AI Enablement Radar week 45: turn feedback into an action

AI Enablement Radar week 45: turn feedback into an action

This week's most useful AI updates help teams act on what they already know. Google Forms gains analysis with suggested next steps, Cursor can build and publish internal apps, and the EU gathers examples public organizations can reuse. For Nordic teams, this is a good moment to choose a recurring question and build a routine that leads to a decision. Here are the signals from November 2–8 and a concrete test for next week.

Top signals this week

  • Google Forms gains Gemini analysis that runs Python for calculations, finds patterns across questions and suggests next steps. Rollout varies by account; it is not available to everyone at once. Source: Google Workspace: actionable insights in Forms.
  • Cursor launched Build Live Apps on November 5. Cloud agents can arrange a database, login and hosting, but publishing requires an explicit user decision. Source: Cursor: Build Live Apps.
  • GitHub Dependabot can now update agent dependencies in Agent Package Manager. These are version updates, not security updates. Source: GitHub: Dependabot and Agent Package Manager.
  • The EU's AI toolbox brings together reusable solutions, training and procurement support. This week's introduction includes form assistance and a text assistant for public administration. Source: Interoperable Europe: AI toolbox.

What organizations are actually doing with AI

Novo Nordisk lets domain experts try ideas

As background, not news from this week, Anthropic describes how Novo Nordisk uses Claude Code to let scientific experts create prototypes. They can show an idea to the development team instead of first translating it into technical tickets. The case study describes a development team of eleven people.

The same source shows why prototyping and integration need to be distinguished: connecting legacy systems for device protocols took several months. This is a useful division of work for a Nordic team. The person who knows the business makes the need concrete; the person responsible for the system assesses how to connect it. These are supplier case-study claims, not an independent evaluation.

Source: Anthropic: Novo Nordisk.

Baden-Württemberg gives text work a shared service

This week's EU introduction features F13, a generative text assistant from Baden-Württemberg's public administration innovation lab. It supports summaries, research and text generation. The useful point is a shared service for recurring work, rather than every employee building a separate solution. The introduction gives no verified productivity gains.

Source: Interoperable Europe: AI toolbox.

The tooling layer: platforms, agents, and workflows

From survey responses to an action

Gemini in Google Forms can analyze responses and recommend next steps directly in the Responses tab. Business Standard and Plus are among eligible editions. Rapid Release rollout began November 2; Scheduled Release starts November 19. Check that the feature appears in your account before planning work around it.

For example, use course feedback to distinguish a recurring problem from an isolated request. Check counts and denominators yourself before changing the course structure.

Source: Google Workspace: actionable insights in Forms.

A shorter route to hosting an internal app

Cursor can build apps from new or existing projects, create Postgres when needed and arrange login through Cursor SSO. The feature is available on paid plans with cloud agents enabled. This makes an internal view of cases or feedback more realistic, but the team must still decide who owns the data and approves publication.

Source: Cursor: Build Live Apps.

On November 6, Cursor also added two-way voice calls with agents. This differs from dictation: conversation can continue while the agent delegates coding work. Treat it as another way to discuss a change, not a reason to skip review.

Source: Cursor: Voice Chat with Agents.

Maintain the instructions too

Agent Package Manager manages items including agent instructions and skills as dependencies. Dependabot can now open update proposals for supported git dependencies in apm.yml and refresh the lockfile when present. Routines the team reuses need version maintenance too.

Source: GitHub: Dependabot and Agent Package Manager.

GitHub's new review metrics need careful interpretation. A comment can count as addressed even when the reply says a fix will arrive in a later change. Use the metric to follow up work, not as proof that defects are resolved.

Source: GitHub: code review addressed rate.

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

Google announced on November 2 that Gemini 2.5 Pro and Flash are being deprecated. From December 1, API requests to them will route to other models at the destination model's prices. Parameters are changing too. Anyone operating an integration needs to test real requests before that date, rather than simply change a model name.

Source: Google: Gemini API changelog.

The EU toolbox, introduced on Interoperable Europe on November 4, supports public organizations with technology selection, skills and procurement. Its F13 example is a text assistant for tasks including summaries and research. Use the material to inform requirements and comparisons, not as approval of your own solution.

Source: Interoperable Europe: AI toolbox.

Google Meet is also starting to roll out administrator-controlled policy notices before meetings. They can appear when recording, transcription or Gemini notes are used. A visible policy helps participants understand the routine, but does not replace an assessment of legal basis or consent where required.

Source: Google Workspace: Meet policy notices.

As Swedish background guidance, Digg recommends connecting AI governance to existing management documents and clear responsibilities. In practice: give the integration scoped permissions, store keys in environment variables or a secret manager, and log runs without exposing secrets. Put human approval where a workflow changes customer commitments or publishes something.

Source: Digg: Inför en AI-policy.

This week's practical Hammer test

Set aside 40 minutes to turn feedback into a decision brief. Use a current survey or an export the team is permitted to analyze in its approved AI tool. If the Forms feature is missing, do the same exercise with a file in the tool you already use.

  1. In the first 5 minutes, choose a decision, such as which part of an introductory course to improve. Name the decision owner.
  2. Over the next 10 minutes, ask AI to group responses, show counts and identify the responses supporting each conclusion. Separate observations from proposals.
  3. Over the next 15 minutes, check two recurring patterns and one outlier against the original. Correct misclassified responses and recalculate if needed.
  4. In the final 10 minutes, choose an action, an owner and a follow-up date. Save which material was used. Do not send or publish anything automatically during the test.

A short instruction to use:

Read the material and ask which decision we need to make if that is unclear. Group recurring feedback, give counts and identify the evidence for each group. Show conflicting responses separately. Suggest an action, but state what we must check before choosing it.

The test is complete when the decision owner can trace the proposal back to the responses and say what should change. Next, connect the same routine to recurring inputs. Tool Forge fits teams ready to connect that routine to their existing systems.

Companies and tools to watch

  • Google Forms: test whether the analysis actually helps you choose an action.
  • Cursor: follow how hosting and ownership work after the first publication.
  • Novo Nordisk: let domain experts show the need with a prototype.
  • GitHub: keep reusable agent instructions updated like other dependencies.

The Forge newsletter

Get new articles in your inbox

Pick the topics you care about. No noise, at most one email a week.

Get new articles in your inbox

We follow GDPR. Unsubscribe anytime.