Building the intelligent organisation in the agentic age

Executive summary

• The first wave of AI improved productivity; the next wave is reshaping how we make decisions.
• As agents move from answering prompts to monitoring, reasoning, recommending and acting on our behalf, the centre of gravity shifts from model access to decision architecture.
• The strongest organisations will be outcome-led: they will start with the business result, identify the decisions that drive it, and build the context, governance and human accountability required to improve it.
• The next frontier will be economic and organisational: creating more value per unit of intelligence consumed, with better context, less waste, stronger governance, and more resilient architecture.
• Competitive advantage will belong to companies that turn these capabilities into a durable performance edge: better decisions, better execution and better outcomes at scale.

‍

Bloomberg recently noted[1] that corporate AI language has moved on. Mentions of "generative" AI on S&P 500 earnings calls are down around 77% from their 2023 peak, while the broader term "AI" was used more than 6,000 times in the latest quarter. "Agentic" has become the new label. Some of this is fashion. Some of it reflects a real shift. An agent can take action on a user's behalf, rather than simply explain what action the user might take.

Deployment remains early. Outside technology, the Financial Times reported[2] that only 9% of businesses have active agents across multiple parts of the organisation, while 17% are running pilots. Early adopters are already learning the central lesson: agents create value when they are inserted into the right process, with the right boundary between automation and human judgment, and applied to use cases that are critical for the business or function.

This requires critical thinking before deployment. Faster work does not always mean better performance. AI can also create longer conversations, more outputs to review, duplicated efforts and polished low-value work. In a recent example, Adecco began by analysing each step of recruitment to identify where an agent added value and where a human remained essential; the result was a 20% time saving in part of the process.

‍

The centre of gravity moves from tools to decisions

The first AI wave accelerated knowledge work. Search became conversational, reporting became automated, content became abundant. The agentic wave changes the rhythm of decision-making.

Traditional analytics is largely request-driven: a human asks, the system answers, a human decides. Agentic systems run in loops: a trigger appears, the agent interprets it, takes or recommends action, observes the result, and iterates. The Harvard Data Science Review describes this as a move from reactive reporting to proactive, goal-driven action; it also cites a forecast that 50% of business decisions will be augmented or automated by AI agents for decision intelligence.[3]

The practical impact can already be seen in operational settings. In one supply-chain case[4], an agentic system reduced manual purchase-order tracking from five to six hours per person per day to half an hour, cut escalations by 99%, and monitored exceptions 24/7 rather than during expert working hours. The point is less the use case than the pattern: agents create value when they compress the loop between signal, interpretation, action and learning.

For CEOs, the useful unit of analysis is the decision loop, not the agent. What outcome matters? Which decision changes it? Which signals and constraints shape that decision? Which parts can be automated, augmented or reserved for human judgment? This is the foundation of an outcome-led approach.

‍

The bottleneck moves from analysis to judgment

AI makes the front end of decision-making cheaper[5]: gathering information, structuring it, recognising patterns, producing a first recommendation. As that work accelerates, the constraint moves downstream. Interpretation, arbitration and execution carry more weight.

A marketing agent can recommend a budget shift. It cannot settle the trade-off between short-term ROI, brand equity, channel conflict and customer lifetime value without a business frame. A pricing agent can identify margin opportunities. It cannot decide how far the company wants to test customer trust or retailer relationships. Intelligence without context produces plausible answers; contextualised reasoning produces useful decisions.

The human layer remains highly variable. MIT Sloan research[6] found that, when senior executives received identical AI advice, some invested up to 18% more in the same strategic initiative depending on their decision-making style. The article groups executives into skeptics, interactors and delegators, showing that AI decision quality depends on how humans interpret and act on the recommendation.

As models become more capable and more widely available, the hard part becomes the ability to capture, structure and use context: understanding the business, translating its reality into data and models, and interpreting outputs in light of strategic and operational constraints. The advantage shifts to organisations that can turn models into decision levers, encode constraints, and keep human judgment close to the places where ambiguity matters.

‍

Local optimization becomes easy. Enterprise coordination becomes hard.

Most AI deployments still sit inside functional boundaries: a campaign optimiser, a demand forecast, a lead-scoring model, a pricing recommendation, a service automation workflow. As agents become more capable, each function can optimise its own objective faster. The enterprise can become locally smarter and globally less coherent.

A media agent may raise short-term ROI while starving brand equity. A pricing agent may protect margin while slowing acquisition. A supply-chain agent may improve resilience while tying up working capital. These are not model failures. They are business trade-offs.

At scale, advantage comes from the quality of orchestration: the ability to turn many local decisions into coherent enterprise action. This requires a reasoning layer between models and execution: the layer that carries business context, constraints, priorities and decision rights across functions. It determines what should be optimised, what can be delegated, what must be escalated, and how competing objectives are arbitrated.

The intelligent organisation designs this layer deliberately. It lets agents act locally, but reasons enterprise-wide.

‍

The intelligent organisation operates around outcomes

Large organisations were built around functions. Agentic AI pushes them toward decisions and outcomes. The intelligent organisation has four characteristics.

First, decisions become continuous. Performance is monitored in motion, weak signals are detected earlier, and scenarios can be simulated before the next quarterly review. The organisation moves from static snapshots to more dynamic decision loops, closer to how business conditions actually evolve.

