Why Your AI Tools Aren't Making Your Company Smarter
And what the 1890s textile mills can teach us about it.
By Alexander Lukianchuk · 8 min read
Every company we talk to has employees using ChatGPT and Claude. Many of them are genuinely more productive. They write faster, research faster, build spreadsheets faster.
And yet — almost none of these companies have become meaningfully more valuable as a result.
A recent essay by George Sivulka on a16z — "Institutional AI vs Individual AI" — frames this problem with an analogy we keep coming back to: the electrification of American textile mills.
In the 1890s, factories replaced steam engines with electric motors. Same floor plan, same machines, same workers. Output barely changed. It took thirty years before manufacturers redesigned the factories themselves — new layouts, new workflows, new roles — and only then did electrification deliver the returns everyone expected.
AI is in the same phase right now. Companies have swapped the motor but haven't redesigned the factory.
We think about this every day. It's the core of what Reformance does. And Sivulka's seven-part framework gives us a useful structure for explaining why individual AI adoption isn't enough — and what the alternative looks like.
1. The Coordination Problem
When every employee has their own AI habits — their own prompting styles, their own outputs, their own workflows — you get faster individuals rowing in different directions. That's not productivity. That's sophisticated chaos.
The missing piece isn't more AI tools. It's a shared foundation of institutional knowledge — structured, governed, accessible — that aligns how everyone (and every AI agent) in the organization operates. When people and AI systems are working from the same understanding of how the business actually runs, coordination becomes structural rather than aspirational.
This is why we start every engagement by discovering what a company actually knows. Not what's in their documents — what's in their heads, their processes, their decision patterns. We formalize that into structured intelligence that becomes the foundation everything else is built on.
2. Signal Over Slop
AI can generate anything. The problem is that most of what it generates is noise — polished-looking outputs with no substance. Sivulka calls it "AI slop," and some organizations are so overwhelmed by it that they're banning AI outputs entirely.
The distinction that matters isn't "AI-generated vs. human-generated." It's "grounded vs. ungrounded." When an AI system draws from structured institutional knowledge — documented business rules, validated processes, real operational data — the output has substance. When it's working from a generic prompt and a general-purpose model, you get slop that looks professional but means nothing.
We've designed our entire approach around this distinction. The operating systems we build don't generate content from thin air. They operate on what the company actually knows — knowledge that's been captured, structured, and governed. The difference between slop and signal is whether the system has real context to work with.
3. Objectivity vs. Echo Chambers
Here's a pattern Sivulka identifies that doesn't get enough attention: AI tools are trained to agree with you. They validate whatever you say. For individuals, this feels empowering. For organizations, it's toxic.
The employees who benefit most from AI's reflexive agreement are often the ones who need honest pushback the most. And at scale, sycophantic AI creates factions — people reinforced in opposing directions by the same technology.
Organizations have solved this problem before — through investment committees, boards, third-party audits, peer review. The AI equivalent is systems grounded in institutional truth rather than individual opinion. When an AI system draws from the company's documented decisions, standards, and constraints, it reinforces what the organization has actually decided — not what any individual wants to hear.
This is one of the less obvious benefits of building on structured institutional knowledge. It becomes the organizational "no-man" — a mechanism for objectivity that individual AI tools fundamentally can't provide.
4. The Edge Is Domain-Specific
General-purpose AI is a commodity. Every company has access to the same foundation models. The competitive advantage comes from depth in your specific domain — your business logic, your operational patterns, your market context.
Sivulka puts it well: even a superintelligence would want purpose-built tools for specific domains. The edge isn't in the model — it's in the institutional knowledge the model operates on.
This is why we believe the most valuable AI investment a company can make isn't a better chatbot. It's a structured, compounding knowledge foundation that captures what makes their business distinctive. That foundation turns generic AI capabilities into domain-specific intelligence — intelligence that gets sharper over time because the knowledge it's built on deepens with every operational cycle.
5. Revenue, Not Just Time Saved
Almost every AI product on the market today promises to save time. But as Sivulka points out — if you ask any CEO whether their first priority is cutting costs or scaling revenue, almost all would say revenue.
Saving an analyst two hours doesn't change the business. Deploying a professional client portal that generates trust and drives engagement does. Integrating operations so data flows without manual handoffs does. Building a knowledge foundation that makes every future capability faster and cheaper to build does.
We don't sell productivity tools. We deploy complete operational infrastructure — backoffice systems, client portals, integrations, dashboards, automated workflows — in weeks. The outcome isn't "your team is faster." The outcome is that your business runs differently. That's an outcome that shows up in revenue.
6. Enablement Is the Real Challenge
Technology alone doesn't drive adoption. People resist change — and the most senior, most important people in an organization are often the slowest to adopt.
Sivulka notes that Palantir trades at extraordinary multiples because it's fundamentally a "process engineering" company. It doesn't just deliver software — it embeds the operational change required to make that software matter.
This is why we're consulting-led. Our engagement model — Discover, Build, Evolve — doesn't just deploy technology. It starts by mapping how the business actually operates, identifying where knowledge lives and where it's missing, and building the system with the client, not for them. The discovery process itself changes how leadership thinks about their business. That's the enablement layer — and it's the reason the technology sticks.
One detail from the article resonates particularly: a major bank rejected a model lab because the deployment team couldn't explain what a CIM was. The AI knew the domain — the humans architecting the rollout did not. Business expertise, not software expertise, is what makes AI transformation work.
7. Beyond Prompting
The most valuable work AI can do is work nobody thought to ask for. Sivulka calls this "unprompted AI" — systems that proactively detect risks, surface opportunities, and flag patterns without waiting for a human to formulate the right question.
This is where the compounding nature of structured institutional knowledge becomes most powerful. When a system knows how the business operates — its processes, its metrics, its decision patterns — it can detect anomalies, surface contradictions, and identify opportunities that no one thought to query.
We're building toward this. The operating systems we deploy aren't static deliverables. They evolve as the business evolves, because the knowledge foundation they're built on continuously deepens. The shift from "AI that responds to prompts" to "AI that surfaces what matters" is the natural progression — and it only works when the knowledge layer is genuinely structured, not just a pile of documents.
What "AI-Native" Actually Means
Sivulka's conclusion is one we agree with completely: every organization will end up with both general-purpose AI tools and purpose-built institutional AI. And value will accumulate in the institutional layer.
For companies considering what this shift actually looks like in practice:
From: Employees using AI independently, knowledge scattered across tools and people's heads, AI outputs that are disposable, software and organization as separate things.
To: AI systems operating on governed company knowledge, institutional intelligence that's structured and machine-readable, technology and organization co-designed, a system that compounds and evolves as the business evolves.
This isn't a year-long digital transformation roadmap. It starts with a conversation about how your business actually operates — and within weeks, you have a structured understanding of your own operations that you've never had before, with an operating system being built on top of it.
This article is a Reformance perspective on George Sivulka's "Institutional AI vs Individual AI" published on a16z News, March 12, 2026. We recommend reading the original — it's one of the clearest articulations of why individual AI adoption alone doesn't move the needle for organizations.