News archive
Weekday briefing September 11, 2026 · 8:00 AM ET

AI agents get longer tasks and tighter controls

OpenAI's new Agents API can run longer jobs, while new data and voice tools bring AI into reporting and customer calls. Anthropic is adding server-side approval checks for every tool action, and Gemini now has a Windows app. These tools are getting more useful, but beta labels, permissions, and full-stack costs still deserve a close look.

AI-assisted and human-reviewed Last updated September 11, 2026
01

The Big Story

OpenAI News

Introducing the Agents API

OpenAI has put its Agents API into public beta for all developers, offering the managed Codex infrastructure behind long-running agents that can use tools, run code, handle files, and coordinate subagents. Developers can choose OpenAI’s hosted sandbox, their own infrastructure, or partners including Cloudflare, DigitalOcean, and Vercel; the options differ in deployment, storage, performance, and cost. There’s no separate API fee, but customers pay for the tokens and tools they use.

Takeaway

This is useful if a developer is building a custom AI workflow, but most small businesses should wait for a packaged application and clear production guidance before putting customer work on a beta API.

Read the original at OpenAI News
02

For Everyone

Google Gemini Blog

The Gemini app is now available for Windows

Google has launched its Gemini desktop app globally for Windows 10 and 11. Pressing Alt + Space opens Gemini over the active app for tasks such as checking a document or brainstorming presentation titles, while its dedicated workspace can pull information from Gmail and Google Drive for project summaries. The app also offers Gemini Spark for multi-step tasks and image and video generation, with more Windows-specific features planned.

Takeaway

Windows teams using Google Workspace can test this as a quicker way to use Gemini, but should review Gmail and Drive permissions before connecting business information.

Read the original at Google Gemini Blog

OpenAI News

How a researcher uses Codex and ChatGPT to search for new antimicrobial molecules

César de la Fuente’s bioengineering lab uses its own deep-learning models to search genome and protein databases for possible antimicrobial molecules, cutting the initial search from years to hours. The team also uses ChatGPT and Codex to brainstorm hypotheses, write and refine code, prepare datasets, analyze results, and help researchers work across biology, chemistry, and computing. The tools can prioritize leads, but laboratory experiments, safety testing, clinical trials, and regulatory review are still required before a candidate could become a medicine.

Takeaway

This isn’t a ready-made business tool, but it’s a useful reminder that AI is strongest at narrowing large research or data tasks, with experts checking every result before decisions are made.

Read the original at OpenAI News
03

For Builders

OpenAI News

Build more natural voice experiences with GPT‑Live‑1 in the API

OpenAI released GPT-Live-1 in its API for voice apps that listen and speak at the same time, handle interruptions and background noise, and can run full-duplex phone agents for reservations or support. The voice layer can pass complex reasoning and tool calls to GPT-6 Astra or third-party models, while providing transcripts, response text, and turn detection. It costs $0.05 per minute for the voice layer, with backend model and agent charges separate; custom voices require contacting sales.

Takeaway

A business with high-volume calls can test a narrowly scoped booking or support agent, but should price the full stack and keep human escalation for requests that need judgment.

Read the original at OpenAI News

AI TLDR

Claude Managed Agents add auto mode: the server checks every tool call

Anthropic added an `auto` permission policy to Claude Managed Agents on September 10, 2026. For every agent or MCP tool call, Anthropic's server decides whether to run it, deny it, or pause for human approval, and records that decision in tool-use events. Its ant CLI can also connect to a live agent session so an operator can watch it, send messages, and approve or deny waiting calls from a terminal or local web viewer.

Takeaway

If you're building a Claude-powered agent that can touch business systems, test whether these approval controls cover the actions you'd otherwise need to supervise yourself.

Read the original at AI TLDR
04

For Business

OpenAI News

Now everyone can put data to work

OpenAI introduced Data, a ChatGPT Work agent that lets employees ask business questions in plain language, investigate changes, and build shareable interactive dashboards from approved company data. It connects to sources including BigQuery, Snowflake, Databricks, Redshift, Google Drive, and SharePoint, while retaining the connected account’s existing table, row, and column permissions. Administrators install it through Workspace settings, choose data-source plugins, and limit access by role; users can review supporting evidence and share results by Slack or email.

Takeaway

If your business uses ChatGPT Work, start with a tightly controlled reporting use case and make sure your metrics and data permissions are in order before giving more staff access.

Read the original at OpenAI News
05

Worth Watching

TechCrunch AI

OpenAI’s feud with mathematicians is only escalating

Twenty-five Fields Medal-winning mathematicians warned that AI labs are rushing to announce proofs without adequate verification, explanation, or credit for prior work. The letter followed allegations that OpenAI pressured an NYU professor over collaborator attribution and concerns that researchers’ Codex use could inform future models. OpenAI also withdrew sponsorship of a Caltech math event after criticism, while its disputed proof remains unverified.

Takeaway

Treat confidential or valuable work entered into AI tools as potentially exposed, and establish clear review, attribution, and data-use policies before employees use them.

Read the original at TechCrunch AI

TechCrunch AI

OpenAI puts Pro subscriptions on hold due to Astra demand

OpenAI paused new subscriptions to its $200-per-month Pro plan after demand for the Astra model strained infrastructure. Existing users retain priority, while the API and lower-cost Go and Plus plans remain available. OpenAI has not provided a restart date; Astra launched September 3 across Pro, Plus, Enterprise, and Business tiers.

Takeaway

Businesses planning Astra-dependent work should avoid assuming immediate Pro availability and test whether Plus, Business, or API access meets their needs while capacity remains constrained.

Read the original at TechCrunch AI