Apr 18, 2026
Industry
Allbirds shell pivots to NewBird AI — plans $50M raise for GPU-as-a-service infrastructure
The Verge reports that the Allbirds corporate shell is being repositioned as NewBird AI, an AI and GPU infrastructure company. The plan involves raising $50 million to acquire GPU assets for a GPU-as-a-service offering. Shares surged as high as $24.31 from a $6.82 open following the announcement.
Why it matters: A footwear brand pivoting to GPU infrastructure captures the speculative frenzy around AI compute — and raises questions about corporate shell manoeuvres in the AI boom.
Read more on The Verge →
Apr 17, 2026
Product
Anthropic announces Claude Opus 4.7 — stronger agentic and coding performance
Anthropic released Claude Opus 4.7, claiming improved multi-step reliability, fewer tool errors, stronger code review performance, and notable gains over Opus 4.6 in production-style usage. The launch post draws on partner evaluations to position the model for complex agentic and coding workflows.
Why it matters: Incremental model improvements in reliability and tool use directly affect enterprise adoption of AI-driven automation pipelines.
Read more on Anthropic →
Apr 17, 2026
Product
Google introduces Gemini 3.1 Flash TTS — text-to-speech with finer control and 70+ languages
Google launched Gemini 3.1 Flash TTS, a new text-to-speech model focused on controllability, expressiveness, and speech quality. It supports over 70 languages, uses SynthID watermarking on generated audio, and is rolling out in preview through the Gemini API, Google AI Studio, Vertex AI, and Google Vids.
Why it matters: Better TTS with watermarking addresses both the quality demands and the provenance concerns shaping how synthetic speech gets deployed at scale.
Read more on Google Blog →
Apr 17, 2026
Product
Google pushes personalised image generation — Personal Intelligence meets Nano Banana 2 and Google Photos
Google is combining Personal Intelligence, Nano Banana 2, and Google Photos to enable more personalised image generation in Gemini. The system uses connected account context and photo libraries to produce tailored images from shorter prompts, whilst Google states that private Photos are not used directly for model training.
Why it matters: Personalised generation blurs the line between AI tools and personal creative assistants — raising both convenience and privacy questions about how user data feeds creative output.
Read more on Google Blog →
Apr 17, 2026
Product
Microsoft launches MAI-Image-2-Efficient — 41% cheaper image generation at flagship quality
Microsoft released MAI-Image-2-Efficient, a smaller and faster version of its flagship image model. Available now in Foundry and MAI Playground, it runs 22% faster, is 4x more efficient, and is priced approximately 41% lower whilst maintaining flagship-quality image generation.
Why it matters: Driving down image generation costs makes AI-generated visuals viable for higher-volume use cases like e-commerce, advertising, and automated content pipelines.
Read more on Microsoft →
Apr 17, 2026
Product
Midjourney V8.1 goes live — 3x faster, 3x cheaper, with native 2K rendering
Midjourney released V8.1, pitching it as a major speed-and-cost improvement over V8. The update restores what the company calls its iconic aesthetic style, adds native 2K HD rendering, and delivers 3x faster generation at 3x lower cost per image.
Why it matters: Significant cost and speed improvements in premium image generation intensify competition across the visual AI market and expand professional adoption.
Read more on Midjourney →
Apr 17, 2026
Research
MiniMax open-sources M2.7 — high-end open model with 56% SWE-Pro score
MiniMax released M2.7 as open source, highlighting performance of 56.22% on SWE-Pro and 57.0% on Terminal Bench 2. The company is positioning the release as a competitive open model targeting advanced software engineering and terminal-based workflows.
Why it matters: High-performing open models in the coding domain keep pressure on proprietary providers and expand the options available for enterprise self-hosting.
Read more on MiniMax →
Apr 17, 2026
Research
Alibaba open-sources Qwen3.6-35B-A3B — sparse MoE with 35B params, 3B active, Apache 2.0
Alibaba's Qwen team released Qwen3.6-35B-A3B as open source under Apache 2.0. The model uses a sparse mixture-of-experts architecture with 35 billion total parameters but only 3 billion active at inference, targeting agentic coding and multimodal capabilities at a fraction of the compute cost of dense equivalents.
Why it matters: Efficient MoE architectures under permissive licences lower the hardware barrier for deploying capable models, accelerating adoption outside the biggest labs.
Read more on Qwen →
Apr 16, 2026
Product
OpenAI expands Codex beyond code — desktop agent with computer use, browser, and 90+ plugins
OpenAI announced that Codex is expanding well beyond code completion into a broader desktop agent for software work. The update adds computer use on macOS, an in-app browser, image generation, memory, automations, SSH access, richer file previews, and more than 90 new plugins, with rollout beginning 16 April for signed-in ChatGPT users.
Why it matters: Codex evolving from code assistant to full desktop agent signals OpenAI's ambition to own the developer workflow end-to-end — and compete with Claude Code, Cursor, and similar tools.
Read more on OpenAI →
Apr 16, 2026
Product
OpenAI introduces GPT-Rosalind — purpose-built model family for life sciences research
OpenAI launched GPT-Rosalind, a model family designed specifically for biology, drug discovery, and translational medicine. It is available in research preview for qualified customers via trusted access, and comes with a Life Sciences research plugin for Codex connecting to more than 50 scientific tools and data sources.
Why it matters: Domain-specific frontier models for life sciences represent a new commercial vertical for AI labs — and could accelerate drug discovery timelines if the research community adopts them.
Read more on OpenAI →
Apr 16, 2026
Product
OpenAI updates Agents SDK — file access, command execution, and longer multi-step workflows
OpenAI released an update to its Agents SDK, expanding agent capabilities to include file access, command execution, code editing, and longer multi-step workflows in controlled environments. The update reflects a shift from chatbot-style interactions toward fully operational software agents.
