Models are getting better, more widely available and increasingly similar at the top end. AI agents can now do more than answer questions. They can retrieve information, use tools, move work and take action. The competitive question is therefore changing. It is no longer simply who has the best model. It is who has built the better system for putting intelligence to work.
For a short time, access to frontier AI looked like the advantage.
That period is ending.
Stanford’s latest 2026 AI Index shows that leading model performance is beginning to cluster. As of March 2026, Anthropic, xAI, Google and OpenAI were within 25 Elo points of one another on the Arena leaderboard. The report also found that AI was being used in at least one business function by 88% of surveyed organisations.
At the same time, the economics have changed dramatically. Stanford’s previous AI Index found that the cost of running a model with roughly GPT 3.5 level performance fell from $20 per million tokens in late 2022 to $0.07 by October 2024, a reduction of more than 280 times.
The model still matters. Capability, reliability, security, latency and cost all matter.
But access to capable AI is becoming the norm.
That changes where advantage sits.
Models are converging. Organisations are not.
The differences between businesses are increasingly found outside the model.
One organisation has clean data, clear decision rights and connected workflows.
Another has six AI pilots, three agents, duplicate data, unclear permissions and no agreement on who is accountable when the system acts.
They may be using similarly capable technology.
They will not get similar results.
This is becoming particularly obvious with AI agents.
Agents expose weaknesses that chat interfaces could hide. A chatbot can give a poor answer and wait for somebody to correct it. An agent may retrieve a customer record, choose a tool, update a system, route a case or trigger the next step in a process.
Suddenly the quality of the workflow matters as much as the quality of the model.
- The rules matter.
- The data matters.
- The permissions matter.
- The handoffs matter.
- The feedback loop matters.
Agents have made the operating model visible
The industry is starting to recognise this.
A 2026 Deloitte study found that only 5% of surveyed organisations described their business processes as highly prepared for AI agents, while only 15% said they had scaled orchestrated, cross functional multi agent adoption. Three quarters of surveyed leaders also agreed that collaboration between humans and agents creates more value than automation alone.
That is an important distinction.
The problem is increasingly not whether an agent can technically perform a task.
The harder question is whether the organisation is ready for the agent to perform it.
McKinsey reached a similar conclusion in its 2026 work on scaling agentic AI. It reported that many organisations had experimented with agents, but fewer than 10% had scaled them to deliver tangible value. Data limitations were cited as a major barrier.
The lesson is straightforward.
You cannot compensate for a weak operating model by adding a more capable model.
You may simply make the weakness move faster.
The operating system becomes the advantage
This is why we think about AI less as a collection of tools and more as an operating system for intelligent change.
Not an operating system in the computer sense. An organisational one.
It connects five things.
- People and channels: Who does the organisation serve? How do customers, staff, students, patients, citizens and partners interact with it?
- Journeys and workflows: Where does work actually happen? Which steps create value, delay, friction or risk?
- AI operating system: Where can agents, automation, knowledge retrieval, business rules and decision support help? Where does a human still need to approve, interpret or decide?
- Business systems: How does intelligence work across CRM, finance, collaboration, marketing, payments, industry platforms and document systems rather than becoming another isolated application?
- Data and knowledge: What facts, records, documents, policies and context does the system depend on?
Trust and governance run through all five. Refer About Us page for more details.
That is the real architecture of useful enterprise AI.
Better decisions, not simply more predictions
Much of the early AI market was built around producing things.
- Generate a paragraph.
- Generate an image.
- Generate some code.
The next phase is increasingly about decisions and execution.
An agent may gather information from several systems, compare it with policy, prepare a recommendation and start the next workflow. That can be enormously useful.
But the value is not the agent making more decisions. It is the organisation making better decisions with better context.
That requires explicit decision rights.
- What can AI decide?
- What can it recommend?
- What can it prepare?
- What requires human judgement?
- What happens when the system is uncertain?
- Who owns the outcome?
These questions become part of the product design.
Better execution, without losing control
The second advantage comes from reducing the drag around work.
