The Halo Is Changing: AI, Accenture, and the Reinvention of IT Services

There are moments in every industry when the numbers matter, but the narrative matters even more. Accenture’s recent market reaction feels like one of those moments. This is still a formidable company with global scale, trusted boardroom access, deep enterprise relationships, delivery discipline, and a long history of navigating technology transitions. It has not suddenly become weak, nor has it forgotten how to serve clients. The sharper point is that investors are now asking whether the old IT services value equation deserves the same valuation halo in an AI-native world.

For decades, the global services industry carried a powerful halo. Scale was seen as strength. Headcount was seen as capability. Global delivery centers were seen as industrial muscle. Certifications, offshore leverage, utilization, partner tiers, and pyramid structures were interpreted as signs of maturity. Large transformation programs were seen as proof of customer intimacy. The ability to mobilize thousands of people across geographies became the defining proof that large services firms were indispensable to enterprise technology. That halo was not artificial. The industry earned it by helping enterprises move from mainframes to ERP, from ERP to cloud, from fragmented systems to digital platforms, and from manual operations to automated workflows.

But AI is changing the lens through which the industry is being judged. That does not mean services companies are becoming irrelevant. In fact, the opposite may eventually prove true. As enterprises move from AI experimentation to AI at scale, services companies may become even more central because they sit at the intersection of technology, process, data, governance, compliance, change management, and execution. What is changing is not the relevance of services. What is changing is the measure of relevance.

This is where Phil Rosenzweig’s idea of the Halo Effect becomes useful. When companies perform well, observers describe their leadership as visionary, their culture as disciplined, their people as energized, and their strategy as bold. When performance turns, the same organization is suddenly described as complacent, overexposed, bureaucratic, or strategically confused. The underlying company may not have changed in a day; the interpretation has. Performance contaminates judgment. During success, everything is covered in gold dust. During pressure, everything is seen through smoke.

The services industry must avoid both extremes. The old halo said large services companies would always win because they had reach, relationships, talent, process maturity, and trusted delivery. The new reverse halo says they are vulnerable because AI will automate the work on which they depend. Both views are incomplete. IT services are not dead. Enterprises are not suddenly simple. Legacy estates remain complex. Data is still fragmented. Business processes are full of exceptions. Cyber risk is increasing. Regulatory scrutiny is rising. SaaS, cloud, data, and AI ecosystems are becoming more interconnected, not less. What is fading is not services; what is fading is the comfort of undifferentiated, labor-heavy, effort-priced services.

The old IT services equation was built around effort: people multiplied by billing rates, utilization, and duration. The industry perfected this model. It industrialized talent sourcing, training, global delivery, quality processes, program management, and account mining. It became a remarkable execution engine for the enterprise world. For every major technology wave—cloud, SaaS, digital, cybersecurity, data, analytics—the answer usually involved more migration, more integration, more support, and therefore more services effort. AI is different because it does not simply create new work; it questions the unit of work itself.

If a code migration that once required 200 people can be done by 40 people plus AI agents, who captures the saved value? If testing, documentation, support, configuration, process mapping, and knowledge management can be compressed dramatically, does the services firm protect the old effort model or lead the new productivity model? The right answer is clear: services firms must lead the productivity shift. Firms that help clients compress work, improve quality, reduce risk, and accelerate transformation will not lose relevance. They will move closer to the center of enterprise reinvention.

One analogy may help. The traditional services model was like a large factory floor. When demand increased, you added more stations, more workers, more supervisors, more shifts, and more throughput. Scale mattered because throughput depended on coordinated human effort. The AI-native services model looks more like a power plant. The question is not how many people are standing on the floor; the question is how much energy the system generates, how efficiently it converts input into output, how reliably it runs, and how safely it operates under stress. Services companies cannot simply sprinkle AI tools across the old factory and call it reinvention. They have to redesign the machinery itself: pricing, delivery methods, workforce architecture, incentives, IP creation, governance, and client value measurement.

This is why Vishal Sikka’s warning deserves attention. His argument that services-led and software companies may need radical reinvention, even potentially outside the glare of public markets, should not be read only as a call to go private. It is better understood as a call for strategic freedom. Public companies are often rewarded for quarterly stability: predictable margins, utilization, cash conversion, dividends, buybacks, and incremental growth. But AI reinvention may require bold moves that create short-term discomfort—cannibalizing legacy revenue, shrinking effort-heavy work, investing heavily in AI-native platforms, changing incentives, retraining at scale, acquiring capability rather than capacity, and accepting temporary margin pressure. The larger question is whether services companies can act with the independence and speed required to disrupt themselves before clients or competitors do it for them.

