/Podcast_cover (3).jpg)
Ambient AI Agents for Smarter SRE and Incident Response
Guest: Viraat Das — Co-founder & CEO, Exla
Your monitoring caught the incident. So why did it take engineers hours to find the cause? We talk with Viraat Das, co-founder and CEO of Exla, about Libra: an ambient AI agent that keeps investigating production systems, spots regressions traditional monitoring can miss, and proposes fixes.
The Enterprise Arc of Agent Sense
Each episode follows the enterprise journey from prototype to production operation — covering the layers that make agents reliable, governed, and safe to scale.
- Readiness: What breaks when agents leave demos
- Boundaries: Where rules, agents, and human judgment belong
- Data: Why bad data makes agents unsafe
- Data layer autonomy: Where autonomous action helps and where it creates risk
- Integration: How agents coordinate across enterprise systems
- Control: Why agents need gateways before reaching core systems
- Execution: What changes when agents move from chat to computer use
- Governed Scale: Why enterprises need catalogs for agents, tools, and MCP servers
- Conversation to Action: How AI moves from summarizing meetings to driving workflow execution
- Ambient Agents: What changes when agents shift from on-demand to always-on, event-driven operation
- Agent Memory: Why RAG alone isn't enough and what a real enterprise memory system — with ingestion, distillation, and graph relationships — actually looks like
- Agent Performance: Why accuracy alone is misleading and how to measure what agents actually deliver — task completion, latency, cost, and user outcomes — and trace failures to the right layer
- Agent ROI: Why successful demos don't guarantee business value and how to define measurable outcomes, map people and systems, and test production readiness before building the agent
- Incident Response: How ambient agents move from detecting an incident to finding its root cause, and where a human approves the fix
Agent Sense Hosts
Co-Host & Creator
Enterprise AI technical practitioner focused on moving agents from demos into governed operating models. Her work spans architecture, data, workflow automation, and adoption across healthcare, financial services, energy, and IBM software. Agent Sense reflects her field perspective on what it takes to make enterprise AI reliable in production.
Driving business outcomes one practical tip a day.
Co-Host & Practitioner
AI technical architect with deep experience in automation, software testing, and enterprise AI implementation. He has led generative AI initiatives and built solutions across SaaS, RPA, and AI-driven applications — bringing a builder and delivery lens to what makes agents testable, scalable, and ready for real business work.
Podcast Episodes
Brains and Guardrails: What Makes an AI Agent Enterprise-Ready?
Most enterprise AI pilots look impressive until they touch real systems, data, and accountability. What separates a prototype from a production-ready agent? Trust, guardrails, and the design decisions enterprises skip too early.
Rules, Agents, Humans — A Practical Model for Agentic Workflows
Not every decision should go to an agent. Using a banking onboarding scenario, this episode maps out when deterministic rules are the right answer, when agents add value, and where human judgment must stay in the loop.
Why IT Service Agents Fail in Production — A Data Readiness Problem
Agent failures in production are usually data failures, not model failures. Using an IT service ticketing example, this episode shows how incomplete ownership records and stale configuration data cause agents to act confidently on wrong information.
Autonomous Databases — Where Autonomy Helps and Where It Hurts
Autonomous databases can self-tune, self-heal, and self-secure — but autonomy at the data layer has limits. Where does it genuinely reduce operational risk, and where does it create new problems?
Integration Will Decide Enterprise AI — MCP and Agent-to-Agent
When agents need to reach enterprise systems and coordinate with each other, integration becomes the bottleneck. MCP standardizes agent-to-system connections. Agent-to-Agent (A2A) protocols let work move across the enterprise without custom wiring for every handoff.
.jpg)
MCP Gateway: The Control Layer for Enterprise Agents
Connecting agents to enterprise systems creates a new exposure: unmanaged access across HR, finance, IT, and operations. This episode covers how an MCP Gateway sits between agents and core systems, enforcing identity, policy, approvals, and audit trails before any action reaches production data.
The Evolution of Agents — From Chat to Computer Use
Agents are no longer just answering questions — they are navigating interfaces, executing tasks, and operating across runtime environments. What changes architecturally when agents move from chat to computer use, and what safety and observability requirements follow.
