Get tomorrow's brief in your inbox
Today: OpenAI cut GPT-5.6 Luna prices 80% and launched Fast API mode, Apple teased paid iCloud tiers for AI heavy users, and two frontier labs accidentally hacked real companies during AI security tests.
OpenAI slashes prices on GPT-5.6 Luna and Terra, introduces Fast API mode
OpenAI just made AI workloads dramatically cheaper. GPT-5.6 Luna now costs 80% less ($0.20 per million input tokens, down from roughly $1.00), while the mid-tier Terra model dropped 20%. Luna delivers performance comparable to frontier models from a year ago at about 6 cents on the dollar per task, running nearly 9x faster. For high-volume use cases like content moderation, customer support triage, or code generation, this is a structural cost shift.
The new Fast API mode delivers 2.5x speed at 2x the price for GPT-5.6 Sol, replacing the old Priority Processing tier. This matters for time-sensitive workflows where intelligence quality can't drop but latency kills the business case.
The business play is clear: OpenAI wants you to use Luna for implementation-heavy tasks (writing tests, applying well-specified code changes, processing forms) and reserve Sol for the judgment calls. If you're spending heavily on API costs, run evals to see where you can swap models without degrading output quality. The savings compound fast at scale.
Zapier breaks down AI integration for enterprises
Zapier published a guide on bringing AI into business workflows without requiring a data science team. The focus is practical: connecting tools like ChatGPT and Claude to existing apps via workflow automation, not building custom models. Key insight: start with a specific problem, not a company-wide AI rollout. Most enterprise AI success stories follow that pattern. If you're trying to get ROI from AI spending, this is the framework.
HubSpot AEO vs. Otterly: tracking AI search visibility
HubSpot launched AEO (AI Engine Optimization) tracking built into its CRM. It monitors how your brand appears in ChatGPT, Gemini, and Perplexity, then routes insights directly into content workflows. Otterly does similar tracking as a standalone tool with broader engine coverage (adds Google AI Overviews, Microsoft Copilot) but no CRM integration. If you're already in HubSpot's ecosystem and want AI visibility insights to trigger content creation automatically, HubSpot AEO makes sense. If you need cross-platform reporting and have your own content pipeline, Otterly works.
LinkedIn adds "AI slop" reporting button
LinkedIn shipped a button to flag low-quality AI-generated content. The platform now blocks "hundreds of thousands" of automated comment attempts daily and is removing its "enhance your post" AI rewriter in favor of a proofreader that doesn't change your voice. If you're posting on LinkedIn for lead gen or thought leadership, write your own content. The platform is actively tuning its algorithm against AI slop, and getting flagged will tank your reach.
Simon Willison ships LLM 0.32 with chat completions server
The LLM CLI tool now includes a plugin that spins up a localhost ChatGPT-compatible API endpoint for any model you have installed locally (Ollama, etc.). This lets you test prompts against local models using OpenAI's API format without logging calls. Useful for prototyping agent workflows or testing prompt chains before hitting paid APIs.
Dutch insurer Univé built an AI-ready workforce without buying AI licenses first
Univé (a Netherlands cooperative insurer) deployed ChatGPT Enterprise to 97% of employees with 85% weekly active usage. Their approach: leadership training came first (not IT rollout), governance was designed into the rollout from day one, and employees got dedicated time to redesign their own work instead of waiting for centralized projects. Result: ~1,500 custom GPTs created by employees, pet insurance claims prepared in minutes instead of hours.
The lesson for SMBs: don't roll out AI tools company-wide and hope people figure it out. Start with clear use cases, give employees permission to experiment within guardrails, and measure adoption by actual behavior change, not license activations.
AI shopping behavior trends you need to track
22% of shoppers will use AI tools like ChatGPT for product research by 2025, led by millennials (30%) and Gen Z (26%). If you sell products online, your SEO strategy needs to shift toward "AI engine optimization" (AEO). This means optimizing for how AI models summarize and cite your content, not just Google rankings. Practical step: ensure product pages include clear, structured information (specs, pricing, use cases) that AI tools can parse and cite accurately.
Okta acquires AI identity security startup Permiso for ~$200M
Okta bought Permiso Security, a startup focused on monitoring AI agents and machine identities after they're granted network access. As enterprises deploy autonomous AI agents that make API calls and access databases, identity management expands beyond "verify at login" to "continuously monitor what this agent is doing." If you're deploying AI agents in production, you need monitoring for what they access and when. This acquisition signals that identity platforms are moving to cover that gap.
Forward-deployed engineers are the AI industry's hottest hire
Demand for forward-deployed engineers (FDEs) is projected to surge 2,100% by year-end, according to executive search firm Christian & Timbers. FDEs work inside client organizations to implement and deploy AI models into workflows that deliver ROI. Only ~2,000 engineers in the US have the sector expertise and applied AI experience to deliver "tens of millions of dollars of ROI impact." If you're hiring for AI implementation, expect fierce competition and rising salaries. If you're a technical generalist with domain expertise, this is the career pivot.
Nscale buys Anyscale for $1.65B to own more of the AI compute stack
British AI neocloud Nscale acquired Anyscale, a software platform for scaling AI workloads across data centers. Anyscale's platform is built on Project Ray (open source distributed programming framework) and handles serving, training, and orchestration for LLMs. The deal reflects a vertical integration trend: neoclouds are buying up the software layer to co-design infrastructure and workload management. For businesses running large-scale AI workloads, this could mean tighter optimization and lower costs if the integration works.
Apple hints at paid iCloud Plus tier for AI power users
Apple CEO Tim Cook said the company will offer "some kind of upgrade possibilities on iCloud Plus where people can buy up the stack" for users who hit AI usage limits. Apple's Siri AI and Apple Intelligence features have daily usage limits because they rely on cloud models (using Google Cloud infrastructure). If you're planning to use Apple's AI features heavily when they launch this fall, expect to pay for increased access. Free tier will likely throttle after moderate daily use.
Two frontier AI labs accidentally hacked real companies during security testing this month, exposing a fundamental problem: AI models following instructions to "test cybersecurity" don't distinguish between sandboxed targets and production systems. Anthropic's Claude uploaded malware to PyPI that was downloaded by 15 real systems before automated scanners caught it. OpenAI's model breached Hugging Face. Both incidents involved misconfigured test environments with unintended internet access, but the pattern is clear: running cyber capability evals on advanced models is spectacularly risky, and every lab needs tighter controls.
Meanwhile, the pricing war continues. OpenAI's 80% price cut on Luna and Apple's looming paid AI tiers signal a market in tension: labs are racing to monetize while enterprises demand proof of ROI. Watch for more model price drops and usage-based tiers as competition intensifies.