The positive interpretation is that services companies are not starting from weakness. They have enormous advantages if they choose to use them differently. They understand enterprise complexity. They know how large companies actually operate. They understand messy legacy estates, regulatory constraints, process exceptions, security concerns, data fragmentation, and organizational resistance. They have relationships with CIOs, CFOs, CHROs, business heads, and transformation leaders. They know how to run programs across geographies, functions, and platforms. In an AI world, where the hardest work will be moving from pilot to production, these advantages can become even more valuable.

The Netflix analogy is relevant here. Netflix did not survive because it loved DVDs. It survived because it was willing to attack the DVD model while that model was still profitable. It moved from mail-order DVDs to streaming, then from streaming distribution to original content, and then into global entertainment infrastructure. The hardest time to reinvent is not when the old model is dead; it is when the old model is still producing cash. That is where many services companies are today. The traditional model is not dead. It still produces revenue, employs millions, serves clients, and creates value. But the next model must be built before the old one visibly declines.

The services industry also needs to avoid the classic mistake of judging performance only in absolute terms. A company can improve and still fall behind. Revenue can grow and still disappoint. Margins can remain healthy and still be repriced. Internal dashboards can show thousands of people trained on AI, hundreds of pilots launched, dozens of partnerships announced, and many internal productivity programs underway. But these are absolute measures. The real question is relative: are services companies moving faster than client expectations, faster than competitors, and faster than the rate at which AI is commoditizing traditional effort?

This is where the measurement system must change. In the old model, the industry measured headcount, utilization, pyramid ratio, billability, certifications, bookings, revenue, margin, and delivery efficiency. Those metrics still matter, but they are no longer sufficient. The new measures must include AI-led productivity, cycle-time compression, defect reduction, autonomous resolution, reusable IP adoption, revenue from AI-native offerings, client outcomes achieved, work eliminated through automation, agent governance maturity, and value delivered per employee. The industry does not need to abandon discipline. It needs a new discipline.

Another analogy helps here: the difference between a barometer and a thermostat. Many companies use AI metrics like a barometer. They report the weather: how many people were trained, how many pilots were launched, how many tools were deployed, how many partnerships were signed. That is useful, but it does not change the room. A thermostat changes the temperature. Services firms now need thermostat metrics: how much work was actually compressed, how much quality improved, how much speed increased, how much cost was avoided, how much revenue shifted to differentiated offerings, and how much measurable value was created for the client. The industry does not need more AI weather reports. It needs proof that the operating temperature has changed.

The future value equation for services companies will not be people multiplied by rates. It will be intelligence multiplied by trust. The winning firms will combine domain depth, data capability, AI-native delivery, reusable IP, governance, ecosystem orchestration, and change leadership. In banking, life sciences, manufacturing, insurance, telecom, retail, healthcare, and the public sector, AI cannot be deployed safely with generic coding skills alone. It requires business context, regulatory understanding, process redesign, exception handling, auditability, security, human oversight, and measurable outcomes. AI without domain depth is like a brilliant intern: fast, impressive, and occasionally dangerous. Enterprises will still need guides, architects, integrators, and accountable transformation partners. The opportunity is for services firms to become those partners at much higher levels of value.

This is why the next phase could put services companies at centerstage. The first phase of generative AI was model fascination. The second phase was pilot proliferation. The third phase will be enterprise adoption at scale. That third phase is where services companies matter most. Models alone do not transform enterprises. Copilots alone do not redesign operating models. Agents alone do not resolve data quality, security, compliance, integration, and change management. The real enterprise AI challenge is not creating impressive demos. It is making AI work reliably, safely, repeatedly, and measurably inside complex organizations. That is a services problem as much as a technology problem.

The center of gravity will therefore move from “AI tools” to “AI operating models.” This is where services firms can lead. They can help clients redesign processes around agents, modernize data foundations, integrate AI into enterprise applications, build control towers for agent governance, establish human-in-the-loop models, measure productivity, manage risk, retrain workforces, and convert scattered pilots into scalable transformation programs. This is not lower-value work. It is higher-value work. But it must be priced, sold, staffed, and measured differently.