Governed Catalog of Agentic AI Assets
As AI scales past the pilot phase, teams build agents and tools in isolation — creating sprawl, duplicate builds, and unclear ownership. A governed catalog gives enterprises visibility, reuse, and trust signals before agents reach production.
Beyond Meeting Summaries: From Conversation to Action
AI can summarize a meeting. Enterprise work starts after the summary. Artem Koren of Sembly AI explains how conversation intelligence moves beyond documentation into real workflow execution — CRM updates, follow-ups, task assignments, and cross-team actions with human judgment in control.
The Ambient Agent Pattern: AI That Works While You Sleep
Ambient agents don't wait to be asked. They run in the background — monitoring signals, detecting patterns, and taking action when conditions are met, with human judgment defining the boundaries of autonomy. This episode breaks down the ambient agent pattern and what it means to move enterprise AI from on-demand to always-on.
/Season2_Episode1.jpg)
Agent Memory Is More Than RAG
Agents sit on massive amounts of enterprise data — but still ask users to repeat themselves. RAG alone doesn't fix this. Ran Aroussi walks through what a real agent memory system looks like: ingestion, distillation, graph relationships, and retrieval that reasons over context — not just similarity.
Agent Performance, Beyond Accuracy
An AI agent can be 95% accurate and still fail the business. It can give the right answer and still take too long, use the wrong tool, cost too much, or fail to complete the task. We talk with Jay Migliaccio, about how enterprises should measure and improve agent performance.
/Podcast_cover (9).png)
The Missing Path From AI Agents to ROI
Your AI agent works, but can you prove its business value? A successful demo can create confidence that doesn't carry through to production when requirements are unclear, dependencies are missed, and nobody has agreed on how success is measured. We talk with Robert Maxwell about what enterprises need to address before scaling AI agents.
Ambient AI Agents for Smarter SRE and Incident Response
Your monitoring caught the incident. So why did it take engineers hours to find the cause? We talk with Viraat Das, co-founder and CEO of Exla, about Libra: an ambient AI agent that keeps investigating production systems, spots regressions traditional monitoring can miss, and proposes fixes.
Guests
Practitioners and builders who have joined Agent Sense to share their field experience on enterprise AI.
Vitalii Duk
Founder & CEO, Dynamiq
Vitalii founded Dynamiq to make agentic AI practical and secure. He knows what it takes to move agents from impressive demos to reliable, observable production systems.
S1 E7 · Chat to Computer Use
Jyotsna K Narayanan
Technical Product Manager, IBM
Jyotsna leads product management for the Domain and Agent Catalog in IBM watsonx Orchestrate. Her focus is governed reuse, so enterprises scale agents instead of rebuilding them pilot by pilot.
S1 E8 · Governed Catalog of Agentic AI Assets
Artem Koren
Co-founder & Chief Product Officer, Sembly AI
Artem co-founded Sembly AI to turn meeting conversations into action: follow-ups, CRM updates and system tasks, with people still steering the outcome.
S1 E9 · From Conversation to Action/RanAroussiPhoto.png){kind=avatar}
Ran Aroussi
Founder, MUXI
Ran has built production systems for more than 30 years. He created yfinance, the most widely used open source financial data library, and founded MUXI, open infrastructure for running agents in production.
S2 E1 · Agent Memory Is More Than RAG
Jay Migliaccio
Technical Product Manager, IBM
Jay builds agentic solutions for enterprise customers on IBM watsonx Orchestrate. He judges agents by their full trajectory: the decisions, the tool calls, and whether the task got done.
S2 E2 · Agent Performance, Beyond Accuracy/RoberMaxwell_ProfilePicture.jpeg){kind=avatar}
Robert Maxwell
Founder & CEO, TGT Solutions
Robert founded TGT Solutions in 1996. He studied computer science and mathematics, holds an MBA and is a Chartered Accountant, and focuses on tying business goals to technology decisions.
S2 E3 · The Missing Path From AI Agents to ROI/Viraat_Das_Headshot.png){kind=avatar}
Viraat Das
Co-founder & CEO, Exla
Viraat co-founded Exla, a Y Combinator backed company building Libra: an ambient AI agent that catches production regressions and proposes code fixes. He was previously a machine learning engineer at Amazon.
S2 E4 · Ambient AI Agents for Smarter SRE