Why it matters: Richer SDK primitives for file and command access push the agent ecosystem closer to production-grade automation rather than conversational prototyping.
Read more on OpenAI →
Apr 16, 2026
Infrastructure
OpenAI expands Trusted Access for cyber defence — segmenting frontier AI by risk level
OpenAI broadened its Trusted Access programme for cybersecurity, providing vetted organisations access to more powerful models — including cyber-focused variants — for vulnerability detection and defence use cases. The expansion signals tighter segmentation of frontier AI capabilities by risk level and user trust.
Why it matters: Tiered access for sensitive domains sets a precedent for how frontier capabilities get distributed — balancing security benefits against the risk of dual-use misuse.
Read more on OpenAI →
Apr 16, 2026
Product
Canva unveils AI 2.0 — prompt-based design and editing across the platform
Canva announced an AI 2.0 refresh centred on prompt-based design and editing. The Verge reports that Canva is positioning itself as a unified AI content-creation hub, with a conversational interface that can orchestrate multiple Canva tools from a single prompt.
Why it matters: Canva shifting to prompt-first design lowers the barrier further for non-designers and puts pressure on traditional creative tool suites to match the conversational UX.
Read more on The Verge →
Apr 15, 2026
Product
Perplexity launches Personal Computer — shifting AI work from cloud to local machine
Perplexity announced its Personal Computer product, presenting it as a move from cloud-style orchestration toward work that runs on the user's own machine across files, apps, and the web. The product is framed as a locally grounded AI assistant rather than a cloud-only chatbot.
Why it matters: Local-first AI products address growing user concerns about data privacy and latency, and open competition for who controls the desktop AI experience.
Read more on Perplexity →
Apr 15, 2026
Product
Google brings Gemini app to macOS — native desktop app with screen sharing
Google released a native Gemini app for macOS, available free on macOS 15 and above. The app uses Option + Space as a keyboard shortcut and supports screen sharing, allowing users to ask Gemini about what is on their screen or in local files.
Why it matters: Google bringing Gemini to the Mac desktop intensifies the race for the default AI assistant on personal computers — directly competing with Apple Intelligence and Claude desktop.
Read more on Google Blog →
Apr 15, 2026
Product
Google launches the Google app for desktop on Windows — AI Mode, local files, and Alt + Space access
Google launched its desktop app globally in English on Windows, bringing Search to the desktop with AI Mode built in. The app supports Alt + Space access to web results, local files, installed apps, and Google Drive files, plus screen sharing for contextual questions.
Why it matters: A Google desktop app with file and app access positions Google Search as an operating-system-level AI layer — competing with Microsoft Copilot on its own platform.
Read more on Google Blog →
Apr 15, 2026
Product
Google adds Skills in Chrome — reusable one-click AI workflows inside the browser
Google announced Skills in Chrome, a feature that turns prompts and workflows into reusable one-click tools inside the browser. Users can discover, save, remix, and instantly repeat AI workflows instead of rewriting the same prompts each time.
Why it matters: Reusable AI workflows embedded in the browser lower the barrier from "prompt engineering" to "tool use" — making AI automation accessible to non-technical users at the browser level.
Read more on Google Blog →
Apr 15, 2026
Industry
Boston Dynamics pairs Spot with Gemini Robotics-ER 1.5 for household cleanup tasks
Boston Dynamics demonstrated how its Spot robot can be paired with Google's Gemini Robotics-ER 1.5 for household-style cleanup tasks using natural-language instructions. The company emphasised that Gemini works through Spot's existing SDK and API, keeping behaviour bounded whilst letting the model sequence tasks more flexibly.
Why it matters: Foundation models layered on top of existing robot APIs demonstrate a practical path to making robots more useful without requiring full autonomy or custom hardware changes.
Read more on Boston Dynamics →
Apr 14, 2026
Product
Anthropic redesigns Claude Code desktop around parallel-agent work
Anthropic announced a redesign of Claude Code on desktop, framing the release around the ability to run more Claude Code tasks at once. The update is positioned as a shift toward parallel-agent workflows, allowing developers to manage multiple concurrent coding sessions from a single interface.
Why it matters: Parallel agent execution in developer tools reflects a broader trend toward concurrency in AI-assisted work — multiplying throughput rather than just improving individual task quality.
Read more on Anthropic →
Apr 14, 2026
Product
Anthropic launches Claude Design — AI-generated presentations, prototypes, and marketing assets
Anthropic launched Claude Design, an AI tool for generating presentations, prototypes, marketing assets, and visual documents directly from prompts. Part of Anthropic's Labs push, it is positioned as a no-design-skills-required product for full creative workflows, competing with tools like Canva and Figma.
Why it matters: An AI lab entering the design-tool market signals that frontier model capabilities are now sufficient to challenge established creative platforms on their core use cases.
Read more on Anthropic →
Apr 14, 2026
Product
Blackmagic Design announces DaVinci Resolve 21 — adds Photo page for still-image colour grading
Blackmagic Design announced DaVinci Resolve 21, a major update that adds a new Photo page bringing DaVinci's professional colour tools to still photographs. The release extends Resolve's scope beyond video editing into photo post-production workflows.
Why it matters: A video editing suite expanding into photo colour grading could attract photographers who want unified colour tools across stills and motion — challenging Lightroom's dominance in that niche.
Read more on Blackmagic Design →
Many of this week's stories touch on agent protocols, MCP, and AI search shifts. The free AI SEO Foundations course walks through what to do about it on your own site — JSON-LD schema, llms.txt, AGENTS.md, the agent-readability frameworks, and the five-step audit. Fifteen lessons, two to three hours, with a final exam and certificate.
Open the course →