Join Subsurface LIVE 2025, How Agents Succeed? Commvault SHIFT was a blast

Hi Data & AI Pros,

If you work with data, this is one event you won’t want to miss! Subsurface LIVE 2025 is the world’s leading virtual conference dedicated to the open data lakehouse ecosystem — and it’s completely free to attend.

December 10, 2025
8AM–2PM PT | 4PM–10PM GMT | 5PM–11PM CET
Virtual | Free to Attend

Join thousands of data professionals from across the globe to explore how the world’s top organizations are advancing open architectures, real-world lakehouse implementations, and the latest innovations in data engineering.

Here’s what’s in store:

Keynote: Enabling the Agentic Enterprise

35+ breakout sessions across 5 tracks from the open source community - including deep dives into Apache Iceberg, Arrow, and Polaris

Panel: The Future of the Lakehouse OSS — featuring leaders from AWS, Oracle, and JP Morgan Chase

Technical sessions from Dremio experts — covering performance gains, architecture insights, and hybrid lakehouse strategies

You’ll hear from data pioneers at Bauplan, Best Buy, Capital Group, JP Morgan Chase, and the creators behind Apache Arrow, Iceberg, and more.

Whether you’re a data engineer, architect, analytics or data leader, Subsurface LIVE 2025 is your chance to learn, connect, and get inspired by the best minds in the industry.

Agents succeed when design is clear and the loop never stops learning.
I broke down a five step lifecycle that takes an agent from input to business impact. Use this as a blueprint for any real workflow.

What the lifecycle covers, in plain terms

Perception
• Goal: collect the right signals from data, apps, and context
• Inputs: customer text, logs, transactions, documents, sensor feeds
• Pitfalls: noisy data, missing context, stale information
• Fix: validate inputs, normalize formats, add basic quality checks

Cognition
• Goal: understand the situation and generate options
• Tools: search, retrieval, analytics, rules, and models
• Pitfalls: hallucinations, outdated facts, weak retrieval
• Fix: ground answers in approved sources, cache facts, show evidence

Decision
• Goal: choose an action you can explain and defend
• Tools: business rules, scorecards, thresholds, human checkpoints
• Pitfalls: hidden logic, conflicting rules, no audit trail
• Fix: keep the policy in one place, log every rule hit, attach rationale

Action
• Goal: execute safely in production systems
• Examples: send an offer, create a ticket, file a claim note, trigger a workflow
• Pitfalls: partial writes, timeouts, no rollback
• Fix: make actions idempotent, confirm side effects, add compensating steps

Learning
• Goal: improve the next decision with evidence
• What to capture: input, chosen action, outcome, time, cost, quality
• Pitfalls: vanity metrics, slow feedback, no ownership
• Fix: define success metrics, review weekly, assign owners to fixes

How leaders can apply this today
• Map one workflow to the five steps. Mark green where you are strong and red where you are blind
• Place human review only where risk is real and outcomes are hard to reverse
• Separate policy from prompts so you can change decisions without rewriting the agent
• Treat logs and evidence as first class data, not leftovers

KPIs that keep you honest
• Decision quality: precision, recall, error rate, and dispute rate
• Speed: median time from input to action
• Safety: rollback count, blocked actions, policy violations
• Value: conversion lift, cost per resolution, time to value

My take
Start with one use case. Write the policy for the decision step first. Wire perception and cognition to feed that policy. Add safe actions. Close the loop with learning. Review outcomes every week. When the loop is stable, scale to the next workflow.

Most companies think they have a backup problem. What they really have is a recovery problem. I spent the day at Commvault SHIFT, on the ground talking to customers, partners and the Commvault team. I also spoke with Darren Thomson, Field CTO at Commvault, for a focused chat on what they announced and what it means in the middle of real attacks and real pressure on IT teams.

🔍 Stay Ahead in AI & Data! Join 137K+ Data & AI professionals who stay updated with the latest trends, insights, and innovations.

📢 Want to sponsor or support this newsletter? Reach out and let's collaborate! 🚀

Best,

Ravit Jain

Founder & Host of The Ravit Show