Morning Digest, April 10, 2026

Friday, April 10, 2026 · 17 newsletters reviewed · 5 overlapping stories identified

---

Top Stories

Anthropic Managed Agents goes to public beta

(6 newsletters)

Anthropic opened a public beta for Managed Agents, a hosted platform that lets developers launch cloud-based agents without building infrastructure. Teams define tasks, tools, and guardrails, while Anthropic handles sandboxing, orchestration, long-running sessions, and permissions. Agents can work autonomously for hours without dropping state, and a multi-agent coordination mode (in preview) lets one agent farm subtasks out to others. Pricing is $0.08 per session hour on top of normal AI usage. Early adopters include Notion, Rakuten, Asana, and Sentry. The deeper story from TLDR AI is architectural: Anthropic designed the system to decouple agent interfaces from underlying model implementations, explicitly building for “programs as yet unthought of” as models continue to improve.

Meta ships Muse Spark, first model from Superintelligence Labs

(5 newsletters)

Meta’s Superintelligence Labs, led by Alexandr Wang (who joined after the $14.3B Scale AI acquisition), released Muse Spark: a proprietary multimodal reasoning model with tool use, visual chain-of-thought, and multi-agent orchestration. Wang says the team rebuilt Meta’s AI stack from scratch. Benchmarks put it competitive with Opus 4.6 and GPT-5.4 on reasoning, though it trails on coding and ARC-AGI 2. A “Contemplating” mode pits multiple agents against each other on hard problems. The model is available on meta.ai now, with a private API preview for developers. Unlike the Llama family, Muse Spark is not open source, though Meta says it hopes to open-source future versions. The consensus across newsletters: not a leader yet, but Meta is back in the game with 3B+ daily users and the resources to build on it.

Amazon’s Jassy drops $200B AI defense in shareholder letter, takes aim at Nvidia

(4 newsletters)

Amazon CEO Andy Jassy published his annual shareholder letter with first-ever revenue figures for the company’s AI and chip divisions, firing back at investor skepticism over its $200B 2026 capex plan. AWS’s AI arm has crossed $15B in annualized revenue (260x where AWS itself stood at the same point). The Trainium, Graviton, and Nitro chip lines have crossed $20B in combined yearly revenue, and Amazon is now considering selling “racks of them to third parties” after two unnamed customers asked to buy the company’s entire Graviton supply for 2026. Trainium 3 demand is described as nearly sold out. Graviton is now used by 98% of the top 1,000 EC2 customers. The subtext across coverage: Nvidia has dominated AI compute, but Amazon’s chip numbers suggest real competition is forming on the supply side of the AI boom.

Perplexity hits $450M ARR after pivoting from search to AI agent

(3 newsletters)

Perplexity’s annualized revenue jumped 50% in a single month to $450M after pivoting its “Computer” product away from search and toward agentic workflows. The latest move: a Plaid integration that connects bank accounts, credit cards, and loans directly into the Computer agent, turning it into a personal finance hub capable of building budgets, net worth trackers, and debt payoff plans from text prompts. The company also recently added a U.S. tax feature that autonomously fills out IRS forms. Perplexity is also running a “Billion Dollar Build” challenge, offering $1M in seed funding to anyone who builds a unicorn-potential startup using Computer by April 14. Coverage frames Perplexity as now competing with Mint, TurboTax, and every other app category it integrates into, not just search.

OpenAI launches $100/month Codex Pro tier as OpenAI vs. Anthropic rivalry intensifies

(3 newsletters)

OpenAI added a new $100/month ChatGPT Pro tier offering 5x more Codex usage than the $20 Plus plan, aimed at heavy agentic coding sessions. The move comes alongside an OpenAI investor memo that reportedly characterized Anthropic as “compute-constrained,” noting OpenAI plans to have 30 gigawatts of compute by 2030 versus Anthropic’s projected 7-8 gigawatts by end of 2027. Both companies are reportedly gearing up for IPOs, possibly this year. Separately, a federal appeals court blocked Anthropic’s bid to temporarily stop the Department of War from blacklisting the company, meaning defense contractors can no longer use Claude for DoD work (though Anthropic can continue with other government agencies while the lawsuit plays out).

