AI Enablement Radar week 39: get the AI report to its recipient

AI can produce a useful report and still leave a colleague to move it into the right system. This week's updates from Google, Anthropic, and GitHub make that handoff worth another look: connect existing tools, choose whose account does the work, and follow the result through to its recipient. For a Nordic team, the next experiment is concrete: get one recurring report all the way to review and delivery.

Week 39 covers September 21–27, 2026. We also include HubSpot's September 16 launch, current customer stories without publication dates, and EU guidance from August. These are context, not this week's announcements. Sources appear beside each item.

Top signals this week

• Google Workspace Studio began rolling out more ways to connect workflows to other services on September 21. Custom starters and steps, third-party integrations in beta, and outgoing webhooks are included. The features are off by default and require administrator enablement. Source: Google Workspace Studio: integrations and webhooks.
• Anthropic opened a Claude plugin submission portal on September 25. Developers can follow review progress and see usage analytics after publication. Reusable integrations get a clearer distribution route. Source: Anthropic: Build plugins for Claude.
• Google Cloud announced a public preview on September 24 that turns existing REST APIs into MCP tools through API Gateway. MCP, Model Context Protocol, is a protocol that lets AI discover and use external tools. Source: Google Cloud: REST APIs as MCP tools.
• Google Meet added Quick notes on September 22: a one-page summary of actions, outcomes, and talking points. The longer record remains under Full notes. Source: Google Meet: Quick notes.
• GitHub added OpenTelemetry to the Copilot app on September 22. It is a standard way to collect operational data, here including sessions and tool calls. Prompt and response content is excluded from exports by default. Source: GitHub: OpenTelemetry in the Copilot app.

What organizations are actually doing with AI

Layer: reporting is fast, delivery still takes time

In Anthropic's customer story, the 25-person Norwegian agency Layer describes completing a client report quoted at ten hours in 20 minutes using Supermetrics and Claude. Of those minutes, 17 went into putting the result in the client's Google Sheet. These are customer-reported figures, not an independent benchmark. The page has no visible publication date.

Measure the work after the AI response too. If analysis is fast but somebody still copies numbers and formats the delivery, investigate that handoff next.

Source: Anthropic: Supermetrics and Layer.

Qonto: prepare the payment, let the customer approve

Qonto describes AI agents preparing invoices and payroll transfers inside its existing app. A batch of 15–20 payslips can become a prepared bulk transfer; the customer approves before money moves. The undated story describes a project that began in November 2025. It illustrates an operating pattern, not availability in Sweden.

For a finance manager, the transferable idea is to put preparation and approval in the same place. AI does not need the final decision to remove a lot of administration.

Source: Anthropic: Qonto financial administration.

The tooling layer: platforms, agents, and workflows

Connect the report to the recipient's tools

Workspace Studio names HubSpot, Slack, Asana, and QuickBooks among its integrations. A webhook sends an event or information from one service to another. For teams already using Google Workspace, this may provide a route from incoming material to a reviewed delivery. Custom steps may need Apps Script assistance. Scheduled Release begins receiving the update on September 30; check the features in your account before building the routine.

Source: Google Workspace Studio: integrations and webhooks.

HubSpot's September 16 Fall Spotlight adds another example: Breeze Assistant gains more actions tied to the customer system, and Deal Progression includes one-click approval. A sales team could test turning meeting material into a proposed customer-record update and a follow-up draft. Check the plan and availability for the feature you choose.

Source: HubSpot: Fall 2026 Spotlight.

Decide whose identity the integration uses

Claude Tag gained personal connector support in Slack channels on September 24. A request can use the requesting person's connectors, while unattended and scheduled runs still need shared channel connectors. Team gets the feature first; Enterprise follows later. Users can choose to review responses before posting. A response posted to the channel is visible to everyone in it.

This affects the design of a shared report: choose an owner and an intended destination, so the routine does not accidentally depend on a colleague's private access.

Source: Anthropic: Claude Tag personal connectors.

For organizations with their own systems, Google Cloud's MCP support is a more technical option. Existing authentication, quota, and logging policies apply to API calls. Tool discovery is open by default, however; configure its access separately before production. This is an integration project for the system owner, not a setting an end user should have to guess.

Source: Google Cloud: REST APIs as MCP tools.

Governance and risk: what needs to be in place before scaling

Review Meet before September 29. Google's new automatic note-taking settings for meetings with at least three people start taking effect then. The default is on for Business Standard and Plus, but off for Enterprise Standard and Plus, Frontline Plus, and Google AI Pro for Education. Earlier Gemini Alpha settings may differ. Ask your administrator to check the actual configuration, participant information, and who receives the notes.

Source: Google Meet: settings taking effect September 29.

Make the handoff traceable. Decide who approves an external delivery, which account writes to the destination, and where the outcome is logged. Use scoped permissions and a secret manager or environment variables for keys; redact unnecessary personal data from logs. GitHub's new telemetry export is an example for development teams, not a requirement for every office to adopt the same technology.

Source: GitHub: OpenTelemetry in the Copilot app.

Separate internal drafts from publication. The European Commission's Article 50 guidance, updated August 6, describes transparency obligations applying from August 2. Duties differ between providers and deploying organizations and cover specified AI interactions and content types. Check the applicable requirements and exemptions when a routine publishes material. A blanket claim that all AI content must always be labelled would be misleading.

Source: European Commission: Article 50 transparency guidance.

This week's practical Hammer test

Set aside 40 minutes for a report you already deliver each week. The goal is a finished review copy where the recipient works. Use an approved AI tool and existing permissions; if an integration is missing, produce a concrete connection proposal rather than claiming automatic delivery.

1. Spend five minutes choosing the report, its source, and the recipient's tool. Write down what the recipient should be able to decide after reading it.
2. Spend ten minutes reviewing the previous delivery. Note where someone copies, formats, or waits for approval. Choose one handoff to improve.
3. Spend fifteen minutes having AI prepare a review copy with source links. Use an already approved connector if one exists. Have a colleague check figures, reporting period, and omissions before any external delivery.
4. Spend ten minutes asking the recipient whether the report is immediately usable. Record the remaining manual steps and who owns the next attempt.

Suggested prompt:

Help us improve the delivery of this recurring report. Read the material first and ask which tool the recipient uses if that information is missing. Identify one manual handoff and propose how to remove it using our existing tools. Then create a review copy with sources and clearly marked gaps. State what still needs to be done manually. Do not send or publish anything without our approval.

The test succeeds when the recipient can review the material without rebuilding it first. If integration needs more work, you still have a concrete brief: source, destination, fields to transfer, and who approves.

Companies and tools to watch

• Supermetrics: follow how sourced data reaches the client's delivery format. Source: Anthropic: Supermetrics and Layer.
• Google Workspace Studio: examine whether existing tools can connect without a separate automation system. Source: Google Workspace Studio: integrations and webhooks.
• Anthropic: follow the plugin portal and which integrations become available for your workflow. Source: Anthropic: Build plugins for Claude.
• HubSpot: test whether follow-up can stay in the customer system rather than a separate chat draft. Source: HubSpot: Fall 2026 Spotlight.

If the report works but its handoff still needs manual effort, that is a concrete Tool Forge assignment: connect the source to the recipient's workflow and keep review where it helps.