Second, knowledge becomes operational. Expertise no longer sits only in individuals, playbooks, slide decks, emails or scattered files. Decisions, past recommendations, business rules, benchmarks, prompts, skills and post-decision feedback become structured, accessible and reusable. Over time, knowledge management moves from storing documents to making enterprise knowledge and memory available inside workflows.

Third, systems become connected. Agents need access to enterprise reality: ERP, CRM, media platforms, commerce systems, supply-chain tools, collaboration channels and external data. Poor integration quickly becomes poor judgment. In the supply-chain example above, the article stresses that weak API connections, inconsistent event definitions and conflicting supplier or carrier feeds can cause agents to reconcile noisy signals and propagate errors into operational tools.

Fourth, leadership evolves. AI accelerates execution faster than many management systems can absorb. HBR reported[7] that 89% of leaders say AI has accelerated the speed of work, 87% of knowledge workers say teams lack time or capacity to coordinate when everyone is in execution mode, and 54% of managers report receiving "AI workslop": polished output without substance.

Leadership therefore becomes more directional, more selective and more accountable. Managers need to identify where value is created, which decisions can be automated or augmented, how roles and responsibilities should evolve, and how knowledge, prompts, skills and agents are governed across the organisation. They also need to remain disciplined on the cost-benefit equation: more agents, more tokens or more outputs only matter if they translate into measurable business impact. In an intelligent organisation, leadership governs the system of decisions.

‍

New constraints define the next frontier

Scaling agents creates new constraints. They are less glamorous than demos, and more decisive for enterprise value.

Trust. People need to know when to rely on agents, when to challenge them and when to override them. Trust grows through traceability, visible assumptions, feedback loops and clear accountability. It also requires awareness of human bias toward AI: the skeptic can reject useful recommendations; the delegator can outsource responsibility to the machine.

Governance. Decision rights need to be explicit. Some decisions can be automated because they are frequent, reversible and low risk. Others should be augmented because they require context or carry material consequences. A small number should remain human-led because ambiguity, ethics, brand, regulation or long-term positioning dominate the calculation. Governance should classify decisions before teams build agents around them.

Impact accountability. Agents reshape the work, risks and responsibilities around decisions. A pricing agent that tests customer trust too aggressively, a hiring agent that reproduces historical bias, or a supply-chain agent that optimises cost while weakening supplier standards can create material reputational and business risk. The intelligent organisation maps consequential decisions in advance, assigns clear accountability, and makes automated reasoning visible and auditable.

Economics, frugality and sustainability. Agentic AI introduces a new operating cost: intelligence consumption. Recent academic work on agentic coding tasks[8] found that they used around 1,000 times more tokens than code reasoning and code chat tasks, with average token usage of 4.17 million per task and an average task cost of $1.857. The same study found that token use can vary sharply, runs on the same task can differ by up to 30 times, and higher token use does not reliably improve accuracy.

That changes the economics of AI at scale. The issue at stake becomes how much business value is created per unit of intelligence consumed. Frugality becomes a design principle: use the smallest effective model for a specific task, reduce unnecessary context, cache intelligently, route tasks across deterministic software, statistical models and generative systems, and define stopping rules. At scale, every decision loop has an environmental signature: compute, energy, water and carbon.

Adaptability. The model landscape will keep changing. Resilient organisations will separate decision logic from any single model, keep architectures modular, instrument agent performance, maintain reusable components, and design for evolution. The aim is not to guess the winning model but to build an organisation that can absorb better models, new regulation, new interfaces and new agent capabilities without creating a new layer of technical debt, cost or risk each time.

‍

Conclusion: organising intelligence

At Ekimetrics, our conviction is that the winning pattern in the agentic age will be a hybrid, outcome-led decision system powered by agents, data, business rules and human expertise.

Agents will matter. Context, orchestration and knowledge management will matter more. The organisations that outperform will be those that know which outcomes they are optimising, which decisions drive them, what context those decisions require, and how humans and agents should work together.

The last twenty years were about digitising information. The next decade will be about organising intelligence. Advantage will belong to companies that turn intelligence into outcome-led action at enterprise scale.

‍

Sources

1. Brody Ford, AI Used to Be Generative. Now It's All About Agents, Bloomberg, 2026.
2. Isabel Berwick, Lessons from the Agentic AI Trailblazers, Financial Times, 2026.
3. Monisha Athi Kesavan Premalatha, AI Agents Are Transforming Decision Making: What Leaders Should Know, Harvard Data Science Review / MIT Press, 2026.
4. Ibid.
5. Tomas Chamorro-Premuzic, How AI Is Changing Decision Making in Organizations, Forbes, 2026.
6. Philip Meissner and Christoph Keding, The Human Factor in AI-Based Decision-Making, MIT Sloan Management Review, 2021.
7. Liz Fosslien and Mollie West Duffy, Managers Are Struggling to Keep Up with the AI Productivity Boom, Harvard Business Review, 2026.
8. Longju Bai, Zhemin Huang, Xingyao Wang, Jiao Sun, Rada Mihalcea, Erik Brynjolfsson, Alex Pentland and Jiaxin Pei, How Do AI Agents Spend Your Money? Analyzing and Predicting Token Consumption in Agentic Coding Tasks, University of Michigan, Stanford University, All Hands AI, Google DeepMind, Microsoft AI & MIT, 2026.

‍

To read more about this subject, explore more articles here and here.