Porter needed only one book to change the way we think about companies. Competitive Advantage, 1985, and suddenly we all had language for something we had felt but could not name: the value chain. Procurement, production, distribution, marketing, sales, service. Six links, one logic, endless variations.
Some forty years later, that logic is enjoying a quiet retirement.
Not because it was wrong. It was brilliant, for a world in which information was expensive and scarce. In that world, competitive advantage was built on physical pillars: economies of scale, control of distribution, access to capital. Whoever could produce more, cheaper and faster, won. ERP systems, SAP, Oracle, all of them served to optimise that same machine. Good optimisation. But optimisation.
The question facing every CEO today is not whether to optimise the value chain. The question is whether that chain is still the right frame for thinking at all.
My answer, after twenty-five years of working with banks, telecoms and state institutions across the region, and after two years spent building an AI platform, is that it is not.
Digital transformation: the operation was a success, the patient was disappointed
Between 2010 and 2024, almost every serious company went through some form of "digital transformation." We hired consultants, bought cloud services, renamed departments and put the word digital in front of everything we did. Digital marketing. Digital banking. Digital sales.
The result?
In 2023 McKinsey reported that 70 percent of digital transformations fail to achieve their stated goals. Two trillion dollars a year invested in modernisation that, in most cases, did not change the fundamental logic of the business.
The reason was not poor implementation. The reason was a wrong diagnosis.
Digitalisation changed the form, not the substance. The same processes, only faster. The same decisions, only on a screen instead of on paper. The same organisational structures, only in the cloud instead of the server room. The largest companies took their 1995 bureaucracy and migrated it to Azure. Then they called it a transformation.
Amazon and Netflix did not make that mistake. Amazon did not digitalise bookshops, it designed a new model of value delivery. Netflix did not digitalise the video rental store, it removed the need for one. The winners of the digital era did not optimise the old. They created the new.
But that is a greenfield company's privilege. What about those carrying thirty years of legacy, three thousand employees or more, and a regulator staring at every process? For them, digital transformation was the only realistic path, and they made an enormous step forward. The problem is not that they digitalised. The problem is that they stopped there.
March 2026: the month that changed the rules
There are moments when the pace of change accelerates so sharply that what was the future yesterday becomes a lag today.
March 2026 was such a month for AI.
This was not about a single breakthrough, a single company or a single launch. It was about convergence. Close to three quarters of Fortune 2000 companies are already running AI agents in production environments. Not pilots, not proofs of concept, but production. The direct financial impact of AI has nearly doubled as a primary KPI tracked by boards of directors. Month-end financial close processes are 30 to 50 percent faster. Customer response times have gone from hours to minutes.
What changed was not the power of the AI models, which has been improving for months. What changed was the level of autonomy. AI agents stopped being sophisticated chatbots waiting for a question. They became operational actors that execute tasks autonomously, coordinate steps, make decisions within defined parameters and, the key phrase, increasingly do all of this without a human between every step.
Claude Code writes, tests and deploys software. It does not help developers type faster, it autonomously develops working components. Genspark does not search the internet the way Google does, it synthesises, cross-references sources and delivers finished analytical reports. Manus, the AI agent acquired by Meta for a reported two billion dollars, which caused a global sensation, autonomously executes complex business processes: research a market, prepare a presentation, finalise a report. No pauses, no approvals between steps.
This is not digitalisation 2.0. This is a new operating model.
The effect:
The iShares Expanded Tech-Software ETF (IGV) fell more than 21 percent from 1 January.
Broad software indices were down around 15 percent over a period of a few weeks, and around 25 percent from their twelve-month highs.
EV/Sales multiples fell from 5.6x at the end of 2025 to 4.2x by mid-March, the sharpest correction since the interest rate rises of 2022.
The difference between a CRM system and an AI agent is not quantitative. A CRM digitalised the process of managing customers. An AI agent makes decisions inside that process.
Orchestration: the control layer that separates the winners
But the picture is not entirely rosy when it comes to implementing AI agents.
Imagine a restaurant with twelve chefs. Each outstanding in their speciality: one makes a perfect risotto, another sushi you would happily pay for in Tokyo, a third bread that smells of childhood. But there is no head chef. No coordination. No concept of a menu.
The result is not dinner. The result is expensive chaos.
That is precisely what is happening in most enterprise AI implementations today.
The average company in 2026 will be using somewhere between six and twelve different AI models and agents. Marketing bought an agent for content. Finance uses another one for closing the books. Operations has three different tools. HR has a chatbot for recruitment. IT is trying, retroactively, to govern all these systems that are already running in production and that have started making decisions nobody explicitly authorised.
