What Happens When Coding Agents See Your Entire System

Video thumbnail: What Happens When Coding Agents See Your Entire System
Oct 7, 20261h 14m 38s video lengthBeyond Coding

The Signal

Architect, an internal system at Adyen built by staff engineer Mat Jones, unifies fragmented operational data into a queryable graph. While initially a service-graph project, it now links code, infrastructure, and ownership data to enable automated release gating and diagnostics. The system’s evolution reflects a tradeoff between extreme ingestion scale and data completeness.

The Case

  • Scaling failed at the start: An initial ingestion path using Kafka Connect collapsed at a live-scale volume of 7 million spans per second, forcing a redesign around Apache Flink.16:56
  • Performance-driven data loss: To handle the load, the team intentionally disabled Flink checkpointing and filters out roughly 70% of ingested spans, retaining only the connective tissue between services.24:02
  • Operational control plane: Architect now informs GitLab deployment pipelines, allowing the team to gate releases based on the criticality of code paths rather than static rules.28:40
  • Governance challenges: Broader visibility is a double-edged sword that exposes sensitive data like DNS chains and merchant counts, necessitating a permission-based rollout to prevent security risks.46:25
  • Interface philosophy: Jones argues against over-relying on the Model Context Protocol (MCP); he prefers dedicated APIs or CLI tools for deterministic operations, keeping MCP limited to exploratory work to avoid tool bloat.67:24
  • Coordination primitives: The team built an internal memory layer called Atrium, which allows multiple AI agents to register file ownership and coordinate work, preventing conflict storms during parallel development.1:34

The 1 Minute Signal Take

Architect proves that real-world operational graph systems require aggressive data reduction and pragmatic tradeoffs rather than perfect telemetry. For organizations looking to replicate this, starting with a service graph—the system's 'skeleton'—is the most effective way to secure buy-in before expanding into more sensitive infrastructure and code-level domains.

Pro Analysis

Why it matters

This system represents a shift from reactive observability to an active, graph-based 'operational control plane.' It demo...

Full analysis always available on Pro.

Time saved:1h 13m 5s

Share this

Tags

Written by: 1 Minute Signal Editorial Team