AI News - The AI Enabled PM

Editorial illustration of an independent evaluator inspecting an AI system inside a lab.

AI News

Anthropic Puts Independent Evaluators Inside the Lab

Anthropic and Accenture plan to embed independent evaluators inside frontier-model development. Access, funding and release authority will determine whether it works.

Editorial illustration of connected apps passing through a permission gate.

AI News

Gemini's Connected Apps Expand the Permission Surface

Gemini now connects to work, creative and lifestyle apps. The product challenge is making every delegated action understandable and recoverable.

Editorial illustration of a prompt cache meter and reusable context blocks.

AI News

OpenAI Makes Cache Hit Rate a Product Metric for Agents

OpenAI added monitoring and controls for GPT-6 prompt caching. Agent teams now need to manage cache hit rate as a product metric.

CRM workflow knowledge is organized into an ordered specialist model process

AI News

Salesforce Koa Turns Workflow Knowledge Into a Model

Salesforce built Koa by training an open model on synthetic CRM workflows. The moat may be the workflow specification, not the base model.

An agent execution trace highlights an anomalous action before a stop gate

AI News

Google Wants to Catch Agents That “Succeed” the Wrong Way

Google’s new anomaly detector audits how enterprise agents reason and use tools. Task completion alone is no longer enough.

A sponsored product card becomes a conversation and hands the user to the advertiser website

AI News

OpenAI Turns ChatGPT Ads Into Conversations

OpenAI is testing Sponsored Agents inside ChatGPT ads. PMs should evaluate the full question-to-handoff journey, not just click-through rate.

An open model library inside a compute-chip frame with multiple cloud and hardware connections.

AI News

NVIDIA's Hugging Face Deal Puts Platform Neutrality on Trial

NVIDIA agreed to acquire Hugging Face for $12.93 billion. Teams should now test whether the platform remains neutral in practice.

A batch of code-quality findings organized into one agent-created pull request for review.

AI News

GitHub Turns Code-Quality Backlogs Into Agent Work Queues

GitHub can now hand up to 25 code-quality findings to Copilot. The useful metric is accepted fixes, not findings closed.

A precise image-editing canvas showing local edits preserved across several visual revisions.

AI News

ChatGPT Images 2.5 Turns Generation Into an Editing Workflow

Images 2.5 improves targeted edits, reference fidelity and iteration speed. The product metric should be time to an approved asset.

An evaluation flask connected through an emergency stop to a live server stack beside the headline “Agent evals carry production risk.”

AI News

OpenAI’s Hugging Face Incident Turns Agent Evals Into Production Risk

The OpenAI–Hugging Face incident shows why agent evaluations need production-grade access controls, monitoring, stop paths, and recovery.

A completed-task card above three token discs beside the headline “Cost per completed task.”

AI News

Gemini 3.6 Flash Makes Cost per Completed Task the Model Metric

Gemini 3.6 Flash may lower token spend. PMs should compare cost per accepted task across model calls, tools, review, and recovery.

Voice and chat sockets connected to a protected agent service module beside the headline “From demos to operations.”

AI News

OpenAI Presence Moves Agents From Demos Into Operations

OpenAI Presence packages the operating layer around enterprise agents. PMs still need clear authority, handoffs, ownership, and accepted-outcome metrics.