place. Application performance monitoring helps teams investigate application behavior, while Log Management supports troubleshooting and Real User Monitoring tracks user journeys and frontend performance. Bits AI adds agents that investigate and remediate issues, extending the platform from observing systems into operational work.
Co-founders Olivier Pomel, CEO, and Alexis Lê-Quôc, CTO, previously worked in technology leadership and operations, respectively, at Wireless Generation. Datadog’s engineering contributions include DDSketch, a compact, mergeable data structure for estimating percentiles across distributed monitoring data. It controls error relative to the measured value, helping teams analyze latency distributions without retaining every observation—a practical foundation for understanding slow requests that averages can obscure.
Datadog acquired Adaptive ML in 2026 to expand Datadog AI Research’s work on world models and agentic language-model post-training for observability. Its deployment options also include Bring Your Own Cloud for running Datadog within customer environments. At the end of Q2 2026, the company reported approximately 4,720 customers with at least $100,000 each in annualized subscription revenue run rate, indicating the scale of its larger customer base.
AI Engineer World's Fair 2026 · 25:38
Affiliations reflect their AIE appearances, not necessarily current employment.
Start with Diamond Bishop's talk to learn how agent workflows extend from alert investigation to proposed code fixes, alongside evaluation challenges and human-agent collaboration.
Diamond BishopAI Engineer Summit 2025
Turn to Diane Lin's talk to understand why cybersecurity alerts near ambiguous decision boundaries can produce contradictory outcomes without necessarily indicating a broken model.
Diane LinAI Engineer World's Fair 2026
Bishop describes an AI On-Call Engineer that combines runbooks, logs, metrics, traces, and hypotheses with tool-based reasoning to investigate alerts.
Lin argues that inconsistent decisions can expose unclear labels or conflicting policies. Semantic knowledge and similar past decisions provide ways to improve consistency while prioritizing human review.
Affiliations reflect each recorded session, not necessarily current employment.