Morning Digest, April 24, 2026

April 24, 2026 · 19 newsletters · 5 overlapping stories


Top Stories

OpenAI Launches GPT-5.5, Codenamed “Spud”

(5 newsletters)

OpenAI released GPT-5.5 to paid subscribers, pitching it as a “new class of intelligence” that tops public benchmarks across reasoning, coding, agentic tasks, and computer use. Several scores are reportedly comparable to Anthropic’s unreleased Mythos model. The model is priced at $5/$30 per million input/output tokens, framed as roughly half the cost of “competitive frontier coding models.” It launches right as Anthropic absorbs its worst week of rate-limit and quality complaints in recent memory, creating a notable sentiment shift. The Rundown called it a “vibe shift” back toward OpenAI after months of Anthropic dominance.

Microsoft and Meta Cut Thousands of Jobs as AI Reshapes Workforces

(4 newsletters)

Microsoft announced its first-ever voluntary retirement buyouts in its 51-year history, offering packages to US employees at senior director level and below whose years of service plus age total at least 70. The move affects up to 7% of the US workforce. On the same day, Meta confirmed it will lay off roughly 10% of its global workforce, around 8,000 jobs, beginning May 20. Both companies cited AI efficiency investments as the driving logic. Multiple newsletters flagged this as part of a broader pattern of tech workforce contraction tied to the AI boom.

Anthropic’s Mythos Model Was Accessed by Unauthorized Users

(3 newsletters)

A private Discord group reportedly gained access to Anthropic’s restricted Mythos cybersecurity model, which the company had deemed too dangerous for public release. The group reportedly guessed the deployment URL using naming conventions leaked in a separate data breach, and one member had contractor credentials. Anthropic says it found no evidence its systems were compromised. The same week, Anthropic published a postmortem on Claude Code quality regressions, attributing degraded responses to three separate unintentional changes that have since been resolved as of April 20.

”Tokenmaxxing” Is the New Productivity Theater

(3 newsletters)

Companies including Meta, Microsoft, and Salesforce have been running internal leaderboards ranking employees by AI token usage, with the theory that more tokens burned equals more productive work. The practice, now dubbed “tokenmaxxing,” has produced predictable results: massive token waste, system outages caused by AI overload, and throwaway work. Ramp customers saw a 13x surge in token spend since January. Meta killed its leaderboard after media backlash. Reid Hoffman suggested a middle ground: track usage, but track how people use it, not just how much.

OpenAI Launches Workspace Agents for Teams

(3 newsletters)

OpenAI introduced Workspace Agents in ChatGPT, Codex-powered shared bots that handle multi-step team workflows across Slack, Gmail, Salesforce, and other connected tools. Teams can build agents once and deploy them workspace-wide, set schedules, configure permissions, and share them with colleagues. OAI is pitching this as an evolution of the old GPT Store concept, now with enterprise governance baked in. Free through May 6 for Business, Enterprise, Edu, and Teachers plans.

Also Worth Knowing

Quick Hits

Shower Thoughts

“Procrastination may be one of humanity’s greatest survival mechanisms.” “Cold macaroni with mayonnaise sounds disgusting, but add pickle relish and suddenly it’s macaroni salad.” Via The Hustle