AI-Native Operations
4SO Autonomous SRE
An AI-native operations platform for observation, evidence-grounded RCA, remediation planning, approval and durable execution.
- Evidence-first AI
- RCA
- Policy
- Approvals
- Durable execution
The problem
Problem
An unconstrained LLM is dangerous in production; hallucination, prompt injection or incomplete context must not become direct mutations.
Why It Is Hard
RCA requires correlation across heterogeneous evidence while remediation must stay explainable, resumable and compatible with approval and recovery boundaries.
The decision
Leadership / Architecture Role
Designed the model of AI for reasoning and deterministic systems for authority, with human-in-the-loop gates for high-risk operations.
Key Decisions
Do not make the LLM authoritative; keep human, CLI, web and MCP interfaces on one domain model; support local models as an option without making normal runtime depend on a model.
The architecture
Architecture
Observation and evidence storage are separate from reasoning; policy and admission precede mutations; a durable executor performs work and post-change verification reads the result back from runtime.
Failure & Recovery Model
Each action has durable state, idempotency and verification; crashes or restarts must not turn a plan into uncontrolled duplicate mutations.
Security / Authority Model
Prompt content is not trusted; tool capability, RBAC, policy, approvals and secret boundaries are enforced independently of the model.
The value
Product Experience
Operators see evidence, hypotheses, impact, proposed actions, approval requirements and verification rather than a chat transcript as the product model.
Value Created
Turns AI from an operations chatbot into a governed engineering capability that accelerates investigation without sacrificing governance.
The boundaries
Trade-offs
Automatic execution is intentionally less aggressive than unguarded automation in exchange for stronger explainability, reviewability and recovery confidence.
What Was Intentionally Excluded
AI as source of truth, policy bypass or mandatory dependency of the normal runtime is intentionally excluded.
Truth boundary
Current Truth / Limitations
AI is never the source of truth or a bypass around RBAC, policy or execution; the normal runtime remains operable without a model.
designed → source implemented → runtime tested → production certified
Each level is claimed separately; a lower level never implies a higher one.
Start a conversation
If technology should move your organization forward rather than hold it back, let’s talk.
The first conversation is about your problem, not about me: where you are stuck, what is at stake and whether I am the right person to solve it.
Bring the problem