The one thing
Salesforce's AI assistant grows 21 times in three months, now inside Teams
Why it mattersIf your staff live in Microsoft Teams, they can hand tasks to Salesforce's AI without opening the , so adoption friction drops.
Seventy: Thursday, 5 November 2026
What happened. Salesforce says usage of its Agentforce Coworker AI assistant grew 21 times in three months, reaching over 185,000 users at firms including The Adecco Group, Fulton Bank and BACA Systems. It is also extending Coworker into Microsoft Teams, mobile and desktop with new abilities.
More from today
02Funding
OpenAI buys a company to improve its AI coding helper, Codex
Why it mattersTeams using OpenAI's Codex get more built-in help with code review and comprehension, so evaluate it against your current coding tools.
What happened. OpenAI is buying Pierre Computer Company to strengthen Codex, its coding product, with tools for writing, explaining and checking software.
03Models
OpenAI's image maker is faster and can now create animated GIFs
Why it mattersTeams that produce marketing visuals can now make animated GIFs and make tighter edits in OpenAI's tools, with fewer outside design steps.
What happened. OpenAI released Images 2.6, which generates pictures more quickly, can produce animated GIFs, and edits existing images with finer control. It is available in ChatGPT and Codex, and developers get extra settings through the API.
04Agents
AWS shows how to stop AI assistants failing midway through long reporting tasks
Why it mattersTeams building AI assistants for reports and data analysis can cut mid-task failures and runaway costs by managing what the agent remembers.
What happened. AWS published a walkthrough of Deep Insight, a working example showing how to keep AI agents from running out of memory during data analysis and report writing. It combines several context-management techniques, and a 200K-token window can be used up within minutes on real tasks.
05Agents
AWS adds steadier pass-or-fail scoring for checking your AI assistants' work
Why it mattersTeams testing AI assistants can score policy and safety checks with a consistent number, so approval thresholds stop shifting between runs.
What happened. AWS now lets Amazon Bedrock AgentCore Evaluations use decision models, which pick from a fixed set of answers and give a calibrated probability for each. They suit narrow checks such as policy compliance, tool-call safety and whether an answer matches retrieved context.
Nothing today qualified as infrastructure and compute. We left the slot empty rather than fill it with something weaker.
That's the whole issue.
5 stories, about 1 min 54 sec. Nothing was left out to fill the time.
Words of the day
- context-management techniques
- Methods for controlling what an AI agent keeps in its working memory, such as summarising, trimming or offloading information, so it doesn't overflow.
- decision models
- Models that choose from a fixed set of answers and output a calibrated probability for each, rather than writing free-form text.
- synthesis routes
- Step-by-step chemical reaction plans for making a given molecule in the lab from available starting materials.
“We are stubborn on vision. We are flexible on details.”