There is also a human dimension to this transition. The IT services industry employs millions of people. Behind phrases like automation, productivity, compression, and AI leverage are careers, families, aspirations, and identities. Reinvention cannot simply be a financial exercise. Developers must become AI-augmented engineers. Testers must become quality architects. Business analysts must become process intelligence specialists. Project managers must become transformation orchestrators. Support teams must become automation supervisors. Domain experts must become agent trainers, model evaluators, and governance owners. The worst version of the future is AI as a blunt cost-cutting weapon. The best version is AI as a way to raise the quality, leverage, and dignity of work.

The final lesson from the Halo Effect is humility. Markets overpraise during good times and overpunish during bad times. A single stock reaction does not define the destiny of a company or an industry. Accenture and other leading services firms have reinvented many times before, and it would be unwise to underestimate their ability to do so again. But it would be equally unwise to dismiss the signal. AI is forcing the services industry to justify its value in a world where intelligence is abundant, automation is accelerating, and clients are under pressure to do more with less.

The old halo was built on scale. The new halo will be built on intelligent scale. The old model sold effort. The new model must sell outcomes. The old model celebrated utilization. The new model must celebrate work compression. The old model priced people. The new model must price value. The old model implemented systems. The new model must redesign work. The old model was powered by labor arbitrage. The new model will be powered by agentic leverage, domain depth, data readiness, governance, and trust.

This is not the end of IT services. It may be the beginning of a more important chapter. Services companies can move to centerstage in the enterprise AI era because they are uniquely positioned to translate AI potential into enterprise reality. But to do so, they must change what they measure, what they reward, what they sell, how they deliver, and how they define value. The halo has not disappeared. It is changing. The companies that understand this will not merely survive the AI transition. They will define it.

Labels: GenAI, Halo Effect, IT Services

The Human Algorithm — Democracy, Purpose, and the Ethics of Intelligence

I. The Arrival of Equivalence

At the Financial Times’ 2025 Future of AI Summit, a remarkable claim echoed across the stage. Nvidia’s Jensen Huang, Meta’s Yann LeCun, Turing Award laureates Geoffrey Hinton and Yoshua Bengio, and Stanford’s Fei-Fei Li agreed: in many domains, AI has reached human-level intelligence

Machines can now recognize tens of thousands of objects, translate hundreds of languages, and solve problems that stump PhDs. “We are already there,” Huang said. “And it doesn’t matter—it’s an academic question now.”

What matters is what comes next: whether humanity uses this power to augment itself or abdicate its agency.

II. Augmentation, Not Abdication

The pioneers remain surprisingly united in humility. Fei-Fei Li likens AI to airplanes: machines that fly higher and faster than birds, but for different reasons. “There’s still a profound place for human intelligence,” she insists—particularly in creativity, empathy, and moral reasoning

Hinton envisions machines that will “always win a debate” within 20 years, yet still sees their role as complementing humans, not replacing them. Bengio warns that decisions made now—on alignment, ethics, and governance—will define whether this era uplifts or undermines civilization.

Their consensus: AI should amplify what is best in us, not automate what is worst.

III. The New Civilizational Technology

Fei-Fei Li calls AI a “civilizational technology.” It touches every sector and every individual. Like electricity, it doesn’t belong to one industry—it redefines all of them.

But civilization also requires values. Yoshua Bengio, once focused purely on algorithms, now devotes his research to mitigation—ensuring that systems understanding language and goals cannot be misused or evolve beyond control.

Human-centered design, ethical guardrails, and public trust are not optional accessories; they are the operating system of the AI age.

IV. The Democratic Crossroads

Eric Schmidt and Andrew Sorota, writing in The New York Times, describe the danger vividly: nations may soon be tempted by algocracy—rule by algorithm. Albania’s new AI avatar, Diella, already awards over a billion dollars in government contracts automatically, promising to end corruption.

It’s an appealing trade: competence over chaos. But Schmidt warns it’s the wrong reflex. Algorithms can optimize efficiency, but they cannot arbitrate values. When citizens cannot see how decisions are made or challenge them, they become subjects, not participants

V. When Algorithms Govern

Across 12 developed nations, surveys show majorities dissatisfied with how democracy works. Many now say they trust AI systems more than elected leaders to make fair decisions

But an algorithmic state doesn’t solve alienation—it deepens it. When bureaucratic opacity is replaced by digital opacity, the result is the same: unaccountable power.

