Skip to content
MustHave.ai
MustHave Hoodie Seal
THE MUSTHAVE NEWSROOM

AI news

What’s new. What matters. What do you think?

Top stories

By member votes

GitHub AI Scan Is Generally Available: Verify Coverage and Advisory Limits

GitHub AI Scan is generally available. Check effective policy, fork and Dependabot exclusions, advisory-only findings and Code Security AI Credit budgets.

Build One MCP App for ChatGPT and Claude on AWS AgentCore

AWS AgentCore shows how one MCP App can work in ChatGPT and Claude. The portable pattern is useful, but the no-auth demo needs production hardening.

AWS Says GPT-5.6 Luna Beats GPT-5.4 Mini on Cost per Correct Answer

AWS compares OpenAI models by cost per correct answer instead of token price. The result is useful, but model settings are not fully matched.

Google ToolGrad Builds Tool-Use Training Data Backwards and Reports a 99.8% Pass Rate

Google ToolGrad builds tool-use training data from a valid answer chain back to a user request. Researchers report a 99.8% generation pass rate.

Adobe Acrobat Productivity Agent Turns PDFs Into Podcasts, Reports and Slides

Adobe Acrobat Productivity Agent can turn PDFs into podcasts, reports and slides. Here is the plan map and a fidelity test for real document work.

ElevenLabs and UMG Plan a Licensed AI Remix Platform for Fans

ElevenLabs UMG AI remix plans cover a separate fan platform for participating music. The agreement is real, but launch timing, artists and rights remain open.

Salesforce AI Control Plane Arrives With Job-Ready Agentforce Agents

Salesforce AI Control Plane promises shared governance for Agentforce and third-party agents, but the unified harness rollout starts in fiscal FY28.

Cognition SWE-2 Reaches Devin With a 64% Cost Claim and One Benchmark Warning

Cognition SWE-2 is reaching Devin with a company-reported 64% cost advantage in one comparison. The wider benchmark table tells a less simple story.

Cursor Projects Launches With Persistent Context, Agent Swarms and Scheduled Work

Cursor Projects adds persistent context, coordinated agent swarms and scheduled work in beta. Here is what builders should test before trusting it.