Latest from Moring
Updated
Technical guidance, research and product updates to help you build and govern AI agents as enterprise adoption accelerates.
Observability for AI Agents: Metrics, Traces, and Guardrails That Matter
Metrics say whether an agent works, traces explain why it decided, guardrails bound what it may do, and all three write to one shared record.
Ashvath NarayanWhen an agent hands work back: designing the exception queue
Design an AI agent exception queue with clear owners, states, escalation timers and validation checks before work resumes.
Rollback is not always undo: recovering from agent actions
Learn what rollback does not undo in AI agents, how to recover side effects, and how to make retries safe with idempotency and compensation.
Latest blogs
When an agent hands work back: designing the exception queue
Design an AI agent exception queue with clear owners, states, escalation timers and validation checks before work resumes.
Rollback is not always undo: recovering from agent actions
Learn what rollback does not undo in AI agents, how to recover side effects, and how to make retries safe with idempotency and compensation.
Your AI agents are running without controls — here is what that costs you
Ungoverned AI agents cost more than risk: rework, shadow spend, stalled pilots, exam exposure. A cost inventory you can run in one day against your systems.
Balaji NagarajAI governance for the board: what directors need to know in 2026
A board-level briefing on AI agents: the fiduciary questions, five questions directors should ask management, and the red flags in AI governance reports.
Balaji NagarajAI governance frameworks, mapped: NIST, ISO 42001, EU AI Act, SR 26-2, DORA, CPS 230 and NAIC in one table
Seven AI governance frameworks compared side by side — NIST AI RMF, ISO 42001, EU AI Act, SR 26-2, DORA, CPS 230, NAIC — mapped to the controls each expects.
Balaji NagarajAI SRE vs AIOps vs agentic incident response: what actually changes?
AIOps, AI SRE and agentic incident response overlap in practice.
Arya SharanWhat FFIEC examiners are now asking about your AI agents
The questions FFIEC examiners now ask about AI agents — who approved what, what evidence exists, how agents are stopped — and the records that answer them.
Balaji NagarajMulti-agent governance: how to govern systems of agents, not one agent at a time
A framework for multi-agent governance: the four controls that scale from one agent to a fleet, what changes when agents delegate, and how to evidence it.
Balaji NagarajHow to prepare for a NAIC AI model audit
What a NAIC AI model audit asks: the governance program, the AI inventory, risk evaluations, and evidence insurers must produce — plus a prep sequence.
Balaji NagarajRoot-cause evidence agents: assemble the case before recommending a fix
How incident agents should gather source-linked evidence, test hypotheses, and preserve contradictions before proposing remediation.
Arya SharanThe 48-hour AI agent bake-off: how to evaluate agents on your own cases
How to run a 48-hour agent evaluation on one real workflow: pick the cases, define done, score accuracy and evidence, and decide with numbers instead of a demo.
Balaji NagarajAI agent governance for credit unions: what NCUA examiners will ask
Credit unions are deploying AI in lending, service and fraud. What governed deployment looks like at credit-union scale and what NCUA examiners ask.
Arya SharanAI agent governance for regional and mid-market banks
Regional and mid-market banks face the same agent questions as the majors without the platform teams. What proportionate governance looks like at $2–50B scale.
Balaji NagarajAI agents in Kubernetes incident response: what changes for platform teams
How AI agents change Kubernetes incident response: triage, correlation and remediation with graduated authority, stop conditions and evidence.
Balaji NagarajDORA compliance for AI agents: what EU financial entities need to evidence
DORA applies to AI agents acting inside EU financial entities — logging, ICT risk, resilience testing, third-party oversight. What to evidence.
Arya SharanEU AI Act for agentic AI in financial services: what applies and when
How the EU AI Act applies to AI agents in financial services: risk classification, record-keeping, human oversight, timeline, and compliant deployment.
Arya SharanForward-deployed engineering for AI agents: why the model beats consulting
FDE puts the engineers who build the agent inside the customer's environment. Why agents made this model necessary, and how it differs from consulting.
Balaji NagarajLangChain in the enterprise: what governance has to add
LangChain and LangGraph are how enterprises build agents. What the framework gives you, what it cannot decide, and the governance layer for production.
Balaji NagarajReplayable AI agent runs: what compliance needs from run records
A replayable agent run records every step: inputs, model calls, tool invocations, policy decisions, approvals. What compliance needs and how to build it.
Balaji NagarajWhy companies roll back AI agents: what the Sinch 74% figure actually teaches
Sinch reports 74% of companies have rolled back an AI agent deployment. Why rollbacks happen, what governed rollback looks like, and how to prevent them.
