Insights
Ideas on AI-native operating models, expanded into full articles from what first started as LinkedIn posts.
Incident Avoidance Has an ROI Problem, Not a Technology Problem
Microsoft’s November 12 outage wasn’t caused by legacy technology. It was caused by an operating model that detects and recovers instead of avoiding, and by an ROI model that can’t see an incident that never happened.
Teaching AI to Unlearn: Why Elimination Starts With Questioning How We Work Today
AI trained on history repeats history. To eliminate work, it first has to unlearn the assumption that the work is necessary.
The Restart That Shouldn’t Have Happened
Decision models don't explain their answers. In enterprise operations, the fix isn't better prompts. It's training the why, verifying state before deciding, and a harness that lets humans see what a decision rested on.
The Context to Eliminate
Snowflake's Business Ontology gives a data platform the top-down context to eliminate work, not just automate it. Why the platforms won't lead with elimination, and what enterprises should do instead.
The PR That Ships Without Comments
The LLM's power is reasoning to a decision, not faster code. The metric that matters is a PR that ships with no review comments, and it starts with setting the objective.
Automate or Augment? Both Skip the First Question
Every major AI-and-jobs study asks whether AI will automate or augment work. None asks whether the work should exist at all. Digital banks show what elimination looks like, and how incumbents can get there with historical and market intelligence.
The Missing Objective: Why AI Keeps Automating Work That Shouldn't Exist
Every LLM leap came from changing what we train models to do. One objective is still missing: elimination. Why enterprises need an elimination office.
Good Is Not Good Enough: The Intelligence Hiding in Your Own Data
Fintech turned business processes into technology services. The next, bigger conversion turns business judgment into intelligence, and most enterprises already hold the playbook in their own historical data.
The Forecast Is Free Now. Your History Isn't.
Time series foundation models have made forecasting a commodity across IT and OT. The differentiator is historical intelligence: business curves, the text behind them, and the infrastructure beneath them, read together.
The Science Fiction Moment Always Arrives Early
From a CAD/CAM-driven lathe in the 1990s, to real-time business process anomaly detection, to NOC-less operations today, every wave of transformation looks like science fiction until the operating model catches up. Physical AI will be no different.
The Model Knows What a Rain Monitor Is. It Doesn't Know Why You Want One.
LLMs understand the words in a request but not the intent, context and specificity behind it. Why enterprise intelligence has to be trained in, not rented.
We Built Our Data Lake Backwards
Enterprises don't need a smarter LLM operating system or a bigger data lake. They need purposeful intelligence — a knowledge graph and vector store scoped to business outcome questions, with a lake where data earns its seat.
Vendors and Service Providers Are Sitting on Decades of Intelligence. Why Are They Still Selling Licenses and Headcount?
Product vendors and MSPs hold decades of operational history. Used as intelligence, it could replace due-diligence-led onboarding and technology-driven transformation, and finally make outcome-based pricing real.
Generative AI Isn't Over. Enterprise Intelligence Hasn't Started.
The hype cycle may be ending, but the real work hasn't begun: building organization-specific intelligence from an enterprise's own history, with the right model for the right job.
Most Enterprise AI Decisions Need a Label, Not an Essay — and Not Just From Text
Cheaper decision models are progress. Decisions that graduate out of models entirely, and work that should never exist, are where the real returns are.
Stop Pricing Effort: Value-Based Contracts for the AI-Native Era
IT services revenue is deflating while tech spend booms. Outcome pricing on IT metrics formalises that deflation. Pricing business KPIs is how CIOs and service providers both win.
We've Seen This Movie Before: Thirty Years of Bolting On Instead of Rebuilding
Every major computing wave of the last three decades has been met with the same response — bolt it onto what already exists. AI won't be the exception unless we choose to make it one.
AI-Native Operations: When Growth No Longer Requires Proportional Headcount
The biggest opportunity from AI may not be removing labor. It may be breaking the assumption that business growth requires human capacity to grow at the same rate.
Distributed Computing, Not Bigger Models: What Enterprise AI Actually Needs
Enterprise AI is quietly re-learning the lessons of distributed computing, one expensive GPU bill at a time. Why size, structure, and time — not a bigger model — is the real architecture question most enterprise AI conversations still skip.
Building Intelligence in Physical AI: A Reference Architecture for Predict, Prevent, Eliminate
A Forbes piece on industrial AI as a decision layer above control, and my own predict/prevent/eliminate framing, both stop one layer short. Here's the retrieval architecture underneath: ground truth by layer, from sensor-confirmed to twin-simulated to human-adjudicated.
The Best Reasoning Model Still Does Not Know How Your Enterprise Should Think
Reasoning models keep getting more capable. But capacity to reason is not a reasoning objective — without proprietary context and an elimination-led framework, even the best model just automates the existing process faster.
Eliminate, Predict, Prevent: What Physical AI Should Actually Be Asked
Most physical AI and IT/OT convergence pitches are built to run an existing workflow faster. The harder question is whether the workflow should exist at all — and predict, prevent, and eliminate are three different problems with three different feedback loops.