Other Notable Stories

• Iran ceasefire holds, but economic damage lingers. The two-week US-Iran ceasefire is holding, but the Strait of Hormuz remains disrupted with traffic at below 10% of pre-war levels. Iran is charging ships $2M to transit the strait. Oil prices are expected to stay elevated through year-end, and natural gas disruptions are spiking fertilizer prices globally. The IMF warned it will lower its 2026 global growth projection, calling some damage “irreversible” even if peace endures. US-brokered talks in Pakistan this weekend will determine whether the ceasefire extends.

• Anthropic launches Project Glasswing with Apple, Google, and AWS. Anthropic recruited Apple, Google, and AWS to a new initiative using the Claude Mythos Preview model to scan for zero-day vulnerabilities and defend against AI-driven cyberattacks. The program includes $100M in usage credits and $4M in open-source security funding.

• Google Gemma 4 crosses 500M total downloads. Google’s Gemma 4 hit 10M downloads in its first 7 days after releasing under Apache 2.0, contributing to 500M total Gemma pulls. Community benchmarks show 25-50 tokens/sec on a 16GB MacBook Air, and builders are fine-tuning text, vision, and audio variants on Apple Silicon.

• Oxford AI predicts heart failure 5 years early. University of Oxford researchers introduced an AI tool that reads invisible changes in fat texture around the heart from routine CT scans, flagging patients at 86% accuracy across 72,000 patients. In the highest-risk group, 1 in 4 developed heart failure within five years. Oxford is working with regulators to bring it to NHS hospitals and plans to extend it to all chest CTs within months.

• Maine becomes first state to ban data center construction. Maine lawmakers passed text of a bill blocking data centers from being built in the state until November 2027, the first statewide ban in the US. The measure could clear its final vote within days and may prompt other states to follow.

• Gen Z is getting angrier about AI. A new Gallup survey shows Gen Z sentiment toward AI shifted sharply in the past year: those who are excited dropped from 36% to 22%, hopeful dropped from 27% to 18%, and angry jumped from 22% to 31%. Employed Gen Zers are more than three times as likely to say AI’s risks at work outweigh the benefits.

• Disney plans to cut up to 1,000 jobs under new CEO. Disney is reportedly preparing layoffs primarily targeting the marketing department, consolidated under a new chief marketing and brand officer role in January. The cuts come as new CEO Josh D’Amaro pursues cost reduction.

• Tesla reportedly developing a smaller, cheaper SUV. After killing the Model 2, Tesla is developing a compact SUV priced substantially below the Model 3. It will measure roughly 14 feet long, use a smaller battery pack and single motor, and production is expected to be based in Shanghai, though not before 2026.

Quick Hits

• Google Colab launched “Learn Mode,” a step-by-step coding tutor powered by Gemini, along with Custom Instructions for personalizing AI behavior in notebooks.

• Cursor’s Bugbot now has a nearly 80% resolution rate after adding self-improvement via learned rules from real-world feedback; over 110,000 repos use the feature.

• HeyGen released Avatar V, claiming it eliminates “identity drift” (AI-generated faces that stop resembling the user over time) and outperforms Google Veo 3.1 on accuracy and lip sync in internal tests.

• xAI is undergoing an engineering reorg ahead of SpaceX’s IPO; Elon Musk revealed xAI has seven new models in training on Colossus 2, including 6T and 10T parameter systems.

• Perplexity’s revenue going parabolic is trending on social after a chart went viral showing the curve after the Computer and usage-based pricing shift.

• Palantir stock dropped sharply after investor Michael Burry (now-deleted post) wrote that “Anthropic is eating Palantir’s lunch.”

• US Q4 GDP was revised down again to 0.5%. US fertility rate hit a new record low, down 20% from two decades ago. American Airlines hiked bag fees due to elevated jet fuel costs from the Iran war.

• USPS is proposing raising first-class stamps to 82 cents (from 78 cents), effective July 12, citing a “severe financial crisis.”

• Claude Cowork launched in general availability with new organization controls for admin oversight.

• Florida’s attorney general opened a probe into OpenAI, issuing subpoenas over allegations that ChatGPT helped plan a campus shooting at Florida State University.

• A gene therapy trial for beta-thalassemia showed all patients going over six months without needing a transfusion by reactivating the fetal version of the hemoglobin gene.

• Bryson DeChambeau debuted a 3D-printed 5-iron he designed himself at the Masters, then triple-bogeyed the 11th hole in Round 1.