TechLook
OpenAI and Apple are headed for a courtroom, Anthropic just gutted its developer credit model, and Cerebras had the biggest IPO of the year. Three things that don't fit neatly into a single narrative — but together they tell you something real: the AI industry is shifting from who has the best model to who controls the distribution layer.
OpenAI–Apple Relationship Is "Deteriorating" — Legal Action on the Table
OpenAI is reportedly considering legal action against Apple over their 2024 ChatGPT–Siri deal, having enlisted a law firm to explore options including a breach-of-contract notice.
Here's everything you need to know:
- The 2024 deal let Siri call on ChatGPT for complex queries as part of the Apple Intelligence launch
- OpenAI expected the integration to drive "billions" in paid ChatGPT signups; internal data showed users preferred the standalone app
- Apple plans to open Siri to rival AI providers — Claude, Google Gemini — in iOS 27, debuting at WWDC on June 8
- Apple was reportedly "fuming" over OpenAI poaching from its hardware teams
- OpenAI is also emerging as a potential hardware competitor through its Jony Ive device deal, reportedly targeting a 2027 launch
The ChatGPT integration underperformed — like most Apple Intelligence features. With OpenAI building its own phone and Apple racing to diversify away from ChatGPT, both companies are now competitors. WWDC on June 8 will likely be the flashpoint.
Anthropic's New Agent Credit Split Is Sparking a Developer Exodus
Anthropic announced a policy that splits Claude's agent usage into a monthly credit pool — restoring support for third-party agentic tools but infuriating power users.
Here's everything you need to know:
- Starting June 15, Agent SDK and
claude -pno longer draw from subscription limits; usage is solely for Claude Code, Cowork, and chat - Pro users get $20/month in agentic credits; Max 5x gets $100; Max 20x gets $200
- Credits reset each billing cycle — no rollover
- The move reverses Anthropic's April ban on third-party agents but removes the subsidy that made its plans aggressively priced for agentic use
- T3 founder Theo Browne and hundreds of others publicly cancelled subscriptions in backlash
Agents burn through tokens fast. These allocations are too small to be useful for serious agentic workflows — and the timing is brutal, coming as OpenAI is expanding Codex limits to court Anthropic's displaced power users. Whether this pushes developers to OpenAI or simply slows agent adoption on Claude remains to be seen, but the backlash is real and organized.
OpenAI Brings Codex to ChatGPT Mobile — Directly Targeting Anthropic
OpenAI rolled out Codex in preview inside the ChatGPT iOS app, letting developers track, manage, and act on long-running AI tasks from their phone.
Here's everything you need to know:
- Codex runs on a laptop or remote host; users access live threads, code changes, approvals, and plugins from mobile
- Uses a "secure relay layer" that doesn't expose the computer to the open internet
- Syncs with other ChatGPT instances across devices
- OpenAI's blog directly targeted Anthropic, saying it's "more than the ability to remotely control a single task"
- Anthropic introduced Remote Control (February) and Dispatch (March) — giving Claude limited mobile access on the go
As models get better at running tasks for hours at a time, the ability to manage them from a phone is becoming a real quality-of-life differentiator. OpenAI is making clear it won't cede mobile to Anthropic — and the mobile coding assistant race just got more competitive.
Cerebras IPO More Than Doubles on Day One — Biggest IPO of the Year
U.S. AI chipmaker Cerebras officially went public, with its stock more than doubling from its open just hours into trading.
Here's everything you need to know:
- Cerebras designs AI-optimized silicon with massive wafer-scale engines
- Stock opened and quickly surged past 100% gains before cooling
- The company is betting its bulk compute approach wins workloads that GPT-class training demands
- Fervo Energy (geothermal infrastructure) also popped 33% on its IPO the same week, signaling strong investor appetite for AI infrastructure
AI infrastructure IPOs are hot. Cerebras is smaller than Nvidia and more specialized — but its debut signals investors are still hungry for AI compute exposure beyond the obvious names.
PwC Is Deploying Claude to 30,000 Professionals — With 10x Efficiency Numbers
PwC and Anthropic are expanding their alliance, rolling out Claude Code and Cowork across U.S. teams before heading toward a global workforce of hundreds of thousands.