The Enterprise Doesn't Have a Chief Intelligence Officer Yet — It Needs One
The tooling to build enterprise intelligence is finally cheap enough. What's missing isn't capability — it's someone accountable for building it. Here's the role, and the mandate it needs.
The Intelligent Enterprise Isn't a Dashboard — It's a Conversation
Most AI-native enterprise efforts today are dashboards with a chat interface bolted on, or task automation with no diagnosis underneath it. Neither is intelligence. Here's what actually is — and why it changes how enterprises should plan, budget, and prioritize.
Consumption Isn't the Point
The industry keeps measuring AI by what it costs to run. That's the wrong instrument for what AI is actually for.
Peak Business Performance Starts With What You Eliminate
The enterprise IT function exists to enable peak business outcomes. But the clusters that block those outcomes have been sitting on the transformation backlog for years — visible, labeled, and never prioritized, while budget moves to agents that babysit the work instead.
Intent Architecture: Why AI Adoption Keeps Producing Capability Without Outcomes
OpenAI's own enterprise ChatGPT data found no statistically significant link between messages or tokens per employee and revenue per employee. AIOps taught the same lesson a decade ago. Why AI transformation has to start with a measurable business outcome, not a technology initiative.
Human Assurance in Practice: Turning Ticket Patterns Into a Board-Ready Cost Take-Out
Pulled from current-state intelligence run across three operations portfolios: Eliminate and Automate consistently capture the majority of ticket volume, MTTR compresses 85-96%, and ticket volume correlates directly with the business KPI leadership actually tracks.
AI-Native IT Operations: A Reference Architecture for Guaranteed Outcomes per Token Invested
Most transformations start with a vendor, a model, or an agent library — not an architecture. Why current-state intelligence has to define future-state objectives, elimination-led execution replaces labor, and SME assurance closes the loop back to intelligence.
What Autonomous Vehicle Remote Assist Teaches Us About Human Assurance in IT Operations
An Operational Design Domain defines where a system is trusted to act alone. Remote assist defines what happens at the edge of it — a human resolving one focused question, not taking over the wheel.
What Autonomous Vehicles Can Teach Us About Autonomous IT Operations
Operational design domains, sensor fusion, calibration, and engineered safety — the principles that make self-driving cars trustworthy are the same ones enterprises need for autonomous IT operations.
The Two Moments That Make Open-Weight Model Adoption Inevitable
You can start training your own open-weight models anytime. But an M&A integration and a new IT vendor RFP each put real operating data and a mandate to change on the table at once — making the investment case for you.
Own the Model, Own the IP: Why "Renting" AI Is Becoming the Riskier Bet
CNBC's report on enterprises shifting to open-weight AI models isn't really a story about cost. It's about turning a rented capability into owned IP — and why elimination-first thinking has to come before self-hosting pays off.
Enterprise AI Is Entering the Age of Governed Digital Labor
ServiceNow's latest moves confirm what's next for enterprise AI: agents doing real work inside governed workflows. But governance alone doesn't transform an operating model — elimination-first thinking does.
The AI Usage Bill Has Arrived. Most IT Organizations Aren't Built to Pay It.
Microsoft's pullback on AI coding tools isn't a story about AI falling short. It's a story about metering, and almost no enterprise has built one.
IT Operations Doesn't Need Forward-Deployed Engineers. It Needs to Build the Ones It Already Has.
The forward-deployed engineering wave validates that implementation determines AI payoff — but its work stops short of IT operations. Why AI-enabled SREs, not imported FDEs, are positioned to eliminate operational workload rather than automate it.
How AI-Native IT Operations Enable Peak Business Performance
The intelligence to run the business better is already sitting in ticket queues and event logs. A four-step sequence — eliminate, empower, harden, automate — turns operational data into peak business performance.
Why Efficiency Is the One Metric Legacy Operating Models Can't Price
AI-native platforms create value as avoided cost and avoided effort, not new billable hours. Why legacy operating models don't know how to price the thing they asked for.
Is Your AI Learning Yesterday's Operating Model?
Historical incident data teaches AI more than solutions — it teaches the constraints of the operating model that produced it. Why platform context has to come before reasoning.
Why Are We Still Building Ticket Factories?
Challenges the industry's assumption that AI should simply make engineers faster. Organizations should eliminate unnecessary operational demand before automating what remains.
Platform Intelligence Before AI Reasoning
AI agents become significantly more effective when they first understand the operational platform they're interacting with — reasoning from platform intelligence, not just symptoms.
Administrative vs Operational Demand
Enterprises have eliminated millions of administrative service requests while still accepting large volumes of avoidable operational work — the next frontier for AI-native operations.
AI Doesn't Replace Engineers
AI should augment engineering expertise by eliminating repetitive operational work — freeing engineers for higher-value innovation, architecture, and business transformation.
Moving Beyond Labor-Centric Operating Models
The transition from labor-based IT services toward AI-native operating models built around digital labor, autonomous operations, and measurable business outcomes.
More of these publish regularly, expanded from ideas first shared on LinkedIn.