Balaji NagarajThe state of agent governance: 11 stats from 2026 research
Eleven data points on why enterprise AI agents stall between pilot and production — Gartner, Deloitte, BCG, with sources and dates included.
Balaji NagarajWhat Gartner's 40% agentic AI cancellation figure means for enterprise AI
Gartner projects over 40% of agentic AI projects will be canceled by 2027. The causes are governance failures, not model failures — and each has a fix.
Balaji NagarajAgentic Process Automation vs RPA: Governed Agents vs Scripts
Agentic process automation reasons where RPA scripts break. Before agents touch production processes they need identity, policy, approval and evidence.
Arya SharanAI Agent Audit Trails: Proving What Happened When Agents Take Action
How to build an AI agent audit trail a CISO and internal audit can trust: correlation IDs, tamper-evidence, clock sync, retention and NIST AU controls.
Arya SharanAI Coding Agent Governance: What to Control Before Agents Write Production Code
AI coding agent governance: one identity per agent, scoped repo and tool access, security checks on every pull request, manifest review and a human merge.
Arya SharanAML Investigation AI Agents: From Alert Triage to the SAR Decision
AML investigation AI agents assemble the evidence file, link related alerts and draft the narrative. The investigator decides; the SAR decision stays human.
Arya SharanHuman-in-the-Loop for AI Agents: How to Design an Approval Gate
How to design a human-in-the-loop approval gate for AI agents: which actions to gate, what reviewers see, timeouts and rejection reasons. A KYC worked example.
Arya SharanNAIC AI Bulletin Compliance: What Insurance Carriers Need to Evidence
The NAIC Model Bulletin on AI tells insurers that AI-supported decisions must follow existing insurance law. Here's what examiners will ask.
Arya SharanPayment Disputes and Fraud Agents: What Banking Teams Need to Govern
AI agents can investigate payment disputes and flag fraud faster. Here's how to govern that workflow for regulatory review.
Arya SharanPrior-Auth Routing Under CMS-0057-F: Where an AI Agent Helps a Payor Meet the Clock
AI prior authorization under CMS-0057-F: how a routing agent helps payors meet 72-hour and 7-day timeframes and give specific denial reasons.
Arya SharanAn AI Agent for Regulatory Reporting After SR 26-2: Who Owns the Evidence?
An AI agent for regulatory reporting sits outside SR 26-2. What it touched, which policy allowed it and who approved the report are the bank's to design.
Arya SharanRevenue Cycle AI Agents: Prepare and Track, Let Your Team Decide
Revenue cycle AI agents check prior auth status, assemble denial evidence and follow up on claims. Staff decide appeals, under HIPAA minimum necessary.
Arya SharanVibe Coding Enterprise Risk: The Problem Is Not That AI Writes Code
Vibe coding enterprise risk: unreviewed AI code, secrets in prompts, ungoverned tool access, no provenance. The fix is a record of what was allowed, not a ban.
Arya SharanShould you build or buy an AI control plane?
Compare building and buying an AI control plane layer by layer, including identity, policy, audit evidence, integration and ongoing maintenance costs.
Arya SharanDeploying AI agents in your own environment: what stays private, and what still leaves?
Deploy AI agents in your environment: map model, tool, identity and telemetry paths to verify data residency, retention and what leaves your boundary.
Arya SharanWhat makes an AI agent production grade?
Assess AI agent production readiness with six evidence checks covering identity, actions, policy, evaluations, stop controls and ownership.
Arya SharanGoverning AI coding agents that touch production code
Govern AI coding agents with dedicated identity, pull request boundaries, human merge approval and evidence that connects each request to release.
Arya SharanAI agent identity and permissions: whose access should a tool call use?
Choose delegated user authority, workload identity or scoped credentials for AI agent tool calls, and retain evidence of who acted for whom.
Arya SharanAI gateway, agent framework or control plane: which layer does your team need?
Compare AI gateways, agent frameworks and control planes. Follow one request to see where routing, tool execution, approvals and evidence belong.
Arya SharanWhy Enterprise AI Agents Stall Between Pilot and Production
Seven reasons enterprise AI agents stall after a successful pilot, from workflow scope and data access to ownership, permissions and production metrics.
Sahel RahmanWhat to Retain When an AI Agent Supports a Decision
A proposed seven-part record checklist for AI-assisted decisions, including human review boundaries and limits for activity outside routed controls.
Arya SharanWhy AI coding guidelines fail, and what habit-level evidence looks like
Evaluate AI coding habits against rollbacks and security findings before policy changes, then keep testing as tools and workflows evolve.
