OpenAI launches workspace agents and free ChatGPT for clinicians; Meta tracks staff to train models as Google unveils enterprise agent platform and new TPUs

AI Tech News Today — Daily Briefing
Hosts: Aurora & Isabelle
Date: 2026-04-23T20:58:12.000Z

Quick roundup: a short summary of the biggest A‑I moves from today — let’s jump in.

1. OpenAI — Workspace agents let businesses run A‑I workflows without engineers
   What happened: OpenAI rolled out workspace agents inside ChatGPT so business and operations teams can build, share, and run agents across real workflows without waiting for engineering.
   Key details:

   • Powered by Codex.
   • Agents can live in ChatGPT and Slack.
   • Tasks include lead research to month‑end close.

   Pitch: Spot a repeatable process, spin up an agent, and stop wasting time in handoffs.

2. Meta — Training A‑I agents by tracking employees’ mouse and keyboard use
   What happened: Meta is installing tracking software on U.S. staff machines to capture mouse movements, clicks, keystrokes, and some screenshots. The program is called the Model Capability Initiative.
   Key details:

   • Monitors a designated set of work apps (e.g., Google, LinkedIn, Wikipedia).
   • Goal: teach models how people actually move through software.

   Implication: Raises clear questions about consent, privacy, and control over workforce data.

3. OpenAI — ChatGPT for Clinicians is free to verified U.S. providers
   What happened: OpenAI launched ChatGPT for Clinicians, free to verified physicians, nurse practitioners, and pharmacists in the United States.
   Key details:

   • Uses frontier models tuned for consultations, documentation, and literature review.
   • On the new HealthBench Professional benchmark, GPT‑5.4 scored 59.0 versus human doctors 43.7 — a striking gap that shows how fast clinical tools are evolving.

4. Google — Highlights from Cloud Next ’26: agents, new chips, security
   What happened: At Cloud Next ’26, Google announced several major items: a Gemini Enterprise Agent Platform, two new 8th‑generation TPUs, and agentic cybersecurity tools built with Wiz.
   Key details & metrics:

   • Gemini Enterprise Agent Platform: manage thousands of agents.
   • Two new 8th‑gen Tensor Processing Unit chips optimized for training and inference.
   • First‑party models now process >16 billion tokens per minute via customer APIs.
   • Inside Google, 75% of new code is now A‑I generated.
   • A complex code migration using agents ran 6× faster than with engineers alone.

Closing: Stay curious, stay skeptical — we’ll keep tracking the A‑I developments that matter.
This is Aurora and Isabelle, signing off from AI Tech News Today.