Dixon AI Newsletter April 2026

Aug 10

12 min read

Welcome to the Dixon AI Newsletter which summarises selected AI stories and releases you might have missed. This issue covers AI news from 28th March 2026 to 24th April 2026.

The AI Macro Environment and Enterprise Insights

● Stanford releases the 2026 AI Index: The 400-page report, published on 13 April, found that generative AI is now used in at least one business function at 70% of organisations and has reached 53% population-level adoption faster than the PC or internet. US and Chinese frontier models have traded the top spot since early 2025, with the leading US model now holding just a 2.7 percentage point edge. Global corporate AI investment hit $581.7 billion in 2025, up 130% year-on-year. See (hai.stanford.edu).

● OpenAI closes the largest funding round in tech history: OpenAI raised $122 billion at a post-money valuation of $852 billion (3 April). ChatGPT now has 900 million weekly active users and 50 million paid subscribers, with revenue running at $2 billion per month and enterprise accounting for more than 40% of the total. The round signals the scale of capital now required to stay in the frontier race, with OpenAI positioning itself as core infrastructure ahead of a potential IPO. See (openai.com).

● OpenAI publishes "Industrial Policy for the Intelligence Age": The paper (7 April) lays out policy proposals tied to AI's economic impact, including a public wealth fund to distribute AI gains to citizens, incentives for four-day work weeks, automatic unemployment benefit triggers when AI-displacement metrics cross thresholds, and faster electrical grid build-out. It marks a shift in how frontier labs see their role, from model builders to active participants in shaping labour and social policy. See (openai.com).

● Trump's tech advisory council takes shape: The President's Council of Advisors on Science and Technology was assembled on 26 March with Mark Zuckerberg, Jensen Huang, Larry Ellison and Sergey Brin among the first members. The panel will advise on AI policy, economic strategy and national security. Notably, the line-up comes as Anthropic is in open litigation with the Pentagon, leaving the rest of the industry closer to the White House than at any point in recent memory. See (theverge.com).

● Anthropic secures court injunction against Pentagon designation: Anthropic won an injunction in its legal dispute with the Trump administration (late March), temporarily blocking the Defense Department's "supply chain risk" designation. The underlying dispute centres on Anthropic's refusal to allow Claude to be used for mass domestic surveillance or fully autonomous weapons. The case will likely shape how AI vendors negotiate red lines with government customers. See (wsj.com).

● Snap cuts 16% of its workforce, citing AI: Snap announced roughly 1,000 redundancies on 15 April, with CEO Evan Spiegel explicitly citing AI automation of repetitive work as the rationale. The company expects over $500 million in savings and a path to net-income profitability by the second half of 2026. It joins a growing list of public tech firms framing job cuts as an AI-driven redesign rather than a cyclical reduction. See (engadget.com).

● Meta installs keystroke and mouse-tracking software on employee machines: Reports on 21 April confirmed Meta has rolled out its Model Capability Initiative, which captures mouse movements, clicks, keystrokes and periodic screenshots from US employees to train agentic AI systems. Legal experts flagged potential GDPR issues for the European workforce, and the move raises a broader question about the line between workplace monitoring and training-data collection. See (tech.yahoo.com).

● Apple concedes the model race and bets on the ecosystem: Apple is shifting strategy away from competing directly with ChatGPT and Gemini (reported 29 March). At WWDC in June, Apple is expected to introduce iOS 27 Extensions, letting users install third-party AI chatbots that plug into Siri, and a separate deal will bring Gemini technology into Siri features. Apple is effectively positioning itself as the distribution platform for AI rather than a foundation-model builder, with a 30% App Store cut of the arrangement. See (bloomberg.com).

● OpenAI to spend $20B+ on Cerebras chips: OpenAI has agreed to spend more than $20 billion over three years on Cerebras-powered servers, with total spend potentially reaching $30 billion and warrants giving OpenAI up to a 10% equity stake in Cerebras (17 April). OpenAI will also contribute around $1 billion to fund supporting data centres. The deal is the clearest signal yet that hyperscalers are actively diversifying away from a single-supplier dependency on Nvidia. See (theinformation.com).

● Anthropic's Economic Index Survey goes monthly, and finds a widening skills gap: Anthropic launched a rolling monthly survey on 22 April and published fresh data showing that experienced Claude users are materially more successful at automating tasks than newcomers, across all model tiers, languages and regions. Customer-service, business-sales and market-research workflows saw API automation double between November 2025 and February 2026, suggesting the next wave of displacement will hit commercial and support roles. See (anthropic.com).

