In this briefing
- 01The procurement gateway starts to aggregate tools, products and services
- 02The development gateway starts to accommodate multiple Agents and enterprise-system objects
- 03Data, runtime policies and on-premises deployment enter the same platform boundary
- 04Cross-vendor reference architecture begins to describe the responsibility chain beyond the gateway
- →What to watch next
- ↗Sources and verification
- Claude Marketplace is live, bringing together more than 2,000 connectors and plugins, third-party Agents and products, and service partners; using part of an Anthropic commitment for partner products remains in limited preview.
- UiPath Coding Agents and Data Fabric entity context are GA, and Automation Suite for Linux is available; advanced agents and LLM-as-Judge Guardrail remain in Preview. The different capability states cannot be combined into a single assurance.
- Okta and several vendors formed the Blueprint Alliance and published a multi-vendor reference architecture covering Agent identity, task scope, runtime monitoring and response; the Alliance's principles do not prove that members' products are already interoperable or have passed common conformance tests.
The procurement gateway starts to aggregate tools, products and services
On 23 September, Anthropic launched Claude Marketplace. The official page divides the offering into three categories: more than 2,000 connectors and plugins; Claude-powered Agents and products; and consulting and systems-integration service partners. Connectors and plugins can be built with Model Context Protocol and Agent Skills, while third-party products and services enter the catalogue through an application or partner network. The catalogue size and partner list come from Anthropic's announcement; they are not the result of a review of every listing's security, performance or delivery quality.
On the same day, the official case-study page said that CodeRabbit expanded its Vercel plan through Marketplace, while Power Digital and ThoughtSpot placed Snowflake capacity or an inference API within their existing Anthropic commitments. The page also makes clear that using commitments in this way remains in limited preview and is subject to eligibility. Statements on budget-approval speed, usage scale and deployment come from the vendors and customers; the page provides no common procurement rules, total-cost comparison or independent audit.
For enterprise deployment, a model supplier's commercial gateway is beginning to shape the procurement path for surrounding data, infrastructure, applications and implementation services. Reusing an existing commitment may reduce the need for a new budget process, but it cannot replace supplier due diligence, data-processing review, exit-cost assessment or service-level verification; catalogue visibility, technical compatibility, compliance approval and production readiness remain distinct states.
The development gateway starts to accommodate multiple Agents and enterprise-system objects
On 23 September, UiPath announced that UiPath Coding Agents, first announced in May (month 5), are now at UiPath Platform GA. The company says enterprises can use UiPath Coding Agents, Claude Code, Cursor, Codex, Antigravity or other supported Agents to access UiPath folders, jobs, queues, assets, audit logs and connections in a logged-in session. Here, 'supported' depends on UiPath's platform integration; it does not mean that these Agents have identical capabilities across every version, permission configuration and business process.
The same announcement also introduced UiPath Delegate and extended coding agents to document-extraction projects. Delegate can save a tester's actions as reusable routines, but final judgement, test strategy, quality decisions and sign-off remain with people. The announcement provides no like-for-like comparison of success rates, error recovery or permission isolation between coding agents, and does not prove that natural-language access will automatically comply with every business authorisation.
If an enterprise automation platform can let multiple Agents recognise the same jobs, assets and audit objects, it may lower the cost of discovering interfaces for each integration and make existing process objects part of the Agent runtime context. Its value still depends on session identity, supported versions, permission mapping and correct result verification; a unified gateway cannot automatically standardise the behaviour, responsibilities and failure modes of different Agents.
Data, runtime policies and on-premises deployment enter the same platform boundary
UiPath's announcement lists Integration Service capabilities including Database Hub, Relay, a third-party credential store and MCP Connector. It says that native entities in Data Fabric are now GA as Agent context and can be reused across automations, applications and workflows. Federated Entities are described as zero-copy connections to external sources including Salesforce and SAP, with an audit trail; these are vendor-published capability boundaries, with no evidence on cross-source semantic consistency, latency or independent security testing.
On governance and deployment, UiPath also lists Model Hub, Runtime Checker, LLM-as-Judge Guardrail, Compliance Packs, and Identity & Access Policies. The official release record for month 9 marks Data Fabric entity context and building Agents through Coding Agents as GA, while advanced agents and LLM-as-Judge Guardrail are in Preview; the announcement separately says that Automation Suite for Linux, containing the full Agentic AI stack, is now available. The different states cannot be combined, and the public material gives no offline operating conditions, resource requirements or independent compliance certification.
The enterprise deployment gateway is extending from Agent construction to data context, credentials, runtime policy and infrastructure location. Placing these objects within one platform may reduce fragmented configuration, but 'zero-copy', 'on-premises deployment' and 'runtime checking' answer different questions and cannot substitute for one another; whether data leaves the source system, whether policy actually blocks an action, whether logs are complete and whether the system can operate offline still need to be verified in the target architecture.
Cross-vendor reference architecture begins to describe the responsibility chain beyond the gateway
On 22 September, Okta and vendors including AWS, CrowdStrike, Databricks, Docker, Google Cloud, Salesforce and ServiceNow announced the formation of the Blueprint Alliance. The Alliance expands an existing blueprint into an open, multi-vendor reference architecture organised around four questions: where Agents are, what they can do, what they are doing and how to respond when they deviate. Its principles include treating each Agent as a distinct identity, limiting access to the task, retaining the delegation chain, continuously monitoring runtime behaviour, and supporting immediate and reversible containment.
The announcement positions the architecture as a shared direction for identity, data, application, infrastructure and security vendors, and says that further members will be added as the blueprint is refined. It remains a set of principles and a reference architecture published by the Alliance; the public page provides no final common protocol, interoperability-test results for member products, conformance certification or cross-platform failure-drill results.
As procurement catalogues and automation platforms concentrate more Agents, tools and data behind a small number of gateways, enterprise governance has to cover the identity, delegation, execution and response chain beyond those gateways. A reference architecture can provide a checklist of acceptance questions, but it cannot by itself prove that a Marketplace listing, platform connector or combination of Alliance members is secure and usable; commercial aggregation, technical integration and cross-vendor governance still require separate evidence.
What to watch next
- Whether Claude Marketplace will publish its listing reviews, permission declarations, exit mechanisms and rules for using committed spend, and move the limited preview for buying partner products with committed spend into a verifiable formal state.
- Whether UiPath will separately state the availability and version compatibility of new controls such as Runtime Checker, the resource requirements for on-premises Linux deployment, and permission and failure comparisons across coding agents.
- Whether the Blueprint Alliance will publish executable protocols, common conformance tests and cross-vendor exercises showing that the reference architecture can be implemented in real product combinations.
Sources and verification
Golden Data has edited this briefing from the public materials listed above. The original sources govern facts and figures. The enterprise relevance sections are Golden Data editorial analysis and do not constitute an endorsement of any third-party product.
← Back to AI Daily Briefing