The AI tools that actually run your business are about to live inside your phone.
On May 14, 2026, OpenAI announced that Codex — its desktop AI agent that can write code, operate applications, and execute tasks on your computer — is now controllable from the ChatGPT mobile app on iOS and Android. The preview is rolling out to all ChatGPT plans, including free users. You can start a task from your phone, and Codex running on your Mac or PC will execute it, streaming back screenshots, terminal output, diffs, and test results in real time. Files, credentials, and local setup stay on the machine. Your phone becomes a remote control for an autonomous worker that never sleeps.
This is not a minor feature update. It is the next phase of the AI agent wars.
What Just Happened
OpenAI has been playing catch-up ever since Anthropic’s Claude Code captured the imagination of developers in early 2026. The response has been aggressive: OpenAI shut down side projects like Sora, restructured its executive team, and refocused on building what it calls a desktop “superapp” — a single interface where AI does not just chat, but acts.
The Codex mobile launch is the latest move in that strategy. It means a business owner can review a Codex coding session while commuting, approve a command from a client meeting, or start an automation task while away from the desk. The friction between “having an idea” and “executing it with AI” has just collapsed to nearly zero.
Meanwhile, Anthropic is tightening the screws on its own ecosystem. From June 15, 2026, paid Claude plans will get a separate monthly credit pool for programmatic usage via claude -p and the Agent SDK — but it will be charged at full API rates. That change effectively ends the hobbyist-friendly era of cheap Claude Code automation and forces serious users into enterprise-style billing. The developer community has already responded with workarounds, but the signal is clear: Anthropic is monetising its lead.
Why This Matters for Your Business
For most business owners, the jargon around “MCP servers” and “agent SDKs” sounds like developer noise. It is not. What is happening here is the emergence of AI systems that can actually operate software on your behalf — not suggest a spreadsheet formula, but build the spreadsheet, connect it to your CRM, schedule the follow-up emails, and update your accounting system.
Codex on mobile means the boundary between “work device” and “personal device” is dissolving. Your team will soon be able to trigger complex, multi-step workflows from a phone notification. That is powerful — and risky.
The key shift is from conversational AI (ChatGPT asking and answering) to operational AI (Codex and Claude Code doing). OpenAI says 200 million people already ask ChatGPT finance questions every month. Now imagine those same users not just asking for budget tips, but instructing an agent to reorganise their entire bookkeeping workflow. That is the trajectory.
There is also a competitive angle. OpenAI is clearly willing to subsidise free-tier access to Codex mobile to win market share. Anthropic is moving upmarket with API-rate pricing. The result is a split market: OpenAI chasing volume, Anthropic chasing depth. For businesses, that means vendor choice is becoming a strategic decision, not just a technical one.
What You Should Do About It
If you run a business that depends on any form of digital operations — which is every business — here are three concrete steps to take this quarter.
First, audit your repeatable tasks. Look for workflows that involve copying data between apps, formatting reports, or updating records. These are the low-hanging fruit for desktop agents like Codex and Claude Code. If a junior employee does it twice a week, an agent can probably do it unsupervised.
Second, test one agent tool properly. Do not just prompt ChatGPT. Install Codex or Claude Code on a sandbox machine and give it a real, contained task — “update last month’s sales figures in our dashboard template” or “refactor this Excel macro.” You will learn more in one hour of hands-on use than in a month of reading about AI.
Third, understand the lock-in risk. These tools are not neutral platforms. They are tied to OpenAI’s or Anthropic’s models, billing structures, and increasingly, their view of what AI should be allowed to do. OpenAI’s recent move to connect ChatGPT to bank accounts via Plaid — letting the model see balances, transactions, and stock portfolios — shows where this is heading. The convenience is enormous. The concentration of trust in a single vendor is enormous too.
The Bottom Line
The desktop AI agent race is no longer a developer curiosity. It is becoming the operating system for knowledge work. OpenAI’s mobile Codex launch and Anthropic’s pricing pivot, both inside the last 48 hours, signal that 2026 is the year these tools cross from early adopter to early mainstream.
Businesses that treat this as a spectator sport will find their competitors have automated the work they are still doing manually. Businesses that engage now — testing, auditing, and building guardrails — will define how these tools get used inside their organisations, rather than having the tools define them.
If you want to explore what this means for your operations, your workflows, and your team, visit callumknox.com/advisory/. We help businesses cut through the hype and build automation strategies that actually stick.
Discover more from Callum Knox
Subscribe to get the latest posts sent to your email.
Ready to implement this?
Every article I write is backed by systems I have actually built. If you want the same results without doing it yourself, let me build it for you.
Discuss Your Project