Machine-Service Market Snapshot — 24 August 2026

Autonomous agents can already discover and buy machine services across multiple marketplaces and indexes.

The harder question is what the market actually looks like when those listings are treated as machine contracts rather than as search results.

Atinamos maintains a private evidence catalogue of machine services observed across agent-commerce sources. The underlying catalogue is not published. This snapshot publishes aggregate counts and methodology so that the claims can be examined without exposing the commercial dataset itself.

Snapshot

As observed by Atinamos on 24 August 2026:

These are Atinamos observations, not claims about the total size of the global agent-service market.

Counts reflect records ingested at the snapshot time. Marketplaces can change between observations.

Source coverage

Resolved source records in the catalogue at snapshot time:

These values must not be added together and described as unique services.

A single canonical service can appear in more than one source.

Source coverage is also not equivalent across platforms. Some expose catalogue-style interfaces, while others expose indexes or ranked/search surfaces.

In particular, the observed 402.ad interface reported a larger result total than it would expose through the search interface during testing. Its records are therefore treated as search-exposure evidence rather than an exhaustive marketplace catalogue.

Virtuals ACP is tracked separately as ranked discovery evidence rather than as an exhaustive catalogue. The snapshot contains 9 ACP discovery runs and 90 ranked results.

Cross-source presence

Of the canonical services with resolved source evidence:

A service appearing in multiple marketplaces is evidence of distribution.

It is not, by itself, independent verification of service quality or fulfilment.

Observed HTTP methods

The canonical register contained:

Atinamos does not infer a precise HTTP method where the source evidence does not support one.

Where listing identity cannot safely be resolved to a canonical executable service, the listing is retained separately as unresolved evidence.

At snapshot time that unresolved register contained:

Total unresolved listings: 749.

This follows a core Atinamos evidence rule:

Observed does not mean resolved.

Missing contract identity should be preserved as missing evidence, not filled in by assumption.

Categories: useful, but not yet canonical

Marketplace-supplied category labels are currently too inconsistent for Atinamos to publish a defensible cross-market category distribution as though it were a single taxonomy.

The snapshot contained 3,495 distinct non-empty source-supplied category labels.

The most frequent labels included:

Category coverage also varied materially by source:

For that reason these labels remain marketplace provenance rather than an Atinamos-assigned canonical classification.

A future normalized taxonomy should preserve the original marketplace label alongside any Atinamos-derived classification.

Longitudinal evidence

The catalogue is not intended to be only a directory.

At snapshot time Atinamos had retained 71,817 resolved source observations.

Those observations provide the basis for detecting changes such as:

• appearance or disappearance from a source;
• price changes;
• contract or schema changes;
• provider metadata changes;
• marketplace-supplied reputation changes;
• changes in ranked discoverability.

Longitudinal evidence matters because an autonomous buyer is making a decision about the service that exists now, not merely the service that once appeared in a directory.

What the figures do not prove

This snapshot does not claim that:

• 60,234 represents every machine service in existence;
• every observed listing is currently operational;
• marketplace presence proves fulfilment quality;
• multiple listings constitute independent trust evidence;
• marketplace reputation scores are Atinamos verification;
• an unresolved listing is unsafe.

External marketplace claims remain attributed to their source.

Independent Atinamos verification is recorded separately.

Why maintain this catalogue?

The commercial question behind the work is simple:

Can an AI agent become a prudent customer?

To answer that, a buyer needs more than service discovery.

It needs to be able to distinguish:

what is advertised → what can be resolved into an executable contract → what has changed → what has actually been independently observed.

That is the role of the Atinamos machine-service evidence catalogue.

Snapshot date: 24 August 2026
Canonical services: 60,234
Underlying catalogue: private
Published data: aggregate observations and methodology only

You can pay to be tested. You cannot pay to be trusted.