AI Enablement Radar week 46: turn working material into a service

This week's AI updates make it easier to give colleagues and customers a useful service built on existing material. Google is starting to let teams publish a separate view from a spreadsheet and share reusable AI instructions. Northern Trust and KT, meanwhile, show how much work goes into connecting AI to real business rules. Week 46, November 9–15, is about turning working material into something other people can use.
Top signals this week
- Google Sheets gains publishable canvases: interactive mini-apps with their own links and audience settings. Visitors do not need access to the source spreadsheet. Source: Google Workspace: publish a Sheets canvas.
- Google Workspace gains skills, reusable instructions with rules, templates and reference files. Shared skills become separate copies, making version ownership important. Source: Google Workspace: skills.
- Northern Trust reports that preparing new clients' transaction data fell from roughly two and a half weeks to three days, including review and validation. Source: Microsoft: Northern Trust.
- GitHub is rolling out Gemini 4 Hydrogen in Copilot. Enterprise administrators control model access, and usage is charged at provider list prices under usage-based billing. Source: GitHub: Gemini 4 Hydrogen in Copilot.
What organizations are actually doing with AI
Northern Trust builds rules around financial work
In Microsoft's November 9 customer story, Northern Trust describes how finance experts and developers built a rule library to interpret transaction histories in different formats. Unclear entries go to people, and the bank completes its usual reconciliation before the client goes live. The time measurement includes analyst review and exception handling.
A smaller finance team can borrow the approach: collect the terminology and exceptions that already govern the work, then use them to shape the integration. The bank's result is a reported customer outcome with a described measurement method but no stated sample size.
Source: Microsoft: Northern Trust.
KT gives operators faster evidence
South Korean operator KT reports that AIO-Net detects network issues in two to three minutes, compared with more than ten minutes previously. Investigation and reporting average five minutes. The system covers 70 critical equipment units; 280 is the next planned step.
Agents can run approved read-only commands and recommend actions. The application enforces that boundary, while people decide and act. This offers a concrete model for connecting AI to operational data while keeping decisions with responsible staff. The figures come from KT in Microsoft's November 11 article.
Source: Microsoft: KT.
Saint-Gobain gathers research for the next experiment
Saint-Gobain describes how Microsoft Discovery helped researchers assemble 62,000 patents and 44,000 abstracts into a knowledge base in a couple of hours. Its research manager estimates that the same compilation would otherwise have taken a year. The work produced possible molecules and hypotheses for lower-carbon cement additives.
The article was published on November 10 and describes earlier work. The year-long comparison is the researcher's estimate, and the results are candidates for further research. Other knowledge-intensive teams can apply the approach by having AI prepare reviewable evidence for the next expert assessment.
Source: Microsoft: Saint-Gobain.
The tooling layer: platforms, agents, and workflows
Publish a view other people can use
A canvas in Google Sheets is an interactive view or mini-app built on spreadsheet data. The new publishing option provides a separate link, its own audience and a choice between automatic updates and a static snapshot. Visitors can interact with the view without editing the source sheet.
A training team could display a course calendar while coordinators keep working in the spreadsheet. Check the information exposed by the view before sharing it. Separate access to the spreadsheet does not replace that review.
Publishing happens on the web and requires edit access plus the organization's permission to publish to the web. Rapid Release rollout began November 11; Scheduled Release starts December 2. Business Standard and Plus are among the eligible plans.
Source: Google Workspace: publish a Sheets canvas.
Share the instruction with clear version ownership
Workspace skills work in Gmail, Docs, Slides, Drive and Chat through a mention using @. They currently support English instructions only and are not available in Sheets or Meet. Shared copies do not synchronize centrally, and access to reference files in Drive must be arranged separately.
This suits a shared proposal structure or document-review routine. Give the instruction an owner and version date, so colleagues know when a copy needs updating. Rapid Release began November 9; Scheduled Release starts November 30.
Source: Google Workspace: skills.
Gmail is also expanding access to AI summaries of long threads and non-personalized drafting assistance. The expansion covers eligible Workspace users aged 18 or over and requires smart features to be enabled. Rollout began November 10 and can take up to 15 days. Check what the team's existing account provides before buying another writing tool.
Source: Google Workspace: Gemini in Gmail.
Choose the model for the task
GitHub's November 13 announcement covers gradual access to Gemini 4 Hydrogen across Copilot environments, including IDEs and coding agents. For teams building their own published view, the next step is to compare a concrete change against the same acceptance criteria and check the cost. The product's model list and your organization's model policy determine what you can select.
Source: GitHub: Gemini 4 Hydrogen in Copilot.
Governance and risk: what needs to be in place before scaling
Google is making it easier to confirm an automatically assigned classification label in Drive, Docs, Sheets and Slides. When a single label is assigned, authorized users can select Accept or Edit directly in the banner. Multiple labels still require review in the side panel. The feature requires existing classification and began rolling out on November 9.
Source: Google Workspace: classification labels.
As Swedish background, Digg's procurement guidance recommends that public organizations start with existing services and agreements and examine data rights, storage and exit terms. For an integration, the practical translation is a named owner, scoped permissions and documented terms. Keep keys in environment variables or a secret manager, redact sensitive log fields and put approval before external publication.
Source: Digg: procuring generative AI under Swedish public procurement law.
For schools, the European Commission's 2026 guidance provides questions and classroom scenarios for assessing AI use. Use them to examine learning benefits and the teacher's review role when introducing a service. This guidance and Digg's advice are background to this week's news.
Source: European Commission: AI guidelines for educators.
This week's practical Hammer test
Set aside 40 minutes to make a course calendar or project overview understandable to a colleague who does not work in the source spreadsheet. Use a canvas in Sheets if your account has the feature. Otherwise, a reviewed view or export in your existing tool is enough to test the approach.
- During the first 5 minutes, choose the audience and write the two questions the view should answer, such as when the next course runs and who is responsible.
- Over the next 10 minutes, ask AI to propose the required columns and layout. Keep internal working information in the source material.
- Spend 15 minutes comparing the view with the original. Check dates, empty fields and a changed entry. Decide whether updates should happen automatically or after review.
- Use the final 10 minutes for a recipient test with the correct permissions. Ask the colleague to answer the questions without your help. The responsible person approves any publication and records the link, audience and update owner.
An instruction to start with:
Read the material and ask who the view is for if that is missing. Propose a view that answers our two questions. Specify which fields to show, which should stay in the working material and what needs checking during updates. Mark missing information. Provide a proposal for review before publication.
The test is complete when the recipient finds the right answers and the team knows how the next change will reach the view. If you need to connect this to existing systems, Tool Forge fits the next step: a working integration with clear ownership.
Companies and tools to watch
- Google Sheets: check whether a published view can replace recurring manual exports.
- Northern Trust: study how business rules and exceptions shape the integration.
- KT: follow how read access helps operators make faster decisions.
- GitHub Copilot: compare quality and cost on one task before changing the team's model selection.
The Forge newsletter
Get new articles in your inbox
Pick the topics you care about. No noise, at most one email a week.
We follow GDPR. Unsubscribe anytime.


