Insights

Deep thinking on platform engineering, MRO data governance, and building software that lasts.

The Identity MRO Never Had

By Raghu Vishwanath, Managing Partner | August 2026 | 8 min read

In MRO master data, “identity” was always a description.

A line of text, typed by a human. Match the text, find the duplicate. Every deduplication tool ever pointed at an MRO catalog worked this way — and it was always the wrong model, because the description is the one field you cannot trust.

Two moments from a single afternoon make the point.

The Grounding Problem

On the layer between intelligence and the world.

A lone cloud-to-ground lightning bolt striking empty plains under a dark sky, the circuit closing between an enormous charge and the earth — potential finally doing work once it finds its path to ground.

The Open Circuit

Stack the voltage as high as you like — it does no work until it finds a path to ground....

Dark storm clouds gathering over open plains with a low horizon and no lightning, illustrating potential and ground that are not yet connected — the open circuit before the strike.

The Grounding Problem

A model can pass every exam and still mishandle one real file. The bottleneck was never intelligence — it's grounding,...

What Good Looks Like

The standard for master data, before the tools define it for you.

A macro of a green leaf's vein network — a central midrib branching into ever-finer veins that reach every cell, a metaphor for asset data where a single correct classification connects the asset to its failure modes and spare parts.

What Good Looks Like for Asset Master Data

An asset record can read "ROTATING" — perfectly classified, and useless. It looks governed and connects to nothing. Good asset...

A dense fish bait ball — thousands of individuals resolving into one coherent body underwater, a metaphor for vendor data where every transaction across sites and spellings resolves to one supplier identity.

What Good Looks Like for Vendor Master Data

A vendor master never fails loudly. One supplier becomes three strangers, leverage leaks out of a gap nobody can see,...

A fill-frame honeycomb in natural backlit tones from deep red through amber to gold — densely packed uniform cells, each the same governed shape and complete, none duplicated, a metaphor for item master data where each thing exists once and means the same to everyone.

What Good Looks Like for Item Master Data

A warehouse holds the same bearing in five places, and nobody knows until all five run out at once. Everyone...

The Layer They Cannot Build

Why MRO master data governance can't be built from inside the incumbents' platforms.

The Identity MRO Never Had

In MRO master data, "identity" was always a description — the one field you cannot trust. Ark reads the part's...

Braided glacial river channels from above, Iceland — copper sediment and green vegetation cut by branching cyan meltwater streams radiating outward — hero image for "What the Layer They Cannot Build Actually Looks Like,"

What the Layer They Cannot Build Actually Looks Like

The first two articles in this series argued what the layer cannot be. It cannot be built inside SAP or...

Why the Data Layer Got Outsourced — and Stayed There

The first MRO master data remediation projects were sold around 2003. Twenty-three years later, the same firms are selling the...

Enterprise Data Engineering

The unglamorous engineering that makes AI agents work.

The Virgin River flowing through Wall Street in Zion's Narrows — water finding its direct path between towering sandstone walls, moving at the pace the terrain demands.

The 90-Day Agentic Readiness Sprint

The board wants agentic AI progress by Q3. The SI sends a team. Fourteen weeks later, nobody has looked at...

Vivid colour-banded mountain ridges at Zhangye Danxia, China — each geological layer distinct and visible, with a lone figure standing on the ridge providing scale against the vast layered landscape.

The Three Layers Between Your AI Agent and Your ERP

McKinsey called ERP "the ugly stepchild" of the AI conversation. This article maps the architecture of what agentic-ready actually looks...

Aerial view of a massive Antarctic iceberg — the visible surface impressive, the submerged structure beneath the turquoise water vastly larger.

Agentic AI Won’t Fail Because of AI

42% of AI initiatives failed in 2025. Not because of the models — because of the data infrastructure beneath them....

Builder's Log

Notes from 30 years of building software that lasts.

The swirling, weathered grain of an ancient twisted tree trunk filling the frame — endurance recorded in the wood of a tree that survived where others did not, a metaphor for staying with a platform long after the industry norm is to move on.

Why We Stay

Most software relationships end at go-live. We built the company around the opposite bet. Fifteen years in, two platforms deep,...

The Shape We Wouldn’t Force

A predefined model ships fast — until the second category of data arrives, and the third. We refused to build...

The Revenue We Refused

The remediation contracts were there for the taking — large, repeatable, predictable. We stopped pitching them. Not because the work...