Much organisational effort is consumed by coordination.
- Finding the right information.
- Moving data between systems.
- Preparing documents.
- Checking status.
- Following up.
- Routing cases.
- Reconciling information.
- Remembering the next step.
Agents can increasingly help with this work.
The opportunity is substantial because it sits between systems and between people, where conventional software often leaves gaps.
Gartner has begun describing agentic AI as a workflow manager and has highlighted the increasing importance of scale and reuse as organisations restructure work around agents.
But intelligent execution still needs boundaries.
A fast system that acts outside its authority is not better execution.
It is simply faster risk.
Better capability is harder to copy
There is a third advantage that receives less attention.
- The organisation learns.
- People become better at identifying where AI is useful.
- Rules become clearer.
- Prompts improve.
- Data quality problems become visible.
- Weak handoffs get redesigned.
- Exceptions reveal missing logic.
- Human overrides show where the system misunderstood the situation.
Over time, this creates organisational capability.
That capability is harder for a competitor to acquire than access to the same model.
Anyone can subscribe to a model.
They cannot instantly acquire your workflows, data relationships, decision rules, institutional knowledge, feedback history and understanding of where human judgement matters.
This is where durable advantage begins to appear.
The operating model is now part of the technology strategy
McKinsey’s September 2026 research makes this point directly: organisations creating more value from AI are not simply selecting a single ideal operating model. They are making deliberate choices about how decisions, work, capabilities and accountability are organised around AI.
That changes the agenda for leaders.
The question is no longer: Which AI platform should we buy?
It becomes:
- Which workflows matter enough to redesign?
- Where should AI act, assist or stop?
- What context does it need?
- Which systems must connect?
- Who remains accountable?
- How will we measure the outcome?
- What will the organisation learn when a person disagrees with the AI?
These are operating questions.
The technology enables the answer. It does not replace it.
Start with one workflow
This does not require a grand transformation programme.
In fact, the better starting point is usually smaller.
- Choose one workflow that matters.
- Understand how it works today.
- Identify the decisions.
- Find the friction.
- Make the business rules visible.
- Decide where AI can help.
- Connect the necessary systems and knowledge.
- Define where people remain in control.
- Measure the result.
- Then improve it.
- Do this repeatedly and something important happens.
You stop accumulating isolated AI experiments.
You start building an operating system.
The new AI advantage
The AI race has not ended.
Models will continue to improve. Costs will change. New architectures will emerge. Open and closed models will continue to trade advantages. Stanford’s 2026 Index itself shows how quickly technical leadership can move.
But for most organisations, winning that race is not the job.
The job is to become better at using the intelligence the market makes available.
- That means better decisions.
- Better execution.
- Better capability.
- Clearer human accountability.
And an organisation that learns each time people and AI work together.
The winners will not necessarily have the most AI.
They will have the best system for putting it to work.
Smarter systems. Human rules.
References
Stanford Institute for Human Centered Artificial Intelligence, 2026 AI Index Report
Model performance, organisational adoption and the changing competitive landscape.
2026 AI Index Report
Stanford Institute for Human Centered Artificial Intelligence, 2025 AI Index Report, Research and Development
Evidence on the rapid fall in model inference costs.
AI Index 2025, Research and Development
Deloitte, AI Agents Are Only the Beginning, August 2026
Research on agent readiness, workflow redesign and human collaboration with agents.
Deloitte agentic AI readiness research
McKinsey, Building the Foundations for Agentic AI at Scale, April 2026
Research on data, workflows and the difficulty of scaling agentic AI into measurable value.
Building the foundations for agentic AI at scale
McKinsey, The Key to AI Value Is Hiding in Plain Sight: Your Operating Model, September 2026
Research on how operating model choices shape the value organisations capture from AI.
The key to AI value is hiding in plain sight: Your operating model
Gartner, Scale, Reusability, and the Rise of Agentic AI as a Workflow Manager, August 2026
Analysis of emerging enterprise patterns around agentic workflows, reuse and scale.
Gartner research on agentic AI as a workflow manager
