Newsletter Archive

Saturday, April 4, 2026

AI Pulse Saturday, April 4, 2026

Visit AI Pulse · Share

Half of planned US data center builds have been delayed or canceled, growth limited by shortages of power infrastructure and parts from China — the AI build-out flips the breakers

Half of planned US AI data center projects have stalled or been shelved as power infrastructure and Chinese component shortages throttle the $650B annual AI infrastructure spend. Cloud giants are hitting a hard constraint: you can't deploy GPUs without the power distribution equipment and transformers to run them.

Tom's Hardware · 1 min read Hardware

How My Agents Self-Heal in Production

LangChain engineer built a self-healing deployment pipeline that detects regressions post-deploy, automatically triages whether code changes caused them, and kicks off an AI agent to open a PR fix—all without manual intervention until review. The system uses GitHub Actions, LangSmith monitoring, and an internal coding agent to close the loop on production issues.

LangChain · 6 min read Tools

Anthropic buys biotech startup Coefficient Bio in $400M deal: reports

Anthropic acquires stealth biotech startup Coefficient Bio for $400M in stock, marking the AI safety firm's entry into AI-driven drug discovery and pharmaceutical research.

TechCrunch · 2 min read Market

Anthropic ramps up its political activities with a new PAC

Anthropic launched a new PAC to back political candidates aligned with its policy agenda ahead of the midterms, escalating the company's influence over AI regulation and governance decisions.

TechCrunch · 3 min read Policy

OpenAI’s AGI boss is taking a leave of absence

Fidji Simo, OpenAI's AGI deployment chief, is taking medical leave for several weeks; Greg Brockman will temporarily oversee product and the company's super app roadmap. CMO Kate Rouch is also stepping down.

The Verge AI · 6 min read Industry

[AINews] Good Friday

Google launched Gemma 4 under Apache 2.0 license, positioning it for reasoning, agentic workflows, and on-device inference. Day-0 ecosystem support from KerasHub and others already live; claims of outperforming 10x-larger models on efficiency benchmarks.

Latent Space · 27 min read Industry

Google DeepMind’s Research Lets an LLM Rewrite Its Own Game Theory Algorithms — And It Outperformed the Experts

Google DeepMind's AlphaEvolve uses an LLM-powered evolutionary agent to automatically discover better Multi-Agent Reinforcement Learning algorithms for imperfect-information games like poker, outperforming hand-designed baselines in both Counterfactual Regret Minimization and Policy Space Response Oracles frameworks.

MarkTechPost · 7 min read Research

The cognitive impact of coding agents

Simon Willison talks about the cognitive load and architectural debt introduced by AI coding agents in a viral podcast episode with Lenny Rachitsky that reached 1.1M views in short-form clips.

Simon Willison · 1 min read Community