Jaime Vengoechea
AI-powered. Shaped by Jaime Vengoechea.

The Applied Future.

AI, work, technology, and what comes next.

Sep 23, 2026Source-linked briefingPrevious editions RSS feed
In this edition

New model releases put efficiency in focus, while Google, AWS and NVIDIA address how AI-powered work is guided, evaluated and deployed. These five selections cover September 22 and 23 announcements. Vendor performance claims remain attributed, and earlier stories retain their original dates in the rolling reading list.

This Edition

Sep 23, 2026 · 5 stories
OpenAI launch artwork: GPT-6 Sol and Luna against a sun and moon scene
Image: OpenAI. Image source & credit.
Today's lead
AI developmentsSource: OpenAI
Model release; vendor evaluations

OpenAI brings GPT-6 to lower-cost everyday workflows

OpenAI released GPT-6 Sol and Luna for the API, Codex and ChatGPT Work, positioning them as faster, more affordable alternatives to Astra. The launch combines lower token prices with caching improvements and vendor-reported gains in coding, factuality and computer use. Availability is not the same across every ChatGPT surface, and benchmark results are not guarantees for a team's own tasks.

Read original source
Context & source check

Re-test a small set of recurring workflows before changing your default model. Compare accepted outputs, correction time and total cost per completed task, not just token prices or a leaderboard position.

Model release; vendor evaluations. Company source. Source checked Sep 23, 2026. Not independently tested.

The reading list

4 stories

More from This Edition

4 stories
Anthropic promotional launch artwork for Claude Opus 5.5
Image: Anthropic. Image source & credit.
Latest edition
AI developmentsSource: Anthropic
Model release; vendor-reported savings

Claude Opus 5.5 puts cost per completed task in the spotlight

Anthropic introduced Opus 5.5 with lower token and cache-read prices, clearer communication and claimed improvements on long-running coding and knowledge work. The company reports roughly 40% lower typical workload costs than Opus 5 at default settings. That is a vendor test result, not a universal saving. AWS separately confirmed availability through Bedrock and Claude Platform on AWS.

Read original source
Context & source check

Savings depend on how often context is reused, how much the model reasons and how many retries a task needs. Include review time and failure recovery in comparisons, while retaining approval boundaries for consequential actions.

Model release; vendor-reported savings. Company source. Source checked Sep 23, 2026. Not independently tested.

Image: Source: Google. Google Ads is a trademark of Google LLC; no affiliation or endorsement. Image source & credit.
Latest edition
MarketingSource: Google
Closed-beta expansion; reporting preview

Google expands AI Brief languages and previews a clearer ad journey report

Google is expanding AI Brief's closed beta to seven more languages, including Spanish, French and Portuguese. Advertisers can supply business, audience and messaging context in natural language. Google also describes a unified report connecting search terms, the creative shown and the landing page. Further reporting availability details are promised later this year; this is not a claim of general access today.

Read original source
Context & source check

Prepare a clear brief and a review process for each language. When the report becomes available, inspect whether search intent, ad copy and landing pages tell a consistent story before judging automation by conversion volume alone.

Closed-beta expansion; reporting preview. Company source. Source checked Sep 23, 2026. Not independently tested.

Video: NVIDIA / Ekumen / YouTube . Loads YouTube when played.
Latest edition
TechnologySource: NVIDIA
Open-source robotics release

Isaac ROS 5.0 brings reusable AI skills into robotics development

NVIDIA released Isaac ROS 5.0 at ROSCon in Toronto, adding agent-ready documentation and reusable development skills alongside ROS Lyrical and Ubuntu 24.04 support. The release spans perception, manipulation and deployment across Jetson hardware. The linked Ekumen demonstration illustrates collision-free motion planning in the ecosystem; it is not an independent test of the entire release.

Read original source
Context & source check

Reusable skills are moving beyond screen-based work. Physical automation still needs reproducible environments, simulation and hardware validation: faster code generation does not remove the need to test safe real-world behavior.

Open-source robotics release. Company source. Source checked Sep 23, 2026. Not independently tested.

AWS article preview: Evaluate skill-equipped agents with Strands Evals and Amazon Bedrock AgentCore
Image: Amazon Web Services. Image source & credit.
Latest edition
Work & wealthSource: Amazon Web Services
Technical implementation guide

AWS shows how to test whether agents follow the right playbook

AWS explains how Strands Evals and Bedrock AgentCore can inspect whether an agent chose an appropriate skill and followed its instructions. Strands also offers a deterministic check that a named skill was loaded. The guide uses recorded steps and traces, not just the final response, to distinguish choosing the wrong procedure from skipping steps in the right one.

Read original source
Context & source check

A polished answer can hide a broken workflow. Add tests for the required procedure, required checks and actions that must not occur, then investigate routing mistakes separately from incomplete execution.

Technical implementation guide. Company source. Source checked Sep 23, 2026. Not independently tested.

Story preview