Get tomorrow's brief in your inbox
May 15, 2026
Today: OpenAI brought Codex to mobile so developers can manage AI coding tasks from their phones. Microsoft started canceling Claude Code licenses to push engineers toward GitHub Copilot CLI. Coding agents made programming languages fungible enough that platform rewrites now take days, not months.
Codex went mobile, and the AI coding war got personal
OpenAI rolled out Codex access through the ChatGPT mobile app on iOS and Android this week, letting developers start tasks, review outputs, approve commands, and monitor progress from their phones while Codex runs on their laptops or remote environments. The feature is in preview now for all ChatGPT plans, including free.
This is different from remote desktop apps. When you connect to a machine running Codex, the app loads live state from that environment, including active threads, screenshots, terminal output, test results, and approval requests. Your credentials and files stay on the machine where Codex operates. Updates flow back to your phone in real time through a secure relay layer that keeps machines reachable without exposing them to the public internet.
For small business owners running technical teams or managing development work, this changes the rhythm of collaboration with AI agents. You can start a bug investigation while waiting for coffee, unblock Codex mid-commute when it needs direction, or spin up a customer briefing from Slack threads before a call. More than 4 million people now use Codex weekly, and OpenAI says those small check-ins prevent unnecessary rework and keep long-running tasks on track. If you've been using Claude Code or other desktop coding agents, Codex mobile gives you similar flexibility without switching tools.
OpenAI announcement | TechCrunch coverage | The Verge coverage
Microsoft pulled the rug on Claude Code
Microsoft is winding down Claude Code licenses across its Experiences + Devices teams (Windows, Microsoft 365, Outlook, Teams, Surface) by June 30th and pushing developers to GitHub Copilot CLI instead. The company says this is about converging on one agentic command line tool, but sources tell The Verge it's also a financial decision, an easy expense cut at the end of the fiscal year.
This matters because Microsoft's own developers favored Claude Code over Copilot CLI in recent months. The company had encouraged employees without coding experience to prototype ideas with Claude Code, and now those designers and project managers will need to transition workflows. Microsoft says it's investing more in Copilot CLI and working closely with GitHub to close feature gaps. Anthropic's models will remain accessible through Copilot CLI alongside OpenAI and internal Microsoft models.
If you're evaluating AI coding tools for your team, this is a reminder that corporate licensing decisions override product preference. Pick tools you can afford to switch away from if pricing or politics change.
Zapier vs. Composio: governance wins
Zapier published a comparison with Composio, the integration layer for AI agents. Both handle authentication so agents can connect to external apps without exposing API keys, but Zapier offers far more than credential brokering. It includes MCP for connecting AI tools like Claude and ChatGPT to 9,000+ apps, Zaps for automated workflows, Tables for structured data storage, and Canvas for process mapping. Composio covers 1,000 toolkits with a developer focus but lacks agents, databases, workflows, and a visual builder.
The governance layer matters more than most people expect. When you switch agent platforms (ChatGPT to Claude, for example), you lose access controls, audit logs, app connections, and context. Zapier's approach keeps those layers portable across models. If you're building agent workflows for a team, centralizing integrations and governance through a platform like Zapier means you can swap models without rebuilding your operational infrastructure.
Zapier uses task-based pricing. Composio prices per tool call, which can get expensive fast if your agents are chatty.
Sea Limited is betting on AI agents to manage complexity at scale
Sea, the Singapore-based tech giant behind Shopee, is rolling out Codex across its developer organization with 87% weekly active usage. David Chen, co-founder and Chief Product Officer, says the company sees AI-assisted development as a structural multiplier for navigating complexity in fragmented, hyper-localized markets across Southeast Asia.
The key insight: developers aren't using Codex to type faster. They're using it to think better. Sea's teams now integrate AI agents into CI/CD pipelines to propose test-driven implementations, surface edge cases in distributed systems, and accelerate debugging loops. By letting AI rapidly prototype alternative implementations and generate exhaustive test coverage, the company is shipping more resilient systems while paying down technical debt.
If you run a technical team, this is the framework to steal. Use agents to improve engineering discipline, not just velocity. Generate test coverage, explore edge cases, and prototype alternatives before committing to an approach.
Personal software is eating the world
The Verge published a deep dive on the "personal software revolution," where AI coding tools let non-developers build custom apps for themselves, not for venture capital or eventual acquisition. With $20 a month and a half-formed idea, Claude Code or similar tools can build functional software tailored exactly to your needs.