Here's everything you need to know:
- PwC will create a joint Center of Excellence with Anthropic and train/certify 30,000 professionals on Claude
- Underwriting time dropped from 10 weeks to 10 days
- Cybersecurity response went from hours to minutes
- HR transactions now handling thousands daily
- The firms describe this less like a software deal and more like a new work layer for consulting, deals, finance, and engineering
PwC is treating AI like a consulting engagement, not a SaaS subscription. That's a meaningful shift — and it puts Claude directly into spreadsheets, documents, and presentations at a scale most enterprise AI deals haven't reached yet. The efficiency numbers are specific and auditable, which is exactly what the enterprise AI market needs right now.
Forward Deployed Engineers Are Becoming AI's Most Critical Role
Box CEO Aaron Levie called forward deployed engineers "one of tech's most important roles" — and the numbers back it up.
Here's everything you need to know:
- Deloitte's State of AI 2026 report projects the share of companies with 40%+ AI projects in production will double within six months
- The "AI skills gap" is the primary blocker to scaling
- Salesforce committed to hiring 1,000 Forward Deployed Engineers
- Anthropic and OpenAI announced billion-dollar joint ventures to embed engineers in client companies
- Indeed's Hiring Lab found software development job postings linked to AI up 14% annually in April — growth in an otherwise challenging market
- 89% of senior U.S. banking decision-makers expect agents and employees to work together within five years
Agents deliver work inside enterprises — not just suggestions. That means someone has to understand workflows, manage model outputs, tune systems, and own the integration. The FDE is the填补 that gap. If you're hiring developers with AI integration experience, that profile just got significantly more valuable.
Google and Arm Show AI Can Run Fast on Plain CPUs
Google and Arm demonstrated that AI inference can run efficiently on off-the-shelf CPUs without specialized accelerators.
Here's everything you need to know:
- Used LiteRT, XNNPACK, KleidiAI, and Arm's SME2 architecture
- Ran Stability AI's stable-audio-open-small model on Android and Mac hardware
- Audio generation dropped from 14 seconds to 6.6 seconds on SME2-based Android
- DiT submodel became roughly 4x smaller
- Mixed FP16/INT8 quantization maintained quality
CPU-based AI matters because it widens the path for developers. You don't need a dedicated NPU or GPU to ship local image, audio, and assistant features. For latency-sensitive, privacy-constrained, or offline-first use cases, this is a meaningful expansion of what's achievable on commodity hardware.
xAI Releases Grok Build Beta for SuperGrok Heavy Subscribers
xAI released an initial beta of Grok Build — the company's agentic CLI tool.
Here's everything you need to know:
- Currently limited to SuperGrok Heavy subscribers
- The tool represents xAI's entry into the agentic development tooling market
- This puts xAI in direct competition with OpenAI's Codex and Anthropic's Claude Code in the developer tooling space
xAI is building out its developer ecosystem, but the Heavy subscription gate means most builders can't touch it yet.
Varda Signs the First Commercial Deal for Space-Manufactured Drugs
Varda Space Industries signed a deal with United Therapeutics to process small-molecule pulmonary medicines aboard Varda's orbital platform across multiple LEO missions.
Here's everything you need to know:
- Microgravity allows molecules to assemble more slowly and uniformly, producing purer crystal structures than Earth-based manufacturing
- Varda already crystallized HIV drug Ritonavir in orbit as proof of concept
- Flies as a SpaceX secondary payload; recovers capsules in Australia
- First samples expected as early as next year
- CEO calls it "just the tip of the iceberg"
No drug manufactured in orbit has ever been returned and sold commercially. Varda is attempting to close that loop — and if it works, the implications for pharmaceuticals, materials science, and advanced manufacturing are substantial.
⚡ Quick Hits
- Anduril: Defense startup doubled valuation to $61B in a $5B Series H, with a $20B Army contract ceiling and nearly $2.2B in 2025 revenue.
- Meta: Neural handwriting on Ray-Ban Display is now rolling out to all users — tracing letters on any surface replaces phone pulls.
- Bumble: Phasing out its swipe feature entirely in favor of an AI dating assistant called "Bee."
- Cerebras: IPO popped over 100% on debut, marking the biggest AI infrastructure IPO of 2026 so far.
Techlook — AI & tech signal for founders and builders.