Nvidia has said it does not plan to reappoint Jensen Huang as CEO, even though he offered to return to the top role after resigning last week. The report centers on leadership at the AI…
Why it matters
Enterprises that depend on Nvidia's AI hardware and software should note possible shifts in strategy and roadmap, but no immediate action is required.
External sandboxes for coding agents I've been exploring new ways to sandbox code that is executed by agents, with a focus on making it safer to try out local models without worrying that a…
Why it matters
Developers running coding agents can isolate agent-executed code in an external sandbox with network allow-listing and credential proxying, which reduces the damage from prompt injection or a bad command when trying local models.
Amazon SageMaker AI real-time endpoints now support KV cache-aware routing, which sends each request to the instance holding the longest cached match for its prompt. The AWS post explains…
Why it matters
Teams serving LLMs on SageMaker can reduce redundant prompt processing and latency by enabling KV cache-aware routing for workloads with shared prompt prefixes, after comparing it against prefix-aware routing.
The AWS AI/ML Blog offers a guide on reducing generative AI system costs by first deciding whether each workflow step needs a model at all. It proposes three options per step: deterministic…
Why it matters
Implementation teams can cut LLM spend by routing rule-based steps to deterministic code and reserving right-sized Bedrock models for steps that need judgment, which makes architecture decisions a cost-control lever.
AWS tested three OAuth patterns for AI agents running on Amazon Bedrock AgentCore that call Snowflake Cortex AI over the Model Context Protocol (MCP). The patterns are Snowflake OAuth…
Why it matters
Organizations connecting AI agents to enterprise data platforms should pick an OAuth pattern that matches their identity provider, audit trail, and compliance obligations before deploying agent-to-data integrations.
AWS published a walkthrough of custom metadata schemas in AWS Agent Registry, which let an administrator define typed, validated fields that every agent, MCP server, and skill must carry…
Why it matters
Enterprises running many AI agents and MCP servers can use enforced metadata schemas and approval workflows in AWS Agent Registry to standardize governance and discovery of AI resources.
Urgency: ImportantCategory: Model ReleaseGitHub Changelog·
OpenAIGitHub Copilot
OpenAI's GPT-6.2 Astra and GPT-6.2 Sol are now generally available in GitHub Copilot. GitHub says both models are designed for agentic coding and code review, with stronger coding judgment…
Why it matters
Developers and teams using Copilot can now select these newer OpenAI models for agentic coding and code review, so they should evaluate them against current defaults and check their model-selection policies.
Gemini's Task mode can help users coordinate calls, log into websites, and even make purchases on the user's behalf.
Why it matters
Enterprises should note that consumer agentic features that place calls and make purchases on users' behalf raise new questions about delegated authority, consent and acceptable-use policies for staff who use Gemini.
Model ML, an AI startup, raised $250M to build agents that automate investment banking work. The company says its agents understand the full context of deals bankers and consultants are…
Why it matters
Banks and consultancies evaluating AI for deal work now have another well-funded vendor targeting financial modeling and presentation tasks, so they should assess its accuracy, data access and governance controls before adopting it.
GitHub has extended Copilot's security review specialist to run GitHub secret scanning on code changes. Developers and AI agents can now check for exposed secrets before changes are merged…
Why it matters
Teams using Copilot code review or AI coding agents can add pre-merge secret detection without a separate workflow, though they should still confirm that scanning covers their repositories and agent-generated commits.
GitHub has moved Copilot Evals into public preview, a feature that lets users evaluate and tune how GitHub Copilot performs as their prompts, models, skills, and tools change. The tool is…
Why it matters
Teams relying on Copilot can now measure and adjust its performance over time, so they should consider piloting evals to validate changes in prompts, models, or tooling before rolling them out.
GitHub has added a Repository insights page, the first item in each repository's Insights tab, that shows how Copilot code review feedback is received in that repository. The page lets teams…
Why it matters
Engineering leaders evaluating Copilot code review can now measure whether its feedback is actually acted on, which gives them concrete data for deciding how far to rely on it.
