Need to know
- Microsoft launched seven in-house MAI models at Build 2026, including a reasoning model that matches Claude Opus on coding benchmarks.
- OpenAI Codex now has 5 million weekly active users and ships enterprise plugins targeting non-developer knowledge workers.
- Google opened Gemini Extended thinking to all free and paid users today alongside a new Gemini 3.5 Flash iteration.
- Meta Business Agent went globally available on WhatsApp, Messenger, and Instagram for businesses of all sizes.
- CrowdStrike appointed former Nvidia AI engineering leader Bartley Richardson as its first Chief AI Officer.
New Releases
Microsoft unveiled seven in-house AI models at Build 2026, led by MAI-Thinking-1, a 35B active-parameter MoE reasoning model built from scratch on commercially licensed data.
- MAI-Code-1-Flash outperforms Claude Haiku 4.5 by 16 points on SWE-Bench Pro while using 60% fewer tokens and is already rolling out inside GitHub Copilot and VS Code.
- Microsoft Frontier Tuning launched alongside the models, letting enterprises create custom AI models and agents tuned to their own data without managing pretraining infrastructure.
OpenAI shipped job-specific enterprise plugins for Codex and a new Sites feature that lets users build and deploy interactive web apps directly from the agent.
- Five million weekly active users now use Codex, a sixfold increase since the desktop app launched; one in five users is not a developer, and that segment is growing three times faster than the developer cohort.
- Six role-specific plugins target analysts, designers, salespeople, and bankers, backed by an internal OpenAI report documenting Codex adoption across knowledge work functions.
Google made Gemini's Extended thinking mode available across all free and paid tiers today, simultaneous with a new Gemini 3.5 Flash iteration that improves endurance on hard tasks.
- Extended thinking lets Gemini pause and reason before responding; Deep Think remains restricted to AI Ultra subscribers, preserving a ceiling for paid tiers.
- Rate limit counters were reset to zero for all users alongside the Flash update, which was pushed to fix output quality drop-offs observed in the Antigravity deployment.
Meta launched Meta Business Agent globally today for WhatsApp Business, Messenger, and Instagram, enabling businesses to automate customer service, product recommendations, and sales closures.
- Initially free, the service will shift to conversation-based pricing on WhatsApp Business Platform, with India-specific rates that signal Meta's priority on high-volume emerging markets.
- Meta Business Agent Platform gives larger enterprises infrastructure to build, customize, and deploy branded agents at scale, competing directly with Salesforce Agentforce and HubSpot's AI tools.
Funding
Mizuho cut Okta to Neutral today, signaling analyst concern that AI-native identity competitors and Microsoft Entra's expanding feature set are eroding Okta's pricing power even as the company posted strong Q1 results on AI agent identity demand.
CrowdStrike appointed Bartley Richardson, former Nvidia engineer behind the NeMo Agent Toolkit, as its Chief AI and Autonomous Systems Officer, betting that AI-native threat detection requires frontier model expertise at the C-suite level.
Case Studies
Critical Start released production results from SOC AI, a ten-agent MDR framework coordinating detection, triage, response, threat hunting, and continuous improvement across the full alert lifecycle.
- Each of the ten agents operates with a discrete function and a complete audit trail, directly addressing the compliance and explainability gap in autonomous security operations.
- Production-proven framing distinguishes this from lab claims; the framework is live inside Critical Start's managed service, not a preview or pilot.
Energy technology firm DrillDocs deployed KeeperPAM to eliminate credential risks across 24/7 engineering workflows on a globally distributed, BYOD workforce with no fixed network perimeter.
- Named outcome: credential risk eliminated across offshore operations where shared devices and shift handoffs created persistent privileged access exposure.
- BYOD at scale in a critical-infrastructure context is the specific use case most PAM vendors struggle to address without requiring MDM enrollment.
Trending on X
- Codex as non-developer work platform Developers on X are debating whether OpenAI's Codex growth stat — one in five users isn't a coder, growing 3x faster — marks the moment AI coding tools become general-purpose enterprise software, with swyx calling it 'AGI' after one-shotting a complex app.
- Trump AI executive order reactions Sam Altman and David Sacks are both amplifying the new EO as the right balance between frontier model development and cyber defense, while critics argue voluntary submission for cybersecurity testing has no enforcement teeth.
- Anthropic vs OpenAI valuation parity After Anthropic's $965B valuation surpassed OpenAI's estimated $900B on nearly identical revenue, the AI bubble debate is back on X with skeptics pointing to both companies' ~$20B revenue against near-trillion-dollar valuations.
- Microsoft MAI model benchmarks scrutiny Practitioners are picking apart Microsoft's claim that MAI-Code-1-Flash beats Claude Haiku on every benchmark, questioning whether SWE-Bench Pro reflects real production workloads or is a cherry-picked metric for the announcement.
- Sophos AI malware lab disclosure Security researchers are reacting to Sophos catching a live attacker lab running Claude Opus 4.5 as a coordinator for multi-agent EDR bypass development, calling it the transition from theoretical AI-assisted malware to observed practice.