Ashvath NarayanPolicy servicing agents: handling endorsements, renewals and exceptions
How policy servicing agents prepare endorsements, renewals and exceptions, check effective dates, and route binding decisions to authorized reviewers.
Arya SharanWhen Should a Remediation Agent Stop and Hand Back to a Human?
Seven stop conditions for production remediation agents, how context-preserving handoff works, and why stopping new work doesn't cancel in-progress tasks.
SR 26-2 and Agentic AI: What Banking Teams Need to Know
SR 26-2 supersedes SR 11-7 but excludes agentic AI from scope. Here's what that gap means for your governance review, by workflow component.
Arya SharanFive Banking Workflows Worth Scoping for AI Agents
A practical scoping guide for banking teams: five AI agent workflows, the readiness checks to run, and the decisions your staff should keep.
Arya SharanAI Agents in Insurance: Five Workflows to Evaluate Before You Deploy
Claims, underwriting and servicing compete for investment. Each workflow asks something different of an AI agent.
Arya SharanAI agents for incident updates: explaining what happened while the investigation continues
Use timestamped evidence to draft incident updates, separate confirmed facts from unknowns, and review the exact message before sending it.
Arya SharanAI-DLC: what changes in your software delivery workflow?
AI-DLC changes how work moves from request through implementation and operation. Code generation is one part of that workflow.
Arya SharanWhat should you budget to run an AI agent in production?
Budget for model use, infrastructure, human review, retries and support. Measure operating cost against completed cases and keep launch costs separate.
Arya SharanGraduated authority for AIOps agents: recommendation, approval, and bounded execution
Decide when an AIOps agent can recommend a change, needs approval, or can act within a bounded grant, and what evidence each action should retain.
Arya SharanHow a KYC Evidence Agent Works in Commercial Onboarding
A KYC evidence agent compares applicant data, documents and registry records, surfaces conflicts to analysts, and retains the evidence behind human decisions.
Arya SharanHow should a bank use AI agents to investigate reconciliation breaks?
Scope a reconciliation agent to evidence: which causes it may name, what it must never do, what the controller receives, and what is kept for the next audit.
Arya SharanWho is accountable when an AI agent makes a bad decision?
Why "a human in the loop" is not an answer, and the four roles to record on every agent run: who initiated, which agent acted, who owns it, who approved.
Arya SharanWhat is AI agent orchestration, and how do you govern it from one place?
The three orchestration patterns enterprises deploy, why traditional governance tools fail at that layer, and five pillars for governing it from one place.
EllakkiaaEnterprise AI Architecture: Where an AI Control Plane Fits
Where an AI control plane sits in the enterprise stack, what it does on each agent action, how it relates to platforms you run, and whether to build or buy.
Ashvath NarayanAgentic AIOps: Who Governs the Ops Agents Now Acting on Their Own
Ops agents now restart services and roll back deploys on their own. The governance questions every deployment must answer, and where today's approaches break.
EllakkiaaObservability for AI Agents: Metrics, Traces, and Guardrails That Matter
Metrics say whether an agent works, traces explain why it decided, guardrails bound what it may do, and all three write to one shared record.
Ashvath NarayanFrom Pilot to Platform: A 90-Day Rollout Plan for Enterprise AI Agents
Why agent rollouts stall after the pilot, and a 90-day plan in three phases: standards as code, routing and secrets gating, then telemetry and a second team.
Balaji NagarajWhy MCP Exists: Connecting AI Agents to Tools and Enterprise Systems
Why a bigger context window doesn't solve live system access, how integration code outgrows the agent, and what MCP actually standardizes.
Vignesh NagarajanWhat an AI Control Plane Is, and the Three-Question Test for Evaluating One
What an AI control plane is in plain terms, and the Deny-Attribute-Replay test for evaluating one, plus where the tools you already own stop short.
Ashvath NarayanTaking MCP to Production: Tool Design, Identity, Gateways, and Governance
Take MCP from a local server to shared infrastructure with workflow-shaped tools, context budgets, on-demand loading, identity and governance.
Vignesh NagarajanAI-DLC vs AI-SDLC: What's the Difference?
AI-SDLC bolts AI onto the lifecycle; AI-DLC governs it. The seven-stage workflow spine, what a native audit trail contains, and how to measure maturity.
Balaji NagarajNo posts in this category yet.
The Enterprise AI Control Plane
A framework for governing agent identity, access, policy and evidence across enterprise workflows.
Balaji Nagaraj Kumar