● Vibe coding drives an 84% surge in App Store submissions: The Information reported (early April) that new app publications on Apple's App Store jumped 84% year-on-year in Q1 2026, reversing a long decline. Replit users alone published nearly 5,000 apps in recent months. The signal matters less for Apple and more as a proxy for how rapidly software production is being democratised, with quality and discovery concerns already following close behind. See (theinformation.com).

● Anthropic accidentally leaks 500,000 lines of its own source code: A debugging file bundled into a routine Claude Code update (reported 31 March) exposed nearly 2,000 files and 500,000 lines of the product's architecture on a public registry. Anthropic confirmed the incident was a packaging error rather than a breach, and no customer data was exposed. Coming from the lab that markets itself on safety posture, the incident is a reminder that operational discipline is now as material as model capability when enterprises evaluate AI vendors. See (theguardian.com).

New and updated AI model and application releases

OpenAI

● GPT-5.5: Most capable model yet for complex agentic coding, research and knowledge work. Rolling out to Plus, Pro, Business and Enterprise. Introducing GPT-5.5 | OpenAI

● ChatGPT Images 2.0: Major upgrade to image generation with improved text rendering, stronger visual reasoning and expanded multilingual script support. Introducing ChatGPT Images 2.0 | OpenAI

● Workspace Agents in ChatGPT: Cloud-based team agents for long-running workflows, integrated with Slack. Available in research preview for Business, Enterprise, Edu and Teachers plans. Introducing workspace agents in ChatGPT | OpenAI

● Codex desktop app update: Background computer use, in-app browser, image generation via gpt-image-1.5, memory, scheduled tasks and over 90 new plugins including Atlassian Rovo, GitLab and Microsoft Suite. Codex for almost everything | OpenAI

● Privacy Filter: Open-weight 1.5B-parameter model for detecting and redacting eight categories of PII locally, Apache 2.0 on Hugging Face. Introducing OpenAI Privacy Filter | OpenAI

● Agents SDK update: Model-native harness, native sandbox execution, configurable memory and MCP integration. GA in Python. The next evolution of the Agents SDK | OpenAI

● GPT-Rosalind: Frontier reasoning model purpose-built for life sciences, drug discovery and translational medicine. Partners include Amgen, Moderna and the Allen Institute. Introducing GPT-Rosalind | OpenAI

● ChatGPT for Clinicians: Free tool for verified US physicians, NPs, PAs and pharmacists with documentation, clinical search and optional HIPAA compliance. Physician advisors rated 99.6% of responses safe and accurate. Making ChatGPT better for clinicians | OpenAI

Google

● Skills in Chrome: Save reusable Gemini prompts and trigger them with a forward slash on any page. Skills in Chrome

● Gemma 4: New open models designed for reasoning tasks and autonomous agent workflows. Gemma 4

● Notebooks in the Gemini app: NotebookLM functionality has been integrated directly into the Gemini app, letting users manage chats and research projects in one place. Not currently available in Workspace accounts. Notebooks in Gemini

● Lyria 3 Pro music generation: Free creation of high-fidelity tracks up to three minutes long, with the ability to mix and customise the output. Lyria 3 Pro in Gemini

● 3D models and charts in Gemini: New capability to turn complex concepts into interactive visuals directly inside the chat. Visualise complex concepts with Gemini

● Gemini Deep Research and Deep Research Max: Two autonomous research agents built on Gemini 3.1 Pro, with MCP support, native chart and infographic generation, and partners including FactSet, S&P Global and PitchBook. Next-generation Gemini Deep Research

● Stitch DESIGN.md: Open-sourced design-rules specification, allowing design systems to be exported and imported across platforms. Stitch DESIGN.md

Microsoft

● 365 Copilot agentic features: Multi-step, app-native actions generally available in Word, Excel and PowerPoint. Copilot's agentic capabilities in Word, Excel and PowerPoint are generally available | Microsoft Community Hub

● Critique and Council: Critique is a multi-model deep research system that pairs a lead model with a reviewer; Council runs Anthropic and OpenAI models side-by-side with a judge model summarising agreement and divergence. Microsoft unveils AI upgrades | Microsoft Community Hub