Robin Sloan, author and technologist, built a family messaging app five years ago and hasn't touched the code since. Now he's using AI to whip up Ruby scripts that pull data from Shopify and USPS to auto-generate shipping labels for his olive oil company. These are weird little tools, one-offs that solve a specific problem and disappear when they're no longer needed.
For small business owners, this is permission to think differently about software. You don't need a subscription SaaS product with features you'll never use. Build a custom meal planner, a family trip scheduler, or a budget tracker with exactly the fields you need. Use it forever, use it once. It's your software.
OpenAI might sue Apple over ChatGPT integration that flopped
OpenAI is exploring legal action against Apple over a ChatGPT integration announced at WWDC in June 2024 that failed to deliver the subscribers and prominence OpenAI expected. The company has enlisted an outside law firm to work through options, including a breach-of-contract notice, though any legal move will likely wait until after the Elon Musk trial concludes.
OpenAI expected the deal might funnel billions in new subscriptions and give the company prime real estate across one of the world's most-used mobile ecosystems. Instead, Bloomberg reports, the integration has been buried, features are hard to find, and revenue is nowhere close to projections. Apple has its own grievances, including concerns about OpenAI's privacy standards and irritation over OpenAI's push into hardware led by former Apple executives including Jony Ive.
This is a reminder that partnering with platform owners is risky. Apple has a long history of embracing partners and then alienating them (Google Maps, Adobe Flash, Spotify). If you're building on someone else's platform, have a backup plan.
Recursive Superintelligence raised $650M to build self-improving AI
Richard Socher (founder of You.com) came out of stealth with Recursive Superintelligence, a San Francisco startup working on AI models that can autonomously identify their own weaknesses and redesign themselves to fix them, without human involvement. The company raised $650 million and includes researchers like Peter Norvig and Tim Shi.
Socher's approach uses open-endedness to achieve recursive self-improvement. Instead of just asking AI to improve something, the entire process of ideation, implementation, and validation of research ideas becomes automatic. The company is focusing on AI research first, eventually expanding to any kind of research, including physical domains.
For business users, this is a "watch and wait" story. Recursive self-improvement is a research moonshot, not a product launch. If it works, it could dramatically accelerate AI capability improvements. If it doesn't, it's still interesting science.
Cisco cut 4,000 jobs despite record revenue
Cisco laid off nearly 4,000 employees (5% of its workforce) despite reporting better-than-expected profit and revenue in its fiscal third quarter. The company says it's changing its cost structure to invest more in AI and cybersecurity. CEO Chuck Robbins touted "record revenue" and "double-digit growth" while acknowledging strategic investments in AI across the company. Robbins was slated to earn more than $52 million in executive compensation during 2025 and declined to comment on whether he plans to reduce his own pay.
This follows a trend of tech companies citing AI spending as a reason to cut staff despite strong financial results. Cloudflare and General Motors have done the same in recent days. For small business owners, this is a signal that enterprise vendors are prioritizing AI tool development over human support. Expect more self-service AI tools and fewer human account managers.
TechCrunch coverage | Ars Technica coverage
Programming languages are no longer lock-in
Mitchell Hashimoto (founder of HashiCorp) pointed out that Bun's migration from Zig to Rust, completed in roughly a week or two with coding agents, proves programming languages are no longer lock-in. If a platform rewrite now takes days instead of months, you can pick the language that works best today and switch later if it becomes a problem.
For small businesses, this means you're no longer stuck with a technology choice forever. If you build something in Python and later decide Go is a better fit, coding agents make the port feasible. Don't overthink the initial language choice.
Simon Willison quote | Mitchell Hashimoto quote
Build an agent harness, not an agent
Zapier published a helpful explainer on what an "agent harness" is: everything wrapped around an AI model that turns it into a working agent. That includes tools, memory, state persistence, context management, sandboxed execution, guardrails, and the logic that ties it all together. The model is just the brain. The harness is what makes it useful for real work.
When you switch agent platforms, you lose the harness: governance, app connections, and context. If you're building AI workflows for a team, focus on keeping the harness (integrations, access controls, audit logs) portable so you can swap models without rebuilding everything.
AI coding tools are shifting from productivity aids to infrastructure layers. The Codex mobile launch and Microsoft's Claude Code pullback show that these tools are now embedded deeply enough in workflows that access policies matter as much as features. Meanwhile, the personal software movement suggests we're entering an era where custom-built, single-use apps become as common as spreadsheets. Watch for more businesses treating AI agents as permanent team members with dedicated governance and tooling, not experimental side projects.