From Digital Enterprise to AI-Native Enterprise

If the missing layer is the intelligence the organisation never wrote down, then the work ahead is not adding AI to what already exists. It is redesigning what exists.

The first enterprise technology revolution was about digitization. We took processes that existed in the physical world and put them into software.

The next revolution is different. We are not simply digitizing existing processes. We are redesigning the processes themselves around a new kind of intelligence.

AI-enabled is not AI-native

That is the difference. An AI-enabled enterprise asks: "Where can we add AI to what we already do?" An AI-native enterprise asks: "If AI could understand, decide and act across this process, how should the process be designed?"

Those are completely different questions. The first adds intelligence to the existing operating model. The second redesigns the operating model around intelligence.

Almost nobody is doing the second one

The US Census Bureau found that 64% of firms using AI made no organizational change at all alongside it. McKinsey finds workflow redesign the factor most strongly correlated with AI reaching enterprise earnings, and only 21% of companies have fundamentally redesigned their operating models around AI.

The economics behind this pattern are old and well measured. Across roughly 30,000 US manufacturing plants, adopting predictive analytics came with about 3% higher productivity, and the measured effect was statistically zero for plants that had not made the complementary investments in systems, skilled people and process design. The technology only pays combined with the redesign.

And this is where the old software paradigm starts to break. For years, enterprise transformation meant buying more software. Then integrating it. Then training people on it. Then changing processes around it. Then building another system to fill the gap left by the previous systems.

The next generation will increasingly work in the opposite direction. Start with the business outcome. Understand the process. Understand the economics. Understand the decisions. Understand the people. Understand the existing technology. Capture the operational intelligence. Then decide what should be software. What should be an agent. What should be automated. What should remain human. And what should be redesigned completely.

Which parts should remain human

That is not a rhetorical question. When US insurance regulators surveyed 193 auto insurers, not one used AI to deny a claim, while 96 used it to inform the adjuster. An entire regulated industry drew the same line independently: automate up to the adverse decision, not through it.

And a meta-analysis of 106 human-AI experiments found the combination often performs worse than the better of the two alone, with the losses concentrated in decision tasks. Where the human sits is a design decision. It has to be made deliberately, process by process.

A coherent layer across the systems

The result is not necessarily fewer enterprise applications. It is something more important: a coherent operating system across them. The ERP still exists. The CRM still exists. The data warehouse still exists. The SaaS applications still exist. But they stop being the thing humans have to understand individually. AI can increasingly become the layer that understands the enterprise across those systems.

That is why companies like Palantir matter to this story. Palantir built its business on the premise that enterprise AI is not simply a model plugged into a database. It requires a representation of the enterprise. Its objects. Its relationships. Its workflows. Its logic. Its constraints. Its decisions. Its actions. The ontology became central because the model needs to understand the world in which it is acting. And the industry is now moving rapidly toward the broader version of the same idea, under the name context engineering.

But there is still a fundamental gap. Most enterprise context is built from what the company has already recorded. Data. Documents. Emails. Meetings. Messages. Transactions. Activity. The decisions that were never written down remain invisible.

That is why the next frontier is not simply collecting more data. It is capturing the intelligence behind the data. And once you can do that, something extraordinary becomes possible. You can begin to turn the organization itself into something that AI can understand.

That is what makes an enterprise truly AI-native. Not having an AI assistant. Not having a hundred agents. Not putting a chatbot on the intranet. But having an operating model in which intelligence is embedded into how the company works.

That is the argument. Enterprises have rebuilt themselves around a technology once before, and that time we can check. The next part starts the test.

Sources

• Census Bureau CES Working Paper 26-25. Bonney et al., 2026. 64% of AI-using firms report no institutional adjustment at all. Only 7 to 8% changed data practices.
• Building AI Advantage. McKinsey, 2026. Workflow redesign is the factor most strongly correlated with enterprise EBIT impact. 21% of companies have fundamentally redesigned their operating models around AI.
• The Power of Prediction. Brynjolfsson, Jin & McElheran, Business Economics 2021. Roughly 30,000 establishments: 2.87% higher productivity from predictive analytics, statistically zero without complementary IT capital, skills and process design.
• The Productivity J-Curve. Brynjolfsson, Rock & Syverson, NBER. Returns to general-purpose technologies lag because the complementary intangible capital must be built first.
• NAIC Private Passenger Auto AI/ML Survey. US insurance regulators, 2022. 0 of 193 auto insurers used AI/ML to deny a claim. 96 used it to inform the adjuster.
• When Combinations of Humans and AI Are Useful. Vaccaro, Almaatouq & Malone, Nature Human Behaviour, 2024. Meta-analysis of 106 experiments: human-AI combinations underperform the better of the two alone on decision tasks.
• Foundry Ontology. Palantir documentation. The operational representation of the enterprise: objects, links, actions.
• Effective Context Engineering for AI Agents. Anthropic, 2025. Context engineering framed as the natural progression of prompt engineering.