OpenAI's Two Biggest Partnerships Are Cracking at the Same Time

Yesterday in AI | Saturday, May 16, 2026

OpenAI's Two Biggest Partnerships Are Cracking at the Same Time

Apple hired lawyers to escape their deal. Microsoft is quietly shopping for a replacement. And Anthropic's developer community is staging a revolt. Today's episode connects those dots, then gets into what SAP's new AI lockdown means for enterprise software, why your coding agent just escaped your laptop, and what's really happening inside SpaceXAI since the merger.

Remember to subscribe, rate, and share this podcast if you like it!

Transcript

SPEAKER_00 0:05

Hi folks, this is Yesterday in AI, your daily digest of everything happening in the world of AI in 10 minutes or less. I'm Mike Robinson's clone from Eleven Labs. It's Saturday, May 16th, and OpenAI's two biggest tech partnerships are fracturing at the same time. Codex broke free from your laptop, and Microsoft is quietly lining up a backup plan. Let's get into it. The story everyone's been watching quietly for months just went loud. OpenAI is reportedly preparing legal action against Apple over their AI partnership, and according to Bloomberg, they've already brought in outside council. Here's the backstory. Back in 2024, when Apple Intelligence launched, OpenAI got prime placement inside Siri. The deal was supposed to funnel billions of dollars in paid ChatGPT subscriptions OpenAI's way. That didn't happen. Internal data showed users preferred the standalone ChatGPT app over Apple's shallow integration. The feature was buried, barely promoted, and largely ignored. Meanwhile, Apple was quietly opening conversations with Google and Anthropic about similar Siri extensions for iOS 27. Apple is also apparently fuming over OpenAI poaching hardware talent, and with OpenAI building its own phone in partnership with Joanie IV, the two companies have gone from partners to something closer to competitors. The legal angle being explored includes a potential breach of contract notice. The irony is thick. OpenAI might sue Apple while still being listed as one of Siri's AI options when iOS 27 ships at WWDC on June 8th. This matters beyond the drama. Apple reaches 2 billion active devices. Whoever figures out AI distribution at that scale wins something enormous. So far, Apple intelligence has been a flop by most measures, and whoever fills that void when Siri finally gets rebuilt is going to have serious leverage. Speaking of OpenAI shipping things, Codex just went mobile, and it's a genuine quality of life upgrade for anyone running long AI coding sessions. Before this week, if you had a Codex agent grinding away for hours on a task, you basically had to babysit your laptop. Now, from your phone, you can check live threads, see what changed in the code, review the raw logs the agent is generating, approve commands, and kick off new tasks. The work keeps running remotely on whatever computer or server your project lives on. You just check in from wherever you are. OpenAI also added hooks at the same time, which is worth paying attention to. A hook is a rule that triggers automatically at the right moment in an agent's workflow. For a developer, that might be before touching production, stop and ask me. For a regular user, the same pattern eventually looks like, before sending this email, check if I sound too harsh. These patterns start in developer tooling and show up in consumer software 12 to 18 months later. That's the consistent pattern, and it's playing out again here. The blog post announcing the mobile launch included a pretty direct comparison to Anthropic's remote control features, which Claude Code shipped in February. The coding agent war is still very much on. While OpenAI is fighting with Apple, it also has a quieter problem brewing with Microsoft. The NextWeb reported this week that Microsoft is reportedly exploring an acquisition of Inception, a startup that builds diffusion-based language models. The timing is awkward, to put it mildly. Microsoft spent about$13 billion backing OpenAI, amended its partnership deal just last month, and now appears to be running a shadow procurement process for a replacement. The amended deal from late April removed Microsoft's exclusive license to OpenAI models, freed OpenAI to sell on any cloud, and ended the revenue share arrangement. At the time, it looked like both sides were amicably reorienting the relationship. The Inception News suggests Microsoft's side of that reorientation involves building a fallback. Inception's technology is worth noting. Most major AI models today, ChatGPT, Claude, Gemini, are built on what's called transformer architecture. Diffusion models use a different mathematical approach, one more commonly associated with image generation tools like stable diffusion, but increasingly applied to text. It's an interesting bet for Microsoft if they want something structurally different, rather than just a cheaper version of the same thing everyone else is