Early Age of AI: 2023–2025
In the early Age of AI, three years packs three decades of action.
In November 2022, ChatGPT stunned the world with a Generative AI that anyone with a browser can converse with to learn, find answers, and generate content. The consumer adoption raced from 0 to 100 million in 2 months and reached 1.8 billion in 18 months (mid-2025).
At the enterprise level, business, IT, and security teams began a different journey. They bought secure access to closed LLMs, moved to private hosting of open models, and ran dozens of controlled pilots. Through this process, they learned the hard lessons on data readiness, quality, retrieval, and access; model tradeoffs; bias and errors; and the critical need for human-in-the-loop management and robust governance across functions.
The experiments were real, and the limited success in scaling to production provided valuable lessons. Crucially, the business cases were often narrative-based hypotheses rather than empirical measurements. By 2025, this experimentation extended into Agentic AI, setting the stage for the next phase of enterprise adoption
Middle Age of AI 2026 onwards - First movers move Enterprise AI to Enterprise Wide AI
Beginning 2026, a clear shift in strategy is emerging. Armed with learnings from the pilot phase, a cohort of first-mover companies is transitioning from “Enterprise AI” to “Enterprise-wide AI.”
This is not merely an operational evolution; it is a strategic course change driven by a leadership vision to create disproportionate competitive advantage. These organisations are choosing to accept and actively manage the risks and challenges of AI, rather than be deterred by them. In an ironic twist, it is the largest, most mature enterprises—the proverbial elephants—that are now learning to dance, demonstrating a new level of agility.
This move to Enterprise-wide AI is materialising at two distinct conceptual levels.
The Anatomy of Level 1: An Emerging Blueprint for Enterprise-Wide Access
Looking under the hood of recent landmark deals, such as Deloitte’s partnership with Claude and PwC’s enterprise agreement with OpenAI, a common architecture appears to be emerging.
For these first movers, the strategy is not simply about buying licenses, but about deploying a comprehensive system. The thinking is built on three pillars: securing access, guiding functionality, and establishing a robust implementation framework.
Access: Securing the Front Door
The foundational goal is to integrate the AI platform into the enterprise’s existing security and IT infrastructure, which is considered a non-negotiable first step.
The plan involves making ubiquitous copilots available through Single Sign-On (SSO) to remove friction for users while maintaining centralized identity management.
Crucially, this would require applying Data Security and Data Loss Prevention (DLP) policies to monitor and block sensitive data.
The system would also be wrapped in comprehensive logging and policy guardrails, giving security teams the necessary visibility and control to manage usage at scale.
Functionality: From Blank Canvas to Guided Capability
The approach recognizes that simply providing a chat window is insufficient. The aim is to unlock real value by shaping the user experience to be immediately relevant and productive for specific roles.
This is being achieved through the development of role and persona kits. Under this model, an auditor or a recruiter would be equipped with a curated library of vetted prompts, pre-approved tools, and workflow templates designed for their function, accelerating adoption and directing AI usage toward high-value activities.
Implementation: The Human Infrastructure for Scale
Technology alone cannot guarantee success, so a core part of the strategy is to build the human and governance infrastructure to support the rollout.
This typically involves establishing a central Center of Excellence (CoE) to set standards and best practices, which is then paired with an “academy at scale” to provide mandatory, tiered training across the organization.
A key principle is that the safety stack must be built-in from day zero, not bolted on as an afterthought. This includes planning for content filters, automated PII controls, transparent model cards, and clear approval flows for new use cases.
The Anatomy of Level 2: An Emerging Blueprint for Embedded Intelligence Across the Enterprise
While Level 1 democratizes AI as a tool, Level 2 industrializes it as a core component of the business engine. The defining characteristic of Level 2 is that AI is stitched directly into systems of record and action. Access happens within the workflow of a call center, a financial reconciliation process, or a compliance check—not in a standalone browser.
The strategy targets areas where the ROI math is cleanest: operations, service, support, and compliance. The reason for this focus is pragmatic: these functions often rely on stable templates, which makes evaluation and setting acceptance thresholds easier. Their high transaction volume provides statistically significant telemetry, and the value can be measured through direct labor savings or cycle-time reduction, creating a CFO-friendly business case.
