Field Notes

Live data in Quick Apps and faster agents: how you save hours in support, BI, and coding

Today is about launches that speed up work: live data right in the AI-built app front end, noticeably faster agents, and a proven path to better conversion.

Key AI news

Amazon Quick: live data in AI-built apps with governance

Quick Apps can now run published apps live against your QuickSight datasets on every open. No more working from the build snapshot. Queries run in the viewer’s context. Existing row and column security (RLS or CLS) apply without a new permission model. SPICE and Direct Query are supported. You need curated datasets in place, Reader Pro roles at minimum, and a consent flow per dataset. The feature has launched.1

OpenAI: Astra Ultrafast is available

GPT-6 Astra Ultrafast is available in the OpenAI API and for eligible ChatGPT Work and Codex users. The provider says Ultrafast delivers tokens up to eight times faster than Astra’s standard mode. That helps in loops with tool calls and code edits. Less waiting time per step. The figures come from the provider. Access and pricing are in the Ultrafast guide.2

AWS case study: personalization with contextual bandits

Amazon Payments deployed a contextual bandit solution (LinUCB) across the full conversion funnel. In a seven-week online test, final conversion rose by a high single-digit percent relative in one target group. Another group showed no effect. The selection rule was deterministic and explainable. The bottleneck was partly content quality, not the model. A contextual bandit replaces rigid A or B tests. It is a method that picks variants by visitor signals and keeps learning.3

Tailored moderation at uniopen with Amazon Nova

uniopen adapted Amazon Nova 2 Lite via supervised fine-tuning in SageMaker and downstream prompt optimization to a two-axis, company-specific moderation policy. Correction data is human-verified. Mandatory tests and alert thresholds govern promotions to production. Evaluation used 737 conversation windows. Training used 3,391. Result: a repeatable, controlled improvement loop without treating generated labels as truth. Model availability depends on the region.4

Highlights for your workday

  • You maintain weekly numbers in chats or slides. Build the query once as a Quick App over your curated datasets. The agent writes the SQL. The app shows fresh values on every open, and each person sees only allowed rows. You eliminate snapshots and manual exports.1

  • Your coding or tool agents often wait on model responses. Switch those to Astra Ultrafast where latency dominates. Compare response time and cost per task against the standard mode. The speed figure comes from the provider.2

  • You personalize funnel content. Start with a contextual bandit instead of the next A or B test. Define 3 to 5 variants. Pick goals per funnel stage. Begin with UCB or LinUCB and watch for groups with no lift. Plan a content pipeline, not just the model.3

Tools and updates

  • QuickSight hierarchy filter: a compact control for multi-level drilldowns, for example region: country: city, up to five levels. Reduces filter sprawl. Works in combination too, for example whole country plus a single city. Author access required.5
  • n8n template: automated review monitoring. Checks listings on a schedule. Summarizes sentiment and statements. Works for competitors too. Useful for reputation management.6
  • n8n Jev model: makes decisions instead of writing. Returns choice plus confidence. Ideal for branching and routing in workflows. Remember to log thresholds.7
  • Claude Platform on AWS, multi-environment access: guide for a central AI services account, cross-account roles for AWS workloads, workspace API keys for devs, and OIDC for external pipelines. Requires an AWS organization and a CPonAWS subscription.8

Try it in five minutes

Requirement: you have QuickSight author access and a dataset with region, country, and city.

1) Open an existing analysis. Under Filters, add a new filter, choose Region, set filter type to Hierarchy filter, then add the fields Region, Country, City in that order. Set scope to Cross-sheet.5 2) Add the filter to the sheet and remove old single filters for region or country or city. Publish the dashboard.5 3) Test the control. Select a whole country and also a single city from another country. Expectation: all visuals and KPIs respond consistently to the combined selection. Check: the sum of filtered revenue matches the map shading.5

Sources

  1. Serve live, governed data in AI-built apps with Amazon Quick (aws.amazon.com)
  2. How NVIDIA GPUs Help Accelerate OpenAI’s GPT-6 Astra Ultrafast (blogs.nvidia.com)
  3. Uplifting conversion across the acquisition funnel with personalization using contextual bandits on AWS (aws.amazon.com)
  4. How uniopen customized Amazon Nova to their retail moderation policies for production deployment (aws.amazon.com)
  5. Simplify dashboard drill-down with the Amazon Quick Sight hierarchy filter (aws.amazon.com)
  6. @n8n_io: New featured template: automated review monitoring. It checks your listings on a schedule and reports the sentiment with a summary of what customers said. Competitor listings work (x.com)
  7. @n8n_io: @typesafeai's Jev model just landed in n8n. Jev’s job isn’t to write, it decides: returning a choice + confidence you can use for branching, sorting and routing in your workflows. (x.com)
  8. Implementing Multi-Environment Access for Claude Platform on AWS (aws.amazon.com)