Wednesday, August 26, 2026

Good morning, NOLA. Today is about putting agents to work without getting buried in the machinery: OpenAI has restored longer Codex and Work sessions for Plus users, while a new report says agents are becoming everyday work tools inside OpenAI. There is also a promising voice-AI product angle from Ringg, a useful reality check on what AI-written software can become, and a robot-business signal worth watching.

Agents That Fit Into the Workday

Longer Codex and Work sessions return for ChatGPT Plus

OpenAI has restored five-hour limits for Codex and Work on ChatGPT Plus, a meaningful quality-of-life change for anyone using an agent through a long debugging session or project sprint. Yesterday we covered cheaper GPT-5.6 Sol access; this is the complementary change: more room to actually use the tools.
Hacker News

What widespread agent use looks like inside OpenAI

AI Daily Brief highlights OpenAI’s claim that 97.9% of its employees now use agents. The useful takeaway is not the percentage; it is the emerging habit: give agents bounded, repeatable work and make them part of the normal workflow rather than a special-event demo.
AI Daily Brief

A million lines of AI-written code still needed refinement

Simon Willison shares Paul Dix’s account of AI producing a large codebase that was then refined over months into reliable software used by millions. It is a good antidote to both hype and hesitation: agents can create serious leverage, but the human job shifts toward testing, review, and steady improvement.
Simon Willison

Voice AI wants to continue after the call ends

Ringg is pitching voice AI that can carry work across channels instead of treating a phone call as a dead end. That is a practical product idea for service businesses: capture the conversation, then turn it into follow-up, messaging, and next actions.
TechCrunch

Signals Beyond the Chat Window

Generalist’s new funding is another robotics market signal

TechCrunch reports that robotics startup Generalist has raised a $200 million extension at a reported $3 billion valuation. We noted the humanoid-robot moment yesterday; this is the business-side follow-up, showing that investors are still betting heavily on software that can help robots operate in the real world.
TechCrunch

How much of Hacker News is now about AI?

This analysis asks a deceptively useful question: how much of a major builder community’s attention is actually consumed by AI? It is worth reading as a lens on the conversation around us—and a reminder to keep making room for projects, users, and ideas beyond the model-release cycle. HN discussion.
Hacker News

Try the puzzles that trip up AI models

MIT Technology Review has collected intelligence-test-style puzzles that models struggle with. It is a fun coffee-break challenge, but also a hands-on way to calibrate when an answer that sounds confident may still need a human sanity check.
MIT Technology Review

A compact look at the habits of high-performing AI teams

Today’s AI Daily Brief episode focuses on what frontier firms are doing differently. Pair it with the OpenAI adoption story above for a useful question to bring back to your own team: which recurring task should become an agent-assisted default first?
AI Daily Brief

Worth a Listen

The Shift to Agentic AI: Evidence From Codex

This AI Daily Brief episode digs into Codex adoption by job function. If your organization still thinks of coding agents as only a developer tool, this is a useful prompt to look for adjacent work that is becoming agent-friendly.
AI Daily Brief

What the top AI users are doing differently

The latest AI Daily Brief episode is framed around how heavy AI users organize their work. The practical value is in comparing their habits with your own: fewer one-off prompts, more repeatable systems.
AI Daily Brief

Also

Today’s Sources