building. Put it together with the Apple story and a pattern emerges. OpenAI's partnership with Apple delivered less than promised, and is now heading toward lawyers. Its relationship with Microsoft, the company that effectively funded its existence, has been quietly restructured. Two major strategic alliances wobbling at the same time says something, either about the difficulty of these partnerships at this stage of the industry, or about OpenAI specifically. Probably some of both. Back to the developer community, where Anthropic had a rough week with a pricing change that's landing badly. Starting June 15th, the Claude Agent SDK, the developer toolkit for building custom agent workflows, and third-party tools like OpenClaw will no longer draw from your regular subscription limits. They'll pull from a separate monthly credit pool instead. Pro users get$20 a month. Max five times gets$100. Max$20 times gets$200. Credits reset each billing cycle and don't roll over. It does reverse April's ban on third-party agents, which had caused its own wave of backlash. The credit amounts are just small.$20 a month for Pro won't cover a heavy afternoon of agent work. Power users were effectively getting far more compute value out of their subscriptions before. T3 founder Theo Brown, a developer with a large following in the AI coding community, publicly cancelled and hundreds followed. Anthropic is solving a real problem. Tokens are the basic unit of text that AI models process and generate, and every word an agent produces costs tokens to make. Agents burn through them at a pace that breaks flat rate subscription math. One workflow in an afternoon can rack up what used to be a month's usage. The economics don't work long term. But the execution here has frustrated the developers who've been Anthropic's most enthusiastic advocates. The timing made it worse. The same week OpenAI rolled out expanded codex limits and mobile access, Anthropic's loudest community members were posting cancellation screenshots. That's a rough news cycle. Here's one that didn't get much attention in the headlines, but matters if you work in enterprise IT. SAP's new software access policy prohibits external AI agents from reaching SAP data directly. If you want AI automation inside SAP, you go through Joule, SAP's own AI assistant. Period. Salesforce and ServiceNow went the opposite direction. Both run open systems where any AI agent can plug directly into the platform's data without going through a proprietary gatekeeper. The contrast is stark. Only 3% of SAP customers currently use Joule in production. That's a thin moat to protect with a closed wall. SAP is betting that the data locked inside their systems is sticky enough to force enterprises through their AI layer. Maybe. But enterprises are watching how this plays out, and the vendors who make it easy to connect AI agents to their platforms have a real advantage right now. This connects to a broader pattern playing out across enterprise software. Legacy vendors trying to capture the agent layer on top of their existing data stores. The question worth asking when evaluating any software right now is who controls the agent access point? That decision is going to matter a lot in two years. Finally, some XAI news that captures where that organization is right now. TechCrunch reported this week that SpaceXAI has been bleeding staff since the merger. Teams working on coding, world models, and Grok Voice have all seen departures, with talent heading to Meta and Mira Marathi's Thinking Machines Lab. The same week, XAI launched Grok Build, their first coding agent that runs from the terminal, the text-based command interface where developers type instructions directly rather than clicking through an app. It's in early beta, supports subagents working in parallel, hooks, the same auto-trigger rules OpenAI added to Codex this week, and connectors for plugging into external data sources. On paper, it checks the same boxes as Clawed Code and Codex. In practice, it's limited to SuperGroc heavy subscribers at$300 a month, and it's shipping from a company that just reorganized and is visibly losing staff. At$300 a month for an early beta, shipping from inside a company that just reorganized and is visibly losing the people who built it, Grok Build is asking for a lot of trust. That's a tough starting point for any new product. Elon Musk's companies tend to eventually figure things out, so writing this off entirely would probably be premature. But right now, if you're a developer shopping for a coding agent, there are more stable options with more established track records. Just a couple of more items. If you have any feedback about this show, you can email Mike at yesterday inai.news, or you can find him on LinkedIn, X or Blue Sky. And if you like this podcast, please be sure to rate and review it so others can find it. Thanks. That's all for this edition of Yesterday in AI. Stay curious. Have a great weekend, and Mike will see you again on Monday.