GitHub Copilot CLI now supports Microsoft Entra authentication in public preview. Users can sign in with their Microsoft Entra account to silently authenticate to a linked GitHub Enterprise…
Why it matters
Enterprise IT and identity teams governing Copilot CLI access can evaluate routing its sign-in through Microsoft Entra, though the feature is still in preview and should be tested before broad rollout.
Enterprises can now stream GitHub Copilot agent activity directly into Microsoft Agent 365, Microsoft's agent control plane. This brings audit data from across Copilot clients into one place…
Why it matters
Enterprise security and governance teams can now centralize audit trails for Copilot agents alongside their other AI agents, which supports compliance and oversight of agentic coding tools.
GitHub has released an update to GitHub Copilot for JetBrains IDEs that adds remote access, on-device voice dictation, and browser use controls. Copilot users in JetBrains get context usage…
Why it matters
Engineering teams using Copilot in JetBrains IDEs should review the new browser use and remote access controls to set policy for how the agent-assisted features are enabled across their developers.
Urgency: ImportantCategory: Model ReleaseGitHub Changelog·
GitHub Copilot
GitHub Copilot now offers Microsoft AI's next-generation coding models, MAI Code 2 and MAI Code 2 Flash. The announcement describes them as a major step forward for Microsoft's MAI coding…
Why it matters
Teams using Copilot should evaluate whether the new MAI Code 2 models fit their coding workflows and cost profile before defaulting to existing models.
GitHub has added new capabilities to Copilot code review, including adaptive effort that scales review depth to each pull request so routine changes get a quick pass and complex ones get a…
Why it matters
Engineering teams using Copilot for pull request review should expect faster reviews on routine changes and an extra layer of credential leak detection, so they may want to revisit their review workflows and secret-scanning policies.
AWS announced its Forward Deployed Engineering service, in which AWS engineers embed with customer teams to move high-value AI workflows from proof of concept to production in 45-day…
Why it matters
Enterprises stuck between AI pilots and production can consider a time-boxed, vendor-embedded engagement as an alternative to building that capability in-house, but should confirm what handover, governance, and ownership terms the sprint actually delivers.
GitHub expanded Copilot Fusion with new underlying models and made it available on more Copilot surfaces. GitHub also published Fusion's results on two public Terminal-Bench leaderboards…
Why it matters
Teams using Copilot can now access Fusion on more surfaces and should check the Terminal-Bench results when comparing coding agents for terminal-based work.
Amazon SageMaker AI now offers data preparation capabilities for building training and evaluation datasets for model development. The tools generate synthetic examples, assess and improve…
Why it matters
Teams building or fine-tuning models on AWS can streamline dataset creation, quality checks and labeling inside their existing workflows, which may reduce reliance on separate labeling vendors and manual curation.
The company uses agentic AI to make supply chains faster and more reliable.
Why it matters
Enterprises running supply chain operations may see more agentic AI vendors like Pedestal AI entering the market, so procurement teams should track whether these tools deliver measurable reliability gains before committing budget.
OpenAI launched Astra for Law, a legal-focused AI offering that ranked first in a launch evaluation covering 24 areas of law. Every ChatGPT Enterprise customer can now request access.
Why it matters
Enterprise legal and compliance teams on ChatGPT Enterprise can request access to a domain-specific legal AI tool, so they should evaluate it against their existing legal research workflows and governance requirements before adopting it.
Sqwish Labs is offering a decision model that customers can train on their own data to improve its performance. The article frames this customization approach as a possible next step as use…
Why it matters
Enterprises evaluating decision models should consider whether vendor-offered customization on proprietary data is worth the added data governance and integration effort compared with off-the-shelf models.
OpenAI has introduced GPT-6.2 Astra and Sol, two new model versions. OpenAI says they improve professional outputs, coding, and computer use, while lowering cost and improving efficiency…
Why it matters
Developers and enterprise teams using OpenAI's API or Codex should evaluate Astra and Sol against their current models for coding and computer-use workloads, and check whether the lower cost changes their model selection and budgets.
Google's Gemini update extending chat retention will roll out globally over the coming weeks.
Why it matters
Organisations whose staff use Gemini consumer chats need to reassess data-governance and retention policies, since chat logs now persist for two months even when history is turned off.