AI Moves From Chat to Doing Real Work
Issue #8 From Google’s Gemma 4 to AI agents handling tasks, here’s what’s actually changing how work gets done.
It’s one of those weeks where nothing screams “huge news”…
But if you look closer, everything is quietly changing.
Less “look what AI can say.”
More: “look what AI can actually do.”
By the end of this newsletter, you’ll see it: AI isn’t just talking anymore; it’s starting to get things done.
TLDR:
Google’s Gemma 4 runs powerful AI on your phone.
Microsoft starts building its own MAI models.
OpenAI goes quiet but moves deeper into enterprise.
Meta opens free multimodal agents; Anthropic hits pause.
AI agents are starting to actually get things done.
Most B2B brands are invisible in AI search.
TOP STORIES
Start here. The 31B model is where the shift becomes obvious.
Google just made strong AI much easier to run
Google released Gemma 4, offering models from small edge versions to a 31B model with a 256K context window. TurboQuant cuts memory usage 6x without hurting performance. This combination makes powerful AI cheaper and easier to run across devices, not just in massive data centers, bringing advanced AI closer to everyday applications.
OpenAI is quietly shifting toward real-world deployment
OpenAI acquired TBPN on April 2 and followed it with an enterprise push on April 8. They also shared a child safety blueprint. No flashy model launch this time. The focus is now on where AI actually gets used: inside companies, systems, and regulated environments that need reliability and accountability.
Microsoft is starting to build more on its own
Microsoft launched MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2, its first serious in-house models outside OpenAI. They target lower error rates and faster outputs. Microsoft is still a partner, but it’s clearly building its own foundation, signaling that real-world performance and independence are becoming key priorities.
Meta and Anthropic made very different calls
Meta released a new multimodal model with free access and built-in reasoning features that coordinate multiple agents. Anthropic, on the other hand, held back Claude Mythos, citing safety concerns. Same week, opposite choices: one pushes access forward, the other slows things down deliberately, highlighting very different approaches to AI deployment and risk.
AI agents are finally starting to feel useful
Tools like C3 Code, Poke, and GLM-5.1 show real progress toward AI that actually does things. They can build apps, manage workflows, and handle multi-step tasks. It’s still early, but this week feels different: AI is moving beyond replies, from helping you think to helping you finish real work.
NEW UPDATES / FEATURES
This phase is where AI starts showing up in tools you already use.
C3 Code turns plain English into working software
C3 AI launched (C3 Code), a platform that lets companies build and deploy full applications using natural language. It handles coding, integration, and deployment through agent-style workflows. Teams can create production-ready software much faster, without needing as much traditional engineering effort, bringing AI into real-world software development.
ElevenLabs is turning into a full creative toolkit
ElevenLabs added image and video generation in beta on top of its voice tools. You can now create visuals, then layer in voice, music, and lip-synced sound effects. Instead of switching between multiple apps, you can do everything in one place, streamlining creative workflows for content creators.
PixVerse is aiming beyond short AI clips
PixVerse C1 launched with a focus on cinematic video: structured scenes, action sequences, and synced audio. It supports text, image, and storyboard inputs and outputs up to 1080p. This tool feels less like experimentation and more like something you could actually use in production pipelines.
Genspark brings AI into tools people already use
Genspark rolled out Claw and Workspace 4.0 with Microsoft Office integration. It connects documents, tasks, and automation in one flow. Instead of asking people to switch platforms, it brings AI into tools teams already know, making adoption much easier and more seamless.
TOOLS TO TRY THIS WEEK
If you want to feel this shift for yourself, start here.
EveryDev.ai: Shortcut to 1,800+ AI tools, curated by developers who actually use them. If you’re building anything, it saves hours of searching for what actually works.
Poke AI: is an agent you can message through WhatsApp or iMessage. It can plan, set reminders, and handle small tasks, no setup required. A simple taste of “agentic AI” in daily life
Runner AI: Runner AI goes further: it builds and runs an entire e-commerce store for you. AI handles the setup, ongoing optimization, and operations. You’re not just getting help; you’re handing over a system. It’s an early glimpse at how AI could manage parts of a business autonomously.
NEW DEVELOPMENTS
Zooming out, a few bigger shifts are worth noticing.
Most companies aren’t showing up in AI results
A new report found 96% of B2B brands don’t appear in AI-driven discovery tools. As more people use AI instead of search engines, the situation becomes a real problem. If AI can’t find your business, there’s a good chance your customers won’t either.
AI rules are starting to take shape
Several US states are pushing new laws around chatbot safety, AI disclosure, and protecting minors. A $224M initiative is also underway to build an AI-ready workforce. The conversation isn’t hypothetical anymore; rules are starting to define how AI gets used in practice.
The real competition is shifting to infrastructure
Intel and SambaNova introduced hybrid AI computing using GPUs, RDUs, and CPUs. Samsung and Mistral are exploring chip partnerships. As models become easier to access, the real edge may come from how efficiently you can run them at scale.
If this helped you make sense of the week in AI, share it with a friend, founder, or builder who follows the space closely.
Don’t miss out; subscribe to 60 Words of AI for structured AI news, where every story is delivered in exactly 60 words.
Before you go, drop your feedback below. It helps make each issue even better.
See you next Thursday!
Bye for now. 👋