Two parallel bridges stretching over a foggy river — representing the deliberate decision to build architectural independence by constructing the same system twice on different foundations

Why I Built the Same Pipeline Twice

The original pipeline worked. The cost was fine. Nobody asked for a rebuild. But depending on a single provider for...

A single diamond held in sharp focus by jeweler's tweezers, with many similar diamonds blurred in the background — representing how retrieval architecture selects the right context from thousands of options before an AI model ever sees it

The Retrieval Layer Nobody Talks About

Everyone debates which AI model to use. Almost nobody talks about the retrieval layer that determines what the model actually...

Hands carefully pruning a bonsai tree on a work table — deliberate constraint shaping growth into something more refined than unconstrained nature would produce.

The 8K Constraint That Made Everything Better

A generous context window lets you be lazy and still get results. An 8K token limit forced a 94% reduction...

AI and MRO

What we learned building AI that understands MRO data

Blue marble caves in Patagonia with light selectively illuminating sculpted surfaces while deeper formations remain in shadow — AI that reveals what it can determine and honestly acknowledges what it can't

AI That Knows When It Doesn’t Know

"PART, MISC, FOR PUMP." Every AI vendor classifies this record confidently. That's the problem. The most important thing your AI...

Master luthier's hands carving the f-hole of a violin — precision craftsmanship representing generational domain expertise that no AI model can replicate, the knowledge that makes MRO data classification accurate

When AI Meets Three Decades of MRO Knowledge

Same bearing. Same valve. Same gasket. Different intelligence. When AI is built by people who understand what a bearing does...

Desert mirage shimmering over a dry salt flat — a convincing illusion of water on the horizon that doesn't exist, like generic AI producing confident but fabricated MRO data classifications

Why Generic AI Fails at MRO Data

Your AI vendor's demo was flawless. Then someone asked: "What happens when we feed it our actual data?" We fed...

The Layers — MRO

Lessons from living inside the MRO data problem

tree and sunrise over mountain in nature

What “Good Enough” Data Actually Looks Like

"We tried the governance route. It didn't stick." I've heard this from more MRO leaders than I can count. They...

$drop of water falling into water$

The Math That Maintenance Managers Don’t See

The builder who asks "why" isn't a new invention. Leonardo was one. The cathedral masters were. The architects of Uber's...

painted hills detail, john day fossil beds national monument, oregon, usa

The Layers Don’t Lie

There's a moment in every long project when you realize you've stopped building what you set out to build. We...

Bluemind Thinking

Vision, philosophy, and the future of platform engineering

The Builder Who Asks Why

The builder who asks "why" isn't a new invention. Leonardo was one. The cathedral masters were. The architects of Uber's...

Where Judgment Lives in Code

Philosophy is cheap. Everyone claims to ask "why." Every pitch deck mentions vision. Every engineering team says they care about...

What the GP Sees That the Surgeon Misses

The patient arrives with a persistent rash. The dermatologist examines it carefully. Considers the pattern, the texture, the distribution. Autoimmune...

Why AI Makes a Perfect Cook (But Never a Chef)

A cook follows recipes. A chef creates them. This isn't a hierarchy of skill. It's a difference in kind. The...

aerial view of a vibrant river delta with intricate patterns

The Platforms That Actually Work

Uber owns no cars. Airbnb owns no hotels. Alibaba owns no inventory. The most valuable platforms in the world own...

What Happens When Building Costs Nothing?

Five years ago, that feature would have taken a quarter. Your team would have scoped it, debated it, staffed it....

assembly line for cars, ford motor company, usa. date: circa 1920s

The Deal That’s Expiring

Depth was your moat. AI is draining it. Your grandfather could learn a trade at eighteen and retire with a...

What Did Leonardo Know That We Forgot?

Leonardo da Vinci designed war machines on Monday and painted the Mona Lisa on Tuesday. He studied anatomy to understand...

Why We Ask “Why”

Five years ago, we took on an engagement that most product shops would have passed on — building a Super...

Insights

Deep thinking on platform engineering, MRO data governance, and building software that lasts.

The Grounding Problem

On the layer between intelligence and the world.

What Good Looks Like

The standard for master data, before the tools define it for you.

The Layer They Cannot Build

Why MRO master data governance can't be built from inside the incumbents' platforms.

Enterprise Data Engineering

The unglamorous engineering that makes AI agents work.

Builder's Log

Notes from 30 years of building software that lasts.

AI and MRO

What we learned building AI that understands MRO data

The Layers — MRO

Lessons from living inside the MRO data problem

Bluemind Thinking

Vision, philosophy, and the future of platform engineering

Ark Series

Data governance, MRO operations, and building foundations that last