PRODUCTS

OpenAI turns a phone into an agent command center

OpenAI's Codex has been on fire lately, with weekly users ramping up to four million. Now you can run it from your phone. 

On Thursday, OpenAI announced a host of upgrades to let you monitor and manage a Codex agent from the ChatGPT app on your iPhone or Android device. This is becoming more and more critical as AI agents like Codex handle long-running, complex tasks that can take hours to complete. 

Codex is part of the new wave of personal AI agents that run on your local system, such as OpenClaw, Claude Code/Cowork, and Perplexity's Personal Computer. However, that means you're tied to the system if it needs feedback to proceed or if you need to stop it to change course. 

One of the biggest breakthroughs of OpenClaw was that you could chat with it via text messaging to keep it on track while you were away from your computer. I've heard OpenClaw founder Peter Steinberger say he originally got this idea because he wanted to interact with his agent from his phone while he was in the kitchen making dinner. 

That's the spirit of what Codex is announcing today, but it's taking it a step further. 

Now, from the ChatGPT app, you can not just message your agent, you can also:

• See all of the tasks your agent is running 

• Review outputs 

• Approve commands

• Change models

• See screenshots

• See terminal output

• Start a new task 

Codex is also getting upgrades that will benefit teams and enterprises. That includes "remote SSH, Hooks, programmatic access tokens for business and enterprises, and HIPAA-compliant use of Codex for eligible ChatGPT Enterprise workspaces," according to OpenAI.

Codex has already been gaining remarkable momentum in the market over the last few months, as OpenAI has pivoted away from its broad consumer focus to lean into coding and helping startups and enterprises leverage the growing capabilities of local agents. The timing has been fortuitous, as its primary rival, Claude Code, has been stymied by Anthropic's choppy uptime issues stemming from its rapid growth. But OpenAI's mobile move with Codex goes beyond what any of its rivals are currently doing and is likely to attract even more AI enthusiasts, who want to be less chained to their desks. We shouldn't be surprised, since Steinberger has been working at OpenAI for the past several months, helping the Codex team broaden and upgrade their AI agent. This move is another signal of the upcoming merger of the Codex, ChatGPT, and the Atlas web browser into the OpenAI superapp.

TOGETHER WITH GUILD

Every engineer is shipping agents. No one is governing them.

Shipping agents is easy. Knowing what each agent can access, what they did, how much they're spending, or who's responsible? Not so much.

Guild is the control plane for AI agents — a unified platform to govern permissions, track actions, control spending, and manage all your agents safely in production.

Bring your own models. Plug in your own tools. And fork and deploy any agent in minutes to share in the spoils. 

Guild is the control plane. The story is yours. Available now.

Get a demo

CULTURE

Americans' disdain for data centers hits new highs

AI already has a publicity problem among the average American. Data centers are making it worse. 

A Gallup poll published this week found that an average of 7 in 10 Americans oppose data center construction in their area, including 48% who strongly oppose such facilities being built locally. Only about a quarter reported favoring local construction of data centers, with just 7% strongly in favor. 

Respondents reported a wide range of concerns about these facilities being built in their neighborhoods: 

• Environmental impacts were an overarching theme, with 50% worried about data centers' impact on resources, 18% about excess water usage, 18% about power grid constraints, and 16% concerned about pollution.  

• Around 22% worried about quality-of-life concerns, including impacts on property values, traffic and housing. And 20% reported concerns about data centers' impacts on costs, such as higher utility bills and cost-of-living increases. 

• Meanwhile, 14% reported that they simply have negative views of AI that broadly shape how they view data centers. 

The poll also found that opposition to data centers was largely bipartisan, with 75% of Democratic respondents, 74% of independent respondents, and 63% of Republican respondents reporting opposition. Gallup noted that this could mean AI infrastructure could become a hot-button campaign issue. 

“Overcoming this opposition stands as a major hurdle in the expansion of AI computing,” Gallup said in its press release. “The intensity of opposition means that proposed data centers are likely to spur grassroots activism from local residents as well as legal challenges.”

But not all of the opinions were negative. Among those who reported being in favor of development, 66% anticipated that data center development would bring local economic benefits, such as job opportunities, tax revenue and improved infrastructure. 

This poll points to an increasingly prevalent trend: Outside Silicon Valley's bubble, non-technical people largely do not like or understand AI. This is also evidenced by the fact that only 16% of people use AI regularly, according to a Microsoft study.

It’s no secret that AI has a PR problem. Practically every day, headlines are bubbling up that AI is poised to shred the economy and force thousands of people out of their professions, especially if they can’t or won’t use it. This narrative is equally fueled by horror stories of people slipping into AI psychosis the more they interact with these models, as well as looming figures related to AI’s environmental impact. But AI faces an existential paradox: Without adoption by a wider user base, today's AI labs will struggle to achieve both the returns and the transformative impact they have promised their technology will deliver. To gain that adoption, AI providers need to change the narrative. The problem, however, is how to do so. Because this technology is so nascent, we truly don’t know the scale of the impact it will have on our economy, society, environment or sense of human purpose. Seeking to quell the mania by explaining how everything will be okay could be a herculean task when it’s too early to accurately make those promises. This will require creativity to find ways to engage more people in constructive conversations about AI. It also wouldn't hurt if the labs could make the tools easier to use and deliver greater benefits in daily life.

