Beyond the Co-Pilot: My Q2 Plan for Multi-Agent Automation

First off, if you noticed complete radio silence on here over the last quarter, I have a pretty solid excuse! Work has been an absolute whirlwind. Between non-stop flights between Australia, Singapore and India, setting up new offshore engineering teams, and establishing new engineering ways of working from scratch, my calendar completely ate my side projects for breakfast.

On top of that, team resource constraints meant I had to jump back into active microservice development alongside my management duties—building out core service APIs, handling edge cases, and pushing production code.

Now that the dust is starting to settle, I’m eager to get back to my AI agentic experiments.

Looking at where quality engineering is heading right now in Q23 2026, the concept of a single “AI co-pilot” sitting in your IDE writing test scripts is already starting to feel outdated. Where I’m spending my cycles over the next few months is diving deep into Specialized Multi-Agent Systems (MAS)—building networks of dedicated agents that communicate, execute, and evaluate test suites dynamically.

What I’m Planning & Experimenting With Next

Instead of relying on one giant prompt to write, run, and debug a test suite, I’m mapping out a coordinated pipeline of specialized, single-responsibility agents:

  • The Analyst Agent: Reads raw user stories, PR diffs, and OpenAPI schemas to map out the exact test surface and required test scenarios.
  • The Authoring Agent: Takes that scope and generates lean, idiomatic test automation scripts across python/playwright or REST clients.
  • The Execution & Healing Agent: Monitors headless browser or API runs, dynamically resolving broken locators or altered parameters on the fly.
  • The Triage Agent: Digs through failure logs, stack traces, and DOM snapshots to present developers with an actionable root-cause diagnosis.

The Next Experiment: LLM-as-a-Judge Evals

To keep everything reliable, the second half of this project will focus on evaluation. How do you automatically verify that an agent-generated test suite actually covers the right business logic without manually reviewing every line?

I’m setting up an evaluation loop using LLM-as-a-Judge patterns to score generated suites against coverage metrics, readability, and execution stability before anything gets merged into our pipelines.

Expect a few technical deep-dives and repository walkthroughs as I get these agent handoffs wired up!