2026-04-09
By Vadym · Generated with AI, curated by me
Google shipped Gemma 4 under a fully open license — multimodal, runs on a single GPU, Apache 2.0. Microsoft released three in-house AI models that bypass OpenAI entirely. Cursor rebuilt its IDE around agents. The distributed layer of AI is closing the gap with the closed frontier faster than most expected.
Google released Gemma 4, its most capable open-weight model family to date, under an Apache 2.0 license. The family comes in four sizes — E2B, E4B, 26B MoE, and 31B Dense — all supporting image and text inputs, with the smaller variants adding audio. Every model supports a 256,000-token context window and runs on a single 80GB NVIDIA H100 GPU. The 31B model ranks #3 on the Arena AI text leaderboard among open models; the 26B MoE ranks #6. Weights are available immediately on HuggingFace, Kaggle, and Ollama.
Microsoft released MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — three proprietary foundational models built internally and deployed on Microsoft Foundry. MAI-Transcribe-1 handles speech-to-text across 25 languages at 2.5x the speed of Azure Fast. MAI-Voice-1 generates 60 seconds of audio in one second and allows custom voice cloning from short samples. MAI-Image-2 ranks top 3 on Arena.ai’s image leaderboard and delivers at least 2x faster generation in Copilot. All three are immediately available to developers via Microsoft Foundry and MAI Playground.
Cursor launched Cursor 3, a complete rebuild of its interface around parallel AI agents. The new Agents Window lets developers run multiple agents simultaneously across local, cloud, git worktree, and remote SSH environments — all from a single command hub. A Design Mode lets developers annotate UI elements directly in the browser and hand them off to agents for targeted changes. Agent Tabs show multiple agent conversations side by side. The stated design premise: most code will be written by agents, and the developer’s job is to orchestrate them.
Eclipse Ventures closed a $1.3B fund split between Fund VI ($720M early-stage) and Early Growth Fund III ($591M growth-stage), bringing its total AUM to roughly $10 billion. The focus: physical AI — robotics, autonomous systems, manufacturing, transportation, energy, defense, and compute infrastructure. Unusually, Eclipse plans to use a portion of the capital to incubate and build startups from scratch, rather than just writing checks. Previous backers include companies like Cerebras.
The tech industry laid off 78,557 workers in Q1 2026, with 47.9% of those cuts — roughly 37,600 positions — attributed to AI-driven efficiency gains and reduced need for human workers. Companies including Block, Atlassian, and Dell explicitly cited AI automation as a contributing factor. Dell’s headcount dropped from 108,000 to 97,000, a reduction of roughly 11,000 jobs. The pace is accelerating: AI-attributed job cuts jumped 25% from February to March alone.
Google released “AI Edge Eloquent” on iOS with no press release — a free dictation app powered by on-device Gemma models. Once downloaded, it transcribes live speech, automatically removes filler words on pause, and offers text reformatting options like “Formal,” “Short,” and “Key Points.” Users can toggle off cloud mode entirely for local-only processing. It can optionally import contact names and jargon from Gmail for accuracy. Android support is not yet available; Google shipped iOS-first, an unusual reversal.
The pattern across today’s stories is decentralization. Google put a frontier-grade multimodal model on a single GPU under a fully open license. Microsoft built its own voice, transcription, and image models — no OpenAI required. Cursor bet its entire product on the idea that developers orchestrate agents rather than write code. And a $1.3B fund is building physical AI companies from scratch because it doesn’t think enough founders exist yet. Meanwhile, 80,000 workers were laid off in Q1, nearly half by AI attribution. The concentrated, closed-frontier story from yesterday is real — but so is this one. Both layers are moving fast. The question is which one compounds harder.
— Boba
Curated by Vadym