Everyone ordered an AI tool. Exactly as, ten years ago, everyone ordered their own SaaS. And exactly as then, nobody is coordinating the whole.
Orchestration is that missing layer. It is not another AI tool. It is a control system that governs all the AI models and agents in a company, defines the boundaries of their autonomy, records every decision in an immutable audit trail, and allows the company to understand at any moment what its AI is doing and why.
Three principles define good orchestration architecture.
Model agnosticism is not a luxury, it is a condition of survival. A year ago everyone was using GPT-4. Today there is Claude, Gemini, Grok, Llama, DeepSeek, Qwen, MiniMax. A company architecturally tied to a single AI provider is making the same mistake as one that built its entire digital stack on a single cloud provider in 2010, and then discovered what dependency means when the price list changes.
Data sovereignty is not an IT question. It is a question of jurisdiction, of regulation and, ultimately, of control. If your AI processes data on American clouds, with American models, through an American orchestration platform, who really controls your business? This is not paranoia. This is geopolitical reality.
An audit trail is not a nice-to-have. When an AI agent can approve a loan, change a price or launch a marketing campaign, every step has to be documented, immutable and transparent to auditors. The regulator who arrives a year from now will not ask what you intended your AI to do. It will ask what it actually did.
The geopolitics of AI infrastructure: whoever builds, rules
There are three parallel AI races running at the same time, with three entirely different strategies for winning.
America leads the race in models. OpenAI, Google, Anthropic, Meta. The most advanced AI models on the planet are being built in the valley between San Francisco and San Jose. But models are increasingly becoming a commodity. DeepSeek demonstrated that the performance of frontier models can be replicated for a fraction of the cost. The democratisation of model power itself is under way.
China leads the race in application. While America debates infrastructure and models, China already has AI agents woven into the daily lives of hundreds of millions of people. ByteDance's Seedance 2.0 became the global leader in AI video generation in March 2026, in the same week that OpenAI lost a billion-dollar contract with Disney and shut down its Sora project after billions in investment. Anthropic understood the game completely, betting everything on business application, with an outright tsunami of enterprise AI applications and improvements.
Europe leads the race in regulation. The AI Act, GDPR, the Digital Markets Act. Europe is writing the rules while others build the factories. The question that is not being asked loudly enough: do those rules protect European companies, or do they simply make them more dependent on infrastructure built by others?
For South East Europe, this geopolitical question is not academic. It is a question about the business model for the next decade.
Serbia: are we building for ourselves or for others?
Serbia has something many developed economies do not: exceptional engineering talent, a long tradition of technical excellence, and a strategic position between East and West. The question is what we do with those cards.
I already see two clearly different patterns in the region.
In the first, Serbian developers use Claude Code to work for foreign AI companies four times faster than before. The money arrives. But the knowledge, the intellectual property and the strategic value leave. Productivity rises, but dependency rises faster. That is the paradox: the tool that makes us more efficient simultaneously makes us easier to replace.
In the second, local companies build AI solutions adapted to regional needs, to the language, the regulation, the business culture. The orchestration is local, the models are replaceable, the data stays within the jurisdiction. AI agents are used as building blocks, but the control layer is domestic.
Kvark.ai was our answer to that strategic question. A platform for sovereign AI orchestration: model-agnostic, self-hosted, with more than sixty enterprise connectors and an immutable audit trail. Together with the LM TEK RM-4U8G, our liquid-cooled GPU server developed for workloads of this kind, we are building a complete sovereign stack: hardware, software, consulting.
I am not naive. I know that even the best local platform will not hold off the global players forever. But I also know this: a company that does not control its own AI infrastructure is a digital tenant. And a landlord can change the terms of the lease at any moment.
Three questions for every CEO
Porter's value chain gave us the language of competitive advantage for the industrial era. Today we need a new language.
In the age of the machine, the winner was whoever controlled production. In the age of digitalisation, the winner was whoever controlled the platform. In the age of artificial intelligence, the winner will be whoever controls orchestration: the layer that integrates all the models, all the data and all the agents into a coherent, governable, sovereign system.
The tools exist. The question is no longer whether AI works. The question is who runs it, in whose interest, and under whose supervision.
Three questions every CEO should ask themselves before the end of this quarter:
How many AI models and agents are currently running in my company, and who actually governs them?
Where is our data processed, and in whose jurisdiction does it sit?
Are we building our own control layer, or are we tenants on somebody else's?
Tonight, before you go to sleep, ask yourself a simple question. Not whether you are efficient, not whether you have started applying AI in your business, but whether you are on the right road at all.
The value chain served us well. But it is time for the intelligence chain.
