Why Data Platforms—Not AI Models—Will Define Enterprise Winners in 2025
How the rise of runtime-generated applications and agentic AI will reshape enterprise value creation—and why CIOs need to act now
The unveiling of OpenAI's o3 model at the end of December 2024 sparked predictable excitement across technology circles and social media. While benchmark discussions dominate current discourse, the more profound implications lie in what this represents for enterprise computing in 2025 and beyond. This moment marks not just another AI advancement, but a fundamental shift in how enterprises will need to approach their technology infrastructure.
The Automation Imperative
When Microsoft CEO Satya Nadella recently declared that "all SaaS apps will become multi-database AI agents," he wasn't merely making a provocative statement about Microsoft's strategic direction. Rather, he was highlighting a crucial truth about enterprise software: much of what we consider "high complexity software" is, in reality, sophisticated CRUD operations wrapped in business logic. This observation matters because it exposes both the opportunity and vulnerability facing enterprises today.
The emergence of o3 and similar AI reasoning models represents more than incremental progress - it signals the practical realisation of the "citizen developer" vision that has tantalised enterprises for decades. Natural language interfaces and sophisticated prompting capabilities will enable rapid micro-app development at a scale previously unimaginable. While this democratisation of development naturally concerns enterprise IT leaders (bringing back memories of "shadow IT"), it introduces what I believe are two critical concepts for the future of enterprise computing: User Abstraction and Application Persistence.
The Adoption Curve: From Human to Agentic Computing
To understand this transformation, we need to examine the enterprise AI adoption curve:
Human Only Operations
Human + High-Friction AI (manual data transfer between systems)
Human + Low-Friction AI (direct enterprise data access)
Human + Native AI (runtime app/service creation within security boundaries)
Human + Agentic AI (outcome-driven orchestration across domains)
This progression isn't merely technical - it represents a fundamental shift in how enterprises create and capture value. To illustrate this, consider a Consumer Packaged Goods (CPG) company operating at Stage 5:
Picture the future of CPG innovation: a harmonious orchestration of AI agents that transforms the traditionally sluggish product development cycle into a real-time symphony of data-driven decisions. A trend-detection agent identifies a newly popular ingredient, instantly triggering an R&D agent to model formulations while a sales prophet runs market simulations. As the ‘supply chain agent’ orchestrates logistics and the ‘finance prophet’ continuously evaluates ROI, a marketing muse crafts targeted campaigns - all before a human has opened their first spreadsheet. This isn't merely automation; it's the manifestation of what I would call "Computational Product Management," where AI agents aggregate decision-making capabilities. But here's the critical insight: success in this new paradigm won't belong to those with the flashiest AI models, but to enterprises that build the foundational Data Platforms enabling these agents to orchestrate and drive the business.
The Data Platform Imperative
While this scenario might seem futuristic, the gap between current capabilities and this future state isn't primarily technological - it's infrastructural. This brings us to the critical role of Data Platforms, which are widely misunderstood in the industry. If AI represents the automobile in our analogy, Data Platforms are the roads, highways, and infrastructure necessary for widespread adoption and value creation.
What Data Platforms Are Not
Simply a Database with a dashboard front-end (although this can be a good start to assuring data quality)
Just a data lake (having unstructured data in your infrastructure can be powerful, but not a prerequisite)
Data infrastructure like Databricks or Snowflake (both are great tools, but they are just tools)
A New Definition
I propose defining a Data Platform as "the ability of Data Infrastructure to serve Data products to end-users." This definition matters because it focuses on capability rather than technology, on outcomes rather than infrastructure; unlike Databricks and Snowflake which currently monopolise the term.
The Abstraction-Persistence Framework
This brings us to two critical concepts that will define the next era of enterprise computing:
Abstraction: The degree of technical expertise required to interact with and leverage data systems. This exists on a spectrum from raw code access (thin abstraction) to templated interfaces (thick abstraction).
Persistence: The durability requirement of data products, ranging from ephemeral runtime generations to hardened, production-grade applications.
This framework reveals why the current moment is so significant: AI-powered no-code tools will enable high-abstraction, low-persistence applications that dramatically reduce the cost and time to value for business initiatives. The economic implications are profound - the unit cost of revenue generation will fundamentally decrease, suggesting significant upside for enterprise valuations.
Consider how Excel became the unofficial operating system of the modern business: Sarah in Finance needs to analyse seasonal pricing trends, so she spins up a spreadsheet, imports some data, adds formulas, and shares her insights. The spreadsheet serves its purpose and likely ends up forgotten in a shared drive. This works because Excel's power lies not in persistence but in its ability to rapidly create disposable analytical tools.
Now imagine Sarah in 2025: Instead of wrestling with pivot tables, she prompts the LLM-powered Data Platform to "build an app that shows how our product pricing correlates with weather patterns across regions." Within seconds, she has a fully functional web app with interactive visualisations, statistical analysis, and even API connections to real-time weather data. She uses it for her presentation, shares insights with colleagues, and once the analysis is complete, the app gracefully disappears - no IT tickets, no maintenance burden, no technical debt. This isn't just faster spreadsheet work; it's the democratisation of sophisticated software development with the same disposable convenience that made Excel indispensable. The key difference? These runtime-generated apps can tap into enterprise data platforms, enforce security policies, and deliver insights that would have required weeks of traditional development.
The CIO's Moment
This transformation places the office of the CIO at the centre of enterprise value creation. While "data-driven" has been a corporate mantra for decades, Data Platforms and AI finally provide a concrete path to realisation. The parallels to urban infrastructure are instructive - like city plumbing, Data Platforms represent a significant upfront investment with ongoing maintenance costs, but the value unlock is transformative. Data Platforms in the short term enable organisations to be truly data-driven, in the medium term, they pave the way to algorithmic and agentically driven businesses. Agentic AI will be to the enterprise, what self-driving is to the car; in effect agentic AI will enable self-driving businesses with minimal human intervention.
Strategic Implications
For incumbents, the threat and opportunity are equally massive. The emergence of micro-teams achieving multi-million dollar revenue through AI leverage will disrupt traditional enterprise models. Success will depend on:
Rapid Data Platform development
Strategic positioning of Platform teams
Organisational transformation led by CIO/CTOs
Conclusion
The best time to start building a Data Platform was last year; the next best time is tomorrow. While the investment is significant, the cost of inaction is existential. Enterprises must view Data Platforms not as a technical initiative but as fundamental business infrastructure for the AI era.