Most discussion about AI agents focuses on intelligence.
Which model reasons better? Which agent can code, browse, negotiate, or plan more effectively?
I think another variable may become just as important: **how much of the digital world an agentic organization can safely and economically operate.**
I use the working term **ADS — Agentic Digital Surface** for one distinct digital action domain: a place where an agent can observe, decide, or act in a way that matters.
An email account is not necessarily one ADS. Reading messages, sending messages, changing filters, managing contacts, and handling calendar invitations may each have different authority and consequences.
A CRM is not necessarily one ADS either. Lead discovery, customer history, forecasting, follow-up, pricing, and opportunity state can be separate action domains.
The point is not to inflate the count.
The point is to measure the territory.
Businesses already have very large digital estates
The average internet user operates across dozens of apps and accounts. Businesses operate across far more.
Zylo’s 2026 data reports an average company portfolio of roughly 305 SaaS applications, rising to about 696 for large enterprises. That is before decomposing each application into the different action domains, permissions, workflows, identities, payment surfaces, and data systems inside it.
At the same time, AI agents are beginning to operate across those systems.
NIST’s 2026 work on software and AI-agent identity explicitly focuses on the risks created when agents gain access to diverse datasets, tools, and applications. Okta’s 2026 research similarly frames agentic AI as a new class of non-human identity operating at machine speed, while many organizations still lack mature governance for those identities.
This suggests a different way to think about agentic capability.
The future advantage may not belong only to the organization with the smartest model.
It may belong to the organization that can govern the largest economically useful digital surface.
Raw surface is not enough
More connectivity is not automatically better.
A thousand integrations without authority boundaries, auditability, or economic purpose are just a larger attack surface.
So I separate the idea into layers:
**ADS** — a distinct digital action domain.
**G-ADS** — a governed ADS: authority, effect boundaries, and receipts are explicit.
**C-EADS** — a candidate economic ADS: there is a plausible economic pathway, but value is not yet demonstrated.
**E-ADS** — an economically evidenced ADS: productive or protective economic utility has actually been demonstrated.
And then an even harder category:
**E-ADS Revenue** — an ADS with a receipt-bound causal link to settled revenue from an independent buyer.
That distinction matters because agentic systems can easily confuse “connected”, “built”, and “valuable”.
They are not the same state.
The organizational shift
Traditional organizations expand their footprint partly by adding people.
More people create more accounts, relationships, workflows, coverage, and operating capacity.
An agentic organization may begin to decouple digital reach from human headcount.
One founder could theoretically coordinate a persistent machine organization whose agents operate across research, development, customer discovery, communications, procurement, payments, support, analytics, and infrastructure.
The agents themselves do not need to be permanent.
The persistent object is the organization:
- identity;
- authority;
- goals;
- capital;
- knowledge;
- relationships;
- digital assets;
- receipts;
- state.
Agents become temporary workers operating portions of that estate.
This leads to a metric I find more interesting than “how many agents do you have?”
> **How much economically useful governed digital territory can the organization operate per unit of human attention?**
That is closer to organizational leverage.
From a stack of tools to an economic metabolism
A mature agentic organization should not merely connect to many services.
Its surfaces should compose into loops.
Research can find demand.
Demand can produce a product requirement.
A build system can produce the artifact.
A distribution surface can reach a buyer.
A relationship surface can preserve the opportunity.
A payment surface can settle the transaction.
Support can generate feedback.
Feedback can flow back into research and product.
That loop is more important than the number of connectors inside it.
A useful agentic organization therefore looks less like a chatbot with plugins and more like a small digital economy.
This is where the “city-state” metaphor becomes useful for me.
A market contains customer and opportunity surfaces.
A workshop contains build and delivery surfaces.
A treasury contains resource and payment surfaces.
An embassy contains relationship and communication surfaces.
A hall of records contains receipts and provenance.
The city expands when it can occupy new digital territory without losing constitutional control over who may do what.
What ANANS is testing
Inside ANANS, we have started treating this as an experimental measurement problem rather than a branding claim.
Our first Reality-Sieved internal baseline currently identifies:
- 18 admitted ADS;
- 17 governed ADS;
- 13 candidate economic surfaces;
- 4 surfaces with evidenced economic utility;
- 2 currently active evidenced economic surfaces;
- 0 autonomous economic surfaces;
- 0 revenue-bearing economic surfaces.
That is not presented as an exhaustive count, and it is not evidence that ADS predicts revenue.
In fact, the most useful thing about the measurement is what it prevents us from pretending.
ANANS has a relatively strong internal loop for research, governed construction, validation, and promotion.
The external economic loop is much weaker.
Discovery, distribution, persistent buyer relationships, merchant routing, settlement, and customer feedback remain comparatively open.
So the next rational move is not “add more agents”.
It is to close the economic loops.
A possible business category
This also suggests a service question.
Most organizations know roughly how many tools they use.
Far fewer can answer:
- Which digital surfaces can our agents actually read?
- Which can they act on?
- Which have explicit authority boundaries?
- Which are economically important?
- Which require human approval?
- Which actions leave usable receipts?
- Where would expanding agent access create real leverage rather than risk?
A first version of an **Agentic Surface Audit** would not need access to the client’s credentials.
The client can supply an inventory of tools and objectives.
The audit can map the estate, classify actionability and authority, identify economic surface gaps, and propose a bounded delegation architecture.
Whether businesses will pay for that remains an open question.
That is exactly why it should be tested before building a large software platform around it.
The larger thesis
My current hypothesis is simple:
> **As agents become cheaper and more capable, governance—not model intelligence—may become the limiting reagent for expanding economically useful digital surface.**
If that is true, the organizations with the greatest advantage will not necessarily have the most agents.
They will have the largest **governed, composable, economically productive surface** relative to human attention and risk.
That is the territory I think is worth measuring.
And then expanding.
---
Notes
ADS, G-ADS, C-EADS and E-ADS are working terms used in ANANS research. They are not presented as industry standards.
External context for this introduction includes Zylo’s 2026 SaaS portfolio data, NIST’s February 2026 concept paper on software and AI-agent identity and authorization, and Okta’s 2026 Businesses at Work and AI Agents at Work research.
No claim is made here that ANANS originated the broad category of agent governance, personal AI operating systems, or cross-application AI agents.