TOGETHER WITH QA TECH

Where AI velocity breaks

Version control is automated. Deployments are automated. Code review is AI-assisted.

Yet most teams still rely on humans to validate releases – or skip it entirely because there isn’t time.

The same AI accelerating your development velocity is making manual QA impossible to scale.

What dozens of engineering teams discovered the hard way, our founders described in their field guide to quality in the AI-SDLC.

Nordea, Pricer, and Upsales use QA.tech to reclaim thousands of hours every month.
Not by hiring more QA engineers – by replacing manual validation with agents that validate every release automatically.

See how it works

WORKFORCE

Claude Cowork: Start with these 3 steps

If the only way you're using AI at this point is through chatbots, then it's time for an upgrade.

Since the end of 2025, local AI agents have greatly multiplied the value that you can get out of these tools. It started with Claude Code, a command line tool for building software that people quickly figured out could be used for automating lots of other things. So Anthropic adapted and created Claude Cowork, a more user-friendly version of the same local agent that was built into the Mac desktop app.

Since then, competitors such as OpenAI's Codex and Perplexity's Personal Computer—and of course, OpenClaw and its offspring—have emerged to make these personal AI agents that run on your local computer the hottest trend of 2026. 

However, a recent poll on The Deep View showed that only 29% of its AI-focused readers have tried Claude Code or Cowork yet. It's safe to say the number would be considerably lower among non-early adopters. So we're going to help you remedy that. 

To begin, you'll need to make sure you have one of Anthropic's paid plans that start at $20 per month. The same is true for using Perplexity's Personal Computer. However, for OpenAI's Codex, there is a limited access version of Codex that you can use even if you're on the free plan.

For this example, we're going to stick to Claude Cowork, because it's the most mature of the personal AI agents (and arguably the easiest to get started with). 

Next, you need to install the Claude desktop app on Mac or Windows. I'm going to also recommend that you install Claude for Chrome because that will allow Claude Cowork to automatically handle tasks for you that require internet browsing. 

Once you've got it installed, open the Claude desktop app, login, and then look up in the left corner of the window. By default, you'll be in the chat interface. Click "Cowork" in the middle of Chat and Code. Now click "Work in a project" and then "Choose a folder."

Here's where you start to decide what you want to do with Cowork. You can give it access to your main Documents folder, for example. Or you could create a folder called "Claude Cowork" where you can drop copies of specific files or folders of files that you want it to work on. I recommend doing this at the start so that Claude doesn't mangle or delete any important files by accident. 

Once you've got that set, here are three of the first ways you can start getting value with Cowork beyond what you'd get from a chatbot alone.  

1. Organize a messy stack of files: By far, the first and most common way that people try out Cowork is to give a stack of files to organize on your computer. Some people who have an overloaded set of files on their desktop will start there. But I'd recommend taking a folder that has a mix of stuff from an old computer or a bunch of photos from a camera or phone, and simply ask Claude Cowork to organize them for you. If you're doing photos, you can also ask Cowork to rename them with a consistent naming convention that uses the data and the information on what the photo is.

2. Summarize a bunch of documents: If you have folders full of Word docs, Excel files, PowerPoints, meetings notes, scanned receipts and other documents that are all on a related topic, then you can ask Cowork to summarize all of the contents in one easy-to-read document. If this sounds like something you'd do already in a chatbot, then there's one big difference to note: Chatbots have limits to the size of files you upload. With local AI agents like Cowork, the sky is the limit. You can even point it at giant video files and have Cowork do a timeline summary for you, for example. 

3. Draft an email from a set documents: You can also take a meeting recording, a bunch of notes, a PowerPoint, a PDF, and literally any other kind of document you can think of, put them all in one folder, point Cowork at it, and then tell it to draft an email for you summarizing the contents for an update to send to your Coworks. This can be great for project updates you need to send to colleagues, for example.

Knowing these tasks and starting to use them, you'll quickly discover other ways you can put a local AI agent to work for you.

LINKS

• Anthropic, Gates Foundation pledge $200 million for AI education

• SpaceX, xAI face exodus of AI researchers 

• Cerebras IPO opens at $350 per share, closes at $95 billion market cap

• OpenAI weighs legal action against Apple over lacking Siri benefits

• Meta opens Ray-Ban display glasses to develop preview

• AI firm Recursive raises $650 million at $4.65 billion valuation

• Runway Agent: The AI video firm has released a “creative partner” for ideation and execution of projects. 

• Recraft V4.1: A new image model from Recraft for more photorealistic humans and “dreamier” gradients

• Notion Developer Platform: The workspace platform has added a developer hub and extended capabilities for AI agents.

• Spellar 3.0: An AI meeting companion that records locally, so you don’t have to add bots to your meetings.

Top LATAM tech talent, vetted and ready to ship.

Build a strong team, without the usual hiring headaches

• Get AI-assisted matching and human vetting tailored to your needs in days.

• Save up to 70% on salary costs while still hiring high-skill professionals.

• Scale with a repeatable process that keeps quality high and effort low.

Scale with Athyna

(sponsored)

GAMES

POLL RESULTS

Do you trust AI model companies to help shape regulation around their technology?

Yes (13%)
Somewhat (37%)
No (44 %)
Other (6%)