Morning Digest, June 24, 2026

12 newsletters, 6 overlapping stories

Top Stories

OpenAI expands Daybreak and ships GPT-5.5-Cyber to patch software at scale

(5 newsletters)

OpenAI broadened its Daybreak security push with an updated Codex Security plugin, a limited release of GPT-5.5-Cyber for vetted defenders, a partner program, and an open-source effort called Patch the Planet. The model scored 85.6% on the CyberGym benchmark, and Codex Security has already scanned more than 30 million commits across 30,000-plus codebases, with over 500,000 findings auto-marked as fixed. The bigger shift is that AI security is moving from finding bugs to landing the actual fix, with projects like cURL, Go, and Python signed on. The timing is notable: Anthropic’s stronger Mythos and Fable models remain pulled under export controls, and OpenAI is seizing the opening.

Anthropic puts Claude inside Slack as an always-on coworker

(4 newsletters)

Anthropic launched Claude Tag, which lets teams tag @Claude in Slack like a teammate to handle tasks asynchronously across channels, codebases, and tools. Claude breaks a request into stages, works through them with approved tools and data, and builds context over time, including an ambient mode that follows up on stale threads. It replaces the existing Claude Slack app and is rolling out to Enterprise and Team customers. Andrej Karpathy called it the third major redesign of LLM interface design, since Slack is where most business context already lives. Anthropic is also preparing Cowork support for mobile.

Meta drops the Ray-Ban brand for $299 in-house smart glasses

(3 newsletters)

Meta unveiled “Meta Glasses,” a cheaper line built with EssilorLuxottica and powered by its Muse Spark AI out of the box, starting at $299. The lineup includes Adventurer, Fury, and a $399 Kylie Jenner edition with a custom chime and her voice option for Meta AI. The hardware is largely unchanged from prior models, so the story is the price and the move toward a two-tier strategy: Ray-Ban for fashion, Meta Glasses for accessibility. Meta still holds roughly 80% of the smart glasses market.

GLM-5.2 lands as the strongest open model and changes the cost math

(3 newsletters)

Z.ai’s GLM-5.2, with a 1M-token context window, is being ranked by many developers on par with Opus 4.8 and GPT-5.5, the first time an open-weights model has passed the developer vibe check. It still trails frontier models on long, messy multi-week projects, and it lacks vision, but it runs at a fraction of the per-token cost. The practical takeaway is routing, not switching: send deep reasoning to a frontier model and route routine, well-defined work to GLM-5.2, which already runs inside Claude Code.

Agent “loops” emerge as the defining workflow trend

(4 newsletters)

Looping, where you treat an agent like an employee by setting a goal, clear instructions, and success benchmarks, then letting it plan, execute, iterate, and verify on its own, is being called AI’s next big shift by voices including Claude Code creator Boris Cherny. The verification step is the critical piece, since without it a loop burns usage limits churning out unusable output. Coverage also flagged the downside: looping can reduce developers to messengers and pile up “prompt debt” when natural language is overused as a specification language. Loop engineering reframes the work around stopping conditions, context rot, and independent verification.

Meta builds a prediction markets app, tanking betting stocks

(2 newsletters)

Mark Zuckerberg directed a small team to build a Polymarket-style prediction markets app known internally as “Arena,” walled off from Facebook, Instagram, WhatsApp, and Messenger. It uses a points system rather than real money for now, though real-money betting has not been ruled out. The report alone sank DraftKings and Robinhood shares.

Also Worth Knowing

Quick Hits

Shower Thoughts

The person you always see at the gym, no matter when you go, who makes you think they must live there, is probably thinking exactly the same thing about you. Source