A Case Study in Level 2: The JPMorgan Chase Blueprint
JPMorgan Chase provides the clearest public blueprint for a Level 2 implementation, demonstrating a strategy built on massive scale, a disciplined focus, and a long-term vision.
The Scale of Ambition: The commitment is staggering. In recent interviews, leadership has confirmed an annual AI spend of $2 billion, which is already generating $2 billion in measurable business value—effectively breaking even on its operational AI investment. This is powered by a team of 2,000 AI-focused staff managing over 600 live use cases, with 150,000 employees using internal AI tools weekly. This is the result of over a decade of sustained investment.
The Strategic Focus: Workflow Over Tools: The core of JPMC’s strategy is a deliberate focus on deep workflow integration (estimated to be 85-90% of their effort) over simple tool access. While internal AI assistants exist, they are secondary. The primary goal is to embed AI into front-office processes like lending approvals and back-office engines for compliance, reconciliation, and fraud detection. As CEO Jamie Dimon notes, the goal is to use AI as an operational “force multiplier,” not just another employee tool.
The Implementation Engine: This strategy is executed via a model-agnostic internal platform that allows the firm to route tasks, evaluate performance, and control costs. This platform is what enables the 600+ use cases to be deeply integrated into core banking systems for real-time decision-making. As industry experts note, this success comes from the difficult work of integrating AI into legacy systems to achieve measurable ROI, a far more defensible strategy than a tool-only approach.
While JPMC is the most transparent leader, they are not alone. As noted earlier, firms like EY are pursuing a similar path by embedding generative AI directly into their core tax and assurance platforms.
Funding, Risk, and the New ROI: From Narrative to Numbers
While first movers are deploying AI at scale, the funding sources remain conventional, drawn from reallocated IT budgets, innovation funds, and modernization programs.
The true novelty lies not in where the money comes from, but in how it’s managed. The unique risks of AI are forcing a new discipline around cost control and benefit measurement.
Managing Runaway Risk and Cost
AI programs are notorious for runaway costs, with data preparation, compute scaling, and ongoing governance all adding significant hidden overhead. In response, leading enterprises are adapting rigorous cost-governance frameworks like FinOps and Tech Business Management (TBM) specifically for AI. They use real-time dashboards to monitor usage, model versions, and unit costs.
A new academic proposal even introduces a metric called LCOAI (Levelized Cost of Artificial Intelligence) to create a normalized, “cost-per-useful-inference” metric, which could further rationalize funding decisions in the future.
The Shift to an Empirical, Data-Driven ROI
The most significant change is the move from a static business case to a live, evidence-based stream of value. This is happening in three ways:
AI Climate Warning: Environmental, Ethical, Human and Social, Concerns
Beyond the balance sheet, the adoption of enterprise-wide AI carries a profound and sobering ledger of external costs , if left unmitigated, the damages could far outweigh the business successes.
Five areas demand urgent attention.
1. The Energy Bill Coming Due
The computational demand for AI is scaling faster than Moore’s Law, with a projected 10x growth by 2027 driving a proportionate increase in energy use. The power draw of large-scale AI systems is already being compared to that of entire cities. Industry forecasts suggest AI-driven data centers could consume 8-10% or more of U.S. electricity by 2030, with some projections reaching even higher (Bloomberg/BNEF estimates). If unchecked, AI’s carbon footprint will create a direct conflict with corporate net-zero pledges, inviting intense scrutiny from communities, consumers, and regulators.
2. The Hidden Thirst: Water Scarcity
AI’s hidden thirst is becoming as alarming as its power draw. Datacenters rely on millions of litres of clean water to cool high-density chips. Microsoft’s water consumption rose 34% in one year, largely due to AI (DatacenterDynamics, 2023), and training a single large model can use an estimated 700,000 litres (Li et al., “Making AI Less Thirsty,” 2023).. Projections suggest global AI data centers could consume 4–6 billion cubic meters of water annually by 2030. In water-stressed regions like Arizona, Singapore, and Spain, this will become a strategic constraint, forcing sustainability metrics to track both carbon and water intensity per AI workload.
