Diamond Bishop

Diamond Bishop is a director of engineering and AI at Datadog, where he leads experimental work on autonomous software agents. He co-founded Augmend, a DevOps automation startup acquired by Datadog, and is developing Dispatch Agents, infrastructure for deploying and operating always-available background agents.

Earlier in his career, Bishop worked on Microsoft Cortana, Amazon Alexa and PyTorch at Meta. At Augmend, he focused on preserving software teams’ operational knowledge and making it usable by automated collaborators. He also co-authored TreeSeg, a method that combines embeddings and divisive clustering to organize lengthy transcripts into hierarchical topics. Following Augmend’s acquisition, he joined Datadog to work on Bits AI.

How Bishop thinks about production agents

• Hypothesis-driven incident investigation: Bits AI’s on-call agent examines alerts, runbooks, logs, metrics and traces; develops and tests potential explanations; and recommends remediation. A companion engineering agent analyzes application errors and can generate proposed fixes and tests for GitHub or VS Code.
• Inspectable human-agent collaboration: Engineers need to see which hypotheses an agent considered, what evidence it collected and which runbook steps it followed. Bishop favors narrowly scoped agents whose work can be verified through offline evaluations, production feedback and evolving test sets.
• Observability for autonomous systems: Multi-step agents require instrumentation across infrastructure, model calls, tools and execution graphs. Visualizing those decision paths helps engineers locate failures that disappear inside undifferentiated activity logs.
• Agents as software customers: Bishop expects agents to use software platforms directly, requiring machine-oriented interfaces, permissions and contextual data. His comments on Datadog’s MCP server frame agents as customers needing structured access to operational telemetry.

With Shah Galeeb Ahmed, Bishop is building Dispatch Agents inside Datadog’s experimental AI products group. Their description of the project outlines managed execution, integrations, monitoring, evaluation and security for background agents. His longer-term goal is self-improving agent infrastructure, including supervisory agents that diagnose and refine other agents.