Adam V.

We’ve all seen impressive agent demos… and then watched them stall in pilot mode. What I like about this playbook is that it focuses on the unsexy but critical parts: governance, lifecycle management, cost discipline, and proving ROI — not just cool model tricks. If you’re building agentic AI for real business workflows, this is a strong operational guide from Day 0 through Day 90 and beyond. #AgenticAI #AIInProduction #EnterpriseAI #AIGovernance #MLOps #AIOperations #DigitalWorkforce #FromPilotToProduction #ResponsibleAI

4

TechPulse #15: Databricks Mastery Series - Day 6 🧵 💰 "We need to cut our compute bill by 50% before next board meeting." My manager at Warner Bros Discovery dropped this on me last year. Our data team was burning through $12K/month on r5.2xlarge clusters. What if I told you one checkbox could cut that to $6K? 🎯 What Are Spot Instances? Think of them like flying standby - you get 70% off regular price, but might occasionally get bumped. AWS sells spare EC2 capacity at massive discounts. Databricks lets you easily configure your clusters to use these spot instances in your VPC. The catch? AWS can reclaim them with 2-minute notice if someone pays full price. ⚡ How to Enable (Using Databricks to Configure Your VPC): 1) Go to "Compute" in your Databricks workspace 2) Click "Create compute" 3) Scroll to "Advanced options" 4) Under "Advanced performance" section 5) Check "Use spot instances" ✅ 6) Databricks configures AWS spot instances in your VPC automatically 7) Driver stays on-demand, workers use spot (smart default!) 💸 Real Numbers (r5.2xlarge in us-east-1) • On-demand: $0.504/hour = $12,096/month ✅ • Spot average: ~$0.15/hour = $3,600/month • Savings: $8,500/month = $102K/year 🛡️ Perfect For: • ETL jobs that can restart • Model training (with checkpointing) • Development/staging environments • Batch processing workloads ❌ Avoid For: • Real-time dashboards • Interactive analysis during business hours • Mission-critical production jobs without fallbacks 📊 My 2-Year WBD Experience: • Total interruptions: 47 (mostly dev clusters) • Jobs that actually failed: 3 (6.4%) • Average downtime: 8 minutes • Money saved: $102K/year Pro tip: Databricks automatically handles AWS spot instance management - if spots can't be acquired in your VPC, on-demand EC2 instances are deployed as fallback, so your jobs complete either way. ✅ Quick Start Checklist: • Enable spot instances ✓ • Driver stays on-demand (automatic) ✓ • Add checkpointing to long-running jobs ✓ • Monitor cluster metrics ✓ • Start with non-critical workloads ✓ Here's the real question: If you could save $100K/year with 30 seconds of setup and 6% risk of minor delays... what's really holding you back? Fear of change? Lack of trust in the technology? Be honest - I want to understand the psychological barriers here. Drop your real concerns below 👇 💡 Tomorrow: TechPulse #16 continues with Day 7! "The auto-scaling trap costing you $3K/month" - Why bigger isn't always better #TechPulse #DatabricksMastery #CostOptimization #DataEngineering #AWS Databricks AWS Ali Ghodsi

2

1 Comment

Today we’re releasing the Pink Paper. Here’s the cleanest signal in the data. When you compare Hermès and LVMH across nearly 20 years, the divergence isn’t brand positioning or demand mix. It’s structural volatility. One metric captures it clearly: Operating Margin Volatility (YoY): • Hermès: ~7% • LVMH: ~14% In other words, LVMH’s operating margins fluctuate at roughly twice the amplitude. This matters because margin volatility reflects how a business absorbs: • Cost shocks • Demand cycles • Capital allocation decisions Lower volatility indicates tighter internal coordination between pricing, production, and growth pacing. Higher volatility reflects exposure introduced by acquisitions, portfolio breadth, and integration cycles. Across margins, cash conversion, long-horizon CAGR, productivity, and volatility, the data consistently points to the same conclusion: Hermès compounds coherence. LVMH compounds scale. The Pink Paper examines why only one of these models produces durable financial stability over cycles.

5

1 Comment

I went down a bit of a rabbit hole on AI spreadsheet ingestion this month, as I continue improving our data pipelines at Valae. There’s a lot of research and products in this space that don’t match what we see inside companies every day. Papers like Microsoft Research’s SpreadsheetLLM achieve high metrics for table extraction, but there’s always a catch. The success is measured on spreadsheets where the data is already pretty clear. Real company spreadsheets aren’t so easy. Sometimes they have three tables crammed into one tab, column headers that are someone's initials, and cell B7 containing "check with Marco re: pricing (NOT final)". In my research, I also dug into the survey "Toward Real-World Table Agents" (Tian et al., 2025) which confirms what I found after reading dozens of these papers: almost all LLM table research uses clean academic datasets. The messy reality of business data is barely ever addressed. That gap is where my team at Valae lives. We're perfecting proprietary processes for turning the messiest possible spreadsheets into clean, machine-readable knowledge, automatically. Despite SpreadsheetLLM, despite LlamaIndex launching LlamaSheets, despite years of research… nothing on the market truly handles the gap between what a spreadsheet contains and what it means. Encoding is a solved-ish problem. Interpretation isn't. And that's the one that matters. https://lnkd.in/er7PdDBP

16

1 Comment