3. The Mandate for Ethics Beyond Compliance
Regulation sets the floor, not the ceiling. The real risks of hallucinations in critical systems, bias in decision-making, and the misuse of opaque models are already making headlines. Ethics must be built into the entire lifecycle, with consent frameworks, transparent reviews, red-teaming, and human override paths. Compliance is what you must do; ethics is what you ought to do. In an age of fragile public trust, it is survival insurance.
4. The Human Capital Bargain
For employees, the rapid integration of AI into daily workflows presents an existential bargain. While AI can elevate baseline competence and augment human capability, the fear of replacement is real and pervasive. Aggressive automation without clear, committed pathways for reskilling and role transition will erode morale, retention, and social trust. Leaders must pair every deployment target with an equally robust plan for psychological safety and human capital development.
5. The Risk of Societal Self-Sabotage
AI poses a direct threat to the first rung of the professional ladder—the interns, new graduates, and junior analysts whose tasks are most easily automated. There is a real risk of hollowing out the entry-level talent pipeline. This is not just inequitable; it is a form of strategic self-sabotage. Today’s new graduates are tomorrow’s leadership bench. A strategy that excludes them in favor of automation is sacrificing its own future.
Enterprise-wide AI Begins Now: Your Strategy Will Define Your Future
The shift from the “Early Age” of AI experimentation to the “Middle Age” of enterprise-wide deployment is happening now.
This transition from narrative-based projects to instrumented, scaled-up AI is creating a new class of leaders. Your strategic choice—whether to be a first mover, a fast follower, or a slow follower—will have profound consequences for your competitive position over the next decade.
For First Movers: Harden Your Advantage
Your challenge is to convert your head start into a sustainable, defensible lead.
Instrument Everything: Move beyond vanity metrics to measure the unit cost per inference, minutes saved per task, and error reduction rates. This empirical data is the foundation for every future investment decision.
Steward the Downsides: Your leadership will now be measured by how you manage the turbulence around energy, water, ethics, and human capital. Publish a responsible AI plan and treat your reskilling programs with the same rigor as your technology roadmap.
Build the Human Infrastructure: Scale your Center of Excellence and internal training academy. The goal is to create a force multiplier effect where AI and human teams make each other smarter and faster.
For Fast Followers: Prepare for a Precision Takeoff
Your advantage is learning from others’ lessons. Use this time to prepare for a rapid and intelligent adoption.
Watch the Radar: Identify the Level 1 and Level 2 leaders in your industry. Analyze their strategies and outcomes to build your own, more refined business case, avoiding their mistakes.
Find Your Workflow: You don’t need to deploy AI everywhere at once. Identify a single, high-value workflow in operations, service, or compliance where the “math is clean” and focus your initial energy there to prove the value quickly.
Prepare Your Foundation: The biggest barrier to AI success is often a poor data and technology foundation. Use this time to clean your data, modernize core systems, and begin building foundational AI literacy in your teams.
For Slow Followers: Mitigate the Risk of Irrelevance
Choosing to be a slow follower is not a low-risk strategy; it’s a high-risk bet on the status quo. If this is your organization’s reality, the immediate priority is to mitigate the downside and prepare for an eventual, necessary shift.
Quantify the Gap: Your most urgent task is to relentlessly monitor your first-mover and fast-follower competitors. Quantify the competitive gap that is opening in terms of cost, speed, and customer experience. This data is the only thing that will catalyze leadership to act.
Focus on No-Regrets Moves: Invest in the absolute basics that will be required no matter what. This means cleaning your data house, modernizing key infrastructure, and investing in broad digital literacy for the workforce.
Cultivate Pockets of Expertise: You don’t need a massive CoE, but you do need a small, dedicated team exploring these technologies. This small group will keep the organization connected to the market and form the seed of a larger team when the time comes to move
What do you think? Share your thoughts and outlook.
If this was helpful do let me know. If someone needs to read this please share and tag.
Have a safe flight!
Best wishes
II. Questioning / Asking
Good conversations flow from well‑sequenced questions—topical, simple, coherent, cohesive.
LLM Conversation Example 1
Q: What are empirical judgments?
A: Empirical judgments are based on observation, experience, or experimentation.
Q: What are moral judgments?
A: Moral judgments are based on ethical principles and values.