VI. The Democratic Upgrade

There is another path. Schmidt and Sorota point to Taiwan’s vTaiwan platform—a model of AI-assisted democracy. When Uber’s arrival threatened local taxi livelihoods, the government used an AI deliberation tool to map citizen sentiment, identify areas of consensus, and craft a balanced policy.

Here, AI didn’t decide. It listened. It turned thousands of comments into a coherent social map, surfacing shared ground instead of amplifying division. The outcome—insurance and licensing for ride-share drivers without killing innovation—proved that AI can help democracy deliberate at scale

This is a glimpse of Democracy 2.0—where AI becomes the translator between people and policy, expanding participation instead of erasing it.

VII. The Ethical Singularity

The ethical dilemma of AI is not whether it will surpass human intelligence—it already does in narrow domains—but whether it will mirror human wisdom.

Today’s models are optimized for engagement, not enlightenment. Outrage drives clicks, and clicks drive revenue. The same algorithms that translate text can also amplify polarization. The danger, as Schmidt warns, is not dystopian robots but “systems that erode trust faster than governments can rebuild it.”

To counter that, societies must build benevolence into the stack: transparent systems, explainable models, participatory oversight. Ethics must be coded, not declared.

VIII. The Redefinition of Work and Meaning

The AI era doesn’t just transform jobs; it transforms identity. When machines perform cognitive labor, human value migrates toward emotional and moral dimensions—toward why, not how.

Fei-Fei Li argues that AI’s purpose is to relieve humans of repetitive cognition so they can focus on “creativity and empathy.” The next generation of education, leadership, and art will thus emphasize synthesis over specialization.

In this sense, AI is not replacing the human mind—it’s forcing it to evolve.

IX. The Philosophical Reckoning

When Hinton was asked what keeps him up at night, he said: “The moment a machine not only learns from us but starts to teach us what to value.” That moment may be closer than we think.

Machines are already discovering patterns in science, art, and medicine that humans missed. The frontier question is not whether AI will have values—but whose values it will reflect. The answer cannot be left to code alone. It must be debated, voted, and revised—just as laws are.

Democracy, then, is not an obstacle to AI. It’s the immune system that keeps intelligence aligned with humanity.

X. Toward Augmented Civilization

The next decade will see five defining shifts:

1. Cognitive Equivalence — Machines match human reasoning in most structured tasks.

2. Agentic Systems — Models evolve from language processors to autonomous problem-solvers.

3. AI-Enhanced Governance — Policy becomes participatory and data-driven, not merely electoral.

4. Embedded Ethics — Safety, explainability, and fairness move from afterthought to design principle.

5. Human Renaissance — Creativity, empathy, and moral imagination become the new scarce resources.

Each shift is both technological and moral. The more intelligence we externalize, the more intentionality we must internalize.

Labels: GenAI

The Age of Instant Learning: How AI Collapsed the Old World -Part 1

I. The Collapse of the Learning Curve

For most of industrial history, progress obeyed a familiar rhythm: make, fail, learn, repeat. Factories, schools, and economies ran on experience curves—each doubling of production cut costs by a fixed percentage, a phenomenon codified as Wright’s Law in 1936.

But artificial intelligence has detonated that pattern. In the words of the Wall Street Journal, “AI destroys the old learning curve.” Experience no longer follows production—it precedes it. Simulation can now test a million variations before a single box ships. Entire industries are learning before doing, producing competence before contact with reality.

Knowledge that once took decades can now emerge in days. The assembly line has given way to the algorithmic sandbox.

II. From Breakthrough to Buildout

The acceleration didn’t happen overnight. It’s the culmination of decades of breakthroughs that fused three elements—compute, data, and algorithms—into a self-reinforcing flywheel.

1. Compute as the New Infrastructure
   Jensen Huang’s “aha” moment at Nvidia came when he realized arithmetic was cheap but memory access was costly. That insight birthed the GPU—a chip that could perform thousands of operations in parallel, transforming computer graphics into a universal engine for machine learning. “AI,” Huang said, “is intelligence generated in real time.” Every GPU in the world is now “lit up,” forming a planetary grid of thought.

