Created: 21 September 2026 · Published: · Last updated:

This article is about the Babco Labs Root Cause Analysis (RCA) Agent.

The Root Cause Analysis (RCA) Agent is a heavy-duty AI universal investigation engine built by Babco Labs. It is designed to determine what actually happened, why it happened, and what evidence supports that conclusion.

The RCA Agent is not limited to software debugging. It can investigate problems in operations, quality, business processes, infrastructure, systems, and other domains where cause and effect can be tested against evidence.

Scale and swarm operation

The current Azure worker architecture is configured to scale to 30 concurrent heavy RCA Agent investigations.

RCA Agent instances can operate independently — different people, systems, or applications investigating unrelated problems — or they can be deliberately coordinated by software or another AI as an RCA Agent swarm examining different parts of a larger problem.

Invocation patterns: Human → RCA Agent · ChatGPT → RCA Agent · ERP → RCA Agent · Your software → RCA Agent · Another AI agent → RCA Agent · AI orchestrator → RCA Agent swarm

The RCA Agent does not live inside ChatGPT. ChatGPT is one interface to the agent. The same RCA Agent can be invoked programmatically from other authorized software, including software with its own AI capable of dispatching and coordinating RCA Agent investigations.

Reasoning model and analysis modes

Every current RCA Agent investigation runs on GPT-5.6 Sol with High reasoning. There is no user-facing lower-model selector.

Rapid Mode does not mean weaker AI, and Scientist Mode does not mean a better model. Both use the same reasoning engine. What changes is the investigative procedure.

Rapid Mode

Rapid Mode is the default. The RCA Agent determines whether the available evidence already supports a defensible causal conclusion. When it does, the investigation closes quickly.

Scientist Mode

When the evidence is ambiguous, the problem is difficult, or a prior remedy failed, the RCA Agent can go deeper. Scientist Mode tests competing hypotheses, derives predictions, searches for contradictions, attempts falsification, and treats failed remedies as evidence against the previous causal model.

The RCA Agent can escalate from Rapid Mode to Scientist Mode automatically, or Scientist Mode can be requested from the beginning.

Engine architecture
GPT-5.6 Sol reasoning + Lean Six Sigma root-cause discipline + the scientific method + evidence gating + deterministic orchestration + durable investigation state + specialized read-only diagnostics + parallel Azure execution

AI performs the causal reasoning. Deterministic software governs the investigation: evidence handling, state, stopping conditions, diagnostic requests, persistence, recovery, and execution.

Around durable job execution, idempotency helps retries and redelivery preserve one logical investigation instead of accidentally creating duplicate work.

Evidence discipline

The RCA Agent is explicitly permitted to conclude “I don’t know.”

If the evidence cannot support a root cause, the RCA Agent does not invent one. It can return INSUFFICIENT EVIDENCE and identify the evidence required to resolve the question.

This is a deliberate design principle: the RCA Agent distinguishes between “The evidence supports this conclusion” and “This is plausible, but the evidence does not prove it.”

Universality

The RCA Agent is designed as a universal investigation capability rather than a software-only debugging utility. Example investigations include a deployment that repeatedly crashes, a warehouse receiving error, an intermittent quality defect, a sudden margin change, an inconsistent business process, or a proposed remedy that failed.

Its job is to find the causal explanation that survives contact with the facts.

Agentic swarm operation

Because the RCA Agent is a standalone service rather than something trapped inside a chat session, software can dispatch many RCA Agent investigations programmatically.

An orchestrating AI could assign parallel investigative roles:

  • RCA Agent #1: investigate infrastructure evidence.
  • RCA Agent #2: investigate application behavior.
  • RCA Agent #3: investigate the timeline.
  • RCA Agent #4: challenge the leading causal theory.
  • RCA Agent #5: analyze why the previous remedy failed.

The orchestrator can collect those durable results and reason across them.

One RCA Agent when you need one investigator. An RCA Agent swarm when the problem deserves many.

Installation and use

Install the Babco Labs RCA (Analysis) Agent v5 plugin in ChatGPT.

  1. Open Plugins in ChatGPT from the sidebar, or go to Settings → Plugins.
  2. Choose Browse plugins and open My Workspace.
  3. Find Babco Labs RCA (Analysis) Agent v5.
  4. Select Install.
  5. Select Connect and complete authorization if ChatGPT prompts you.
Starter prompt

Copy and paste the following prompt into ChatGPT:

@Babco Labs RCA (Analysis) Agent v5 Investigate this problem using the evidence available in this conversation. Use Rapid Mode by default, escalate to Scientist Mode if warranted, and do not claim a root cause unless the evidence supports it. If the evidence is insufficient, tell me exactly what evidence you need next.