OpenAI and AWS extend AI systems into workplace memory and monitoring
AI systems are moving into retained desktop activity, multi-cloud agent monitoring and consulting workflows, making deployment controls as visible as model features.
Artificial Intelligence··Morning
Desktop activity becomes ChatGPT memory
OpenAI has added an opt-in feature called Computer History to its macOS desktop app. It turns activity across applications and websites into a searchable memory and timeline that ChatGPT and Codex can read, and it ships switched off by default, Unite.AI reports. The feature does not take screenshots; it collects events such as clicks, typing, keyboard shortcuts and app switches through macOS accessibility interfaces. An ephemeral Codex session condenses those events into text summaries stored as local Markdown files. The raw event stream is held locally, processed on OpenAI servers and deleted after 48 hours, while the summary files sit unencrypted on the local file system. Users can exclude specific apps and websites, pause collection and clear history by time period. The feature is rolling out to ChatGPT Pro, Business and Enterprise users, with the European Economic Area, Switzerland and the United Kingdom to follow in the coming weeks. The system can also propose an automation when it detects a repeatable workflow.[1]
IBM folds OpenAI tools into consulting work
IBM and OpenAI have agreed to jointly market enterprise AI offerings and build industry-specific solutions, TechCrunch reports. IBM Consulting is setting up a dedicated OpenAI practice and plans to train and certify tens of thousands of consultants—mostly existing employees—on OpenAI technologies. The work focuses on financial services, government, telecommunications and retail. GPT-5.6, Codex and ChatGPT Work will be folded into the IBM Consulting Advantage platform, and specialist teams called Forward Deployed Experts will be created through OpenAI’s partner network. The agreement expands a cybersecurity partnership the two companies formed in June 2026. Mike Healy, a managing partner at IBM Consulting, said the training effort is focused on the next several months. Financial terms were not disclosed, and IBM continues to run its own watsonx platform and Granite models. That package makes the consulting channel—training, certification and industry packages—as visible as the model names being installed.[2]
Agent telemetry reaches beyond a single cloud
Amazon Bedrock AgentCore Observability now collects telemetry from agents running outside AWS, the AWS Machine Learning Blog says. Sessions, traces, spans, token usage and latency from agents on on-premises servers, on Google Cloud and Azure, or on developer machines all flow into the same dashboard. Before this month, AgentCore Observability supported only agents deployed on the AgentCore runtime inside AWS. The setup described uses AWS Distro for OpenTelemetry version 0.10.0 or later, the Strands Agents package built with OpenTelemetry support, and Python 3.10 or later; authentication runs through SigV4, and the data is read in Amazon CloudWatch with Transaction Search and AWS X-Ray. The sample agent runs on the Claude Haiku model. Together with Computer History’s retained desktop memory and IBM’s OpenAI consulting practice, multi-cloud agent monitoring makes deployment controls—what is collected, where agents run, who is trained to operate them—as concrete as model features. Workplace AI is no longer only a chat window; it is a memory timeline, a certified consulting stack and a telemetry path that can follow an agent off a single cloud.[3], [1], [2]