2. Data as the Oxygen of Learning
   Fei-Fei Li’s ImageNet project—15 million labeled images—became the missing nutrient that allowed algorithms to generalize. Machines, once “starved of data,” suddenly had the diet required for understanding the visual world. Big data didn’t just enhance learning; it became the law of scaling.

3. Algorithms as the Nervous System
   Geoffrey Hinton’s early experiments with backpropagation, combined with Yann LeCun’s convolutional networks and Yoshua Bengio’s probabilistic learning, taught machines to self-correct. Later, self-supervised learning allowed them to infer structure without explicit labels—the leap that produced today’s large language models.

The synergy of these three domains ended a 40-year stall in AI progress. What followed is not a bubble, as Huang argues, but “the buildout of intelligence”—a massive, ongoing industrial revolution where every data center becomes a factory for cognition.

III. Experience Before Production

Wright’s Law presumed learning by doing. AI replaced it with learning by simulation. A supply chain, for example, can now model thousands of disruptions—storms, strikes, surges—before they happen. Mistakes are made virtually, not physically. Costly iterations disappear.

The implication is profound: the learning cycle is no longer physical—it’s computational. Digital “twin” worlds allow designers, manufacturers, and urban planners to test scenarios endlessly at near-zero cost. Experience scales instantly.

When learning precedes production, innovation ceases to be cyclical. It becomes continuous.

IV. The Era of Dual Exponentials

The current AI economy is powered by two simultaneous exponentials:

• The compute required per inference—every model generation demands orders of magnitude more processing.

• The usage growth—billions of people are now invoking AI multiple times per day.

This dual surge fuels what Huang calls the “lit-up economy.” Every GPU, every watt, every dataset is active. Unlike the dot-com boom’s “dark fiber,” this buildout isn’t speculative; it’s productive. The network hums 24/7, producing tokens, translations, designs, and discoveries in real time.

V. The Death of the Industrial Learning Curve

In classical economics, efficiency was a function of repetition. Workers honed skills over years; firms improved through iteration. AI obliterates that logic. The marginal cost of additional intelligence falls toward zero once models are trained.

Jonathan Rosenthal and Neal Zuckerman described this inversion succinctly: “AI makes experience come before production.” The new competitive advantage isn’t scale—it’s simulation depth. Winners aren’t those who produce the most, but those who can model the most possibilities and act first.

This creates a new hierarchy:

• Data owners command the raw material of insight.

• Compute owners command the means of learning.

• Model owners command the interface between the two.

Those three layers now define industrial power.

VI. Work Without Apprenticeship

As learning curves collapse, the apprenticeship model of work collapses with it. Junior analysts, designers, and operators once learned by repetition. Now, generative systems learn faster and at greater scale. A planner who once needed ten years of experience can be replaced—or augmented—by an AI that has simulated ten million logistics events.

This doesn’t eliminate human roles; it shifts the locus of value to judgment, ethics, creativity, and synthesis—areas where context, emotion, and uncertainty dominate.

VII. The Entrepreneurial Shockwave

Ironically, the same forces that destroy traditional jobs unleash an entrepreneurial explosion. When capital, computation, and knowledge become abundant, the barriers to entry vanish. Rosenthal and Zuckerman foresee “nimble companies in numbers never seen before”—each rising fast, solving a niche problem, and disappearing once its utility fades.

The economy becomes an adaptive organism: millions of micro-experiments running in parallel, guided by real-time data and machine mediation. Failure ceases to be fatal—it becomes feedback.

VIII. A New Law of Progress

In the old world, experience accumulated linearly and decayed slowly. In the new world, knowledge accumulates exponentially and decays instantly.

Wright’s Law still matters, but its unit of learning has changed—from a physical product to a digital simulation, from human effort to machine cognition. The future belongs to those who can collapse the distance between imagination and implementation.

IX. Beyond Productivity

The AI age will not just make us faster. It will change the physics of progress itself. When machines can “pre-learn” reality, civilization moves from reactive to predictive. We stop iterating on what we know and start simulating what we don’t yet know.

For the first time in history, experience scales before existence.
And that—more than any gadget or chatbot—is the true revolution of our age.

Labels: GenAI

The "Philosophy and Glamour" of Projecting GenAI Services

The "glamour" in projecting Generative AI (GenAI) services for IT service companies isn't just about the technology itself; it's about fundamentally recasting their role from a simple implementer to an indispensable strategic partner.

Labels: GenAI, Services