Anthropic

● Claude Opus 4.7: GA across Claude products, the API, Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry. Notable improvements in advanced software engineering, higher-resolution vision and creative output. Pricing unchanged at $5/$25 per million tokens. Claude Opus 4.7 \ Anthropic

● Claude Design: Anthropic Labs product for collaborative visual work: slides, prototypes, pitch decks, marketing collateral. Exports to Canva, PDF, PPTX, HTML. Research preview for Pro, Max, Team and Enterprise. Claude Design, Anthropic Labs \ Anthropic

● Claude for Word: Microsoft Word integration for Pro and Max subscribers, bringing Claude's writing and editing into Word documents. Claude for Word \ Anthropic

● Claude Code Routines: Research preview of repeatable automations that run on a schedule, via API call, or in response to GitHub events. Introducing Routines in Claude Code \ Anthropic

● Claude Code desktop redesign: New sidebar for parallel agent sessions, integrated terminal, in-app file editor, faster diff viewer and expanded SSH support on Mac. Claude Code desktop redesign \ Anthropic

● Project Glasswing and Claude Mythos system card: Initiative with AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft and Nvidia to find and fix vulnerabilities in critical software using an unreleased Claude Mythos model. Project Glasswing \ Anthropic

● Live Artifacts in Cowork: Interactive dashboards and trackers that auto-refresh with current data from connected apps and files. Live Artifacts in Cowork \ Anthropic

● Everyday connectors: Directory expanded with AllTrails, Audible, Booking.com, Instacart, Intuit TurboTax and Credit Karma, Resy, Spotify, StubHub, Taskrabbit, Thumbtack, TripAdvisor, Uber, Uber Eats and Viator. Connectors for everyday life \ Anthropic

Notable Other Model Releases

● DeepSeek V4 (Pro and Flash): Preview release of the long-awaited V4 series, both open-source under MIT with 1M context. V4-Pro is a 1.6T total / 49B active parameter MoE, V4-Flash is 284B / 13B. Both support dual Thinking and Non-Thinking modes. Notably trained and served on Huawei Ascend and Cambricon chips rather than Nvidia, positioning V4-Pro as the cheapest of the larger frontier models. DeepSeek V4 Preview

● Meta Muse Spark: Meta Superintelligence Labs' first flagship model since the restructuring, purpose-built for Meta's product ecosystem, with text and image inputs and Instant/Thinking modes. Meta Muse Spark launch

● Kimi K2.6 (Moonshot): Open-source coding model with state-of-the-art results on and 4,000+ step long-horizon capability. Kimi K2.6 release

Notable Application Releases

● Canva AI 2.0: Biggest platform overhaul since moving design to the browser, introducing prompt-based editing, an orchestration layer across tools, persistent memory for brand consistency and HTML import. Canva AI 2.0

● Perplexity Personal Computer for Mac: Multi-model AI orchestration across local files, native apps, connectors and the web, integrated with Notes, iMessage and email. Personal Computer is here

● Opera Browser Connector: Free MCP-powered connector that gives ChatGPT or Claude access to open tabs, screenshots and optional browsing history inside Opera One and Opera GX. Opera Browser Connector

● Adobe Firefly AI Assistant + Claude: Creative agent coming to the Firefly app, plus a partnership bringing Firefly capabilities into Claude, alongside expanded model support including Kling 3.0, Veo 3.1 and Runway Gen-4.5. Adobe Firefly AI Assistant

From Dixon AI

AI Drop In: one-hour, live sessions held twice a week.
- Join here: https://www.dixonai.com/ai-drop-in
  - w/c 27th Apr: What’s Happened in AI this month
  - w/c 4th May: Copilot Cowork
  - w/c 11th May: ChatGPT Agents
AI Courses (full details on our website):
- Our next four-week Copilot Masterclass starts on Friday 8th May
- Our next three-week AI Literacy Essentials course starts Thursday 7th May
- Our new four-week Claude Masterclass starts on Wednesday 6th May
- For more information about our online courses, or to sign up to a course, please visit https://www.dixonai.com/training

A word from Rob

The capability overhang closes where people build

I have spent the past month back in rooms with organisations I have worked with before. Vistage groups, CEO cohorts, client teams I first met twelve or eighteen months ago. The pattern across those revisits has been hard to miss, and it is the reason for this month's note.

The organisations that heard the Dixon AI message early, and acted on it, are notably further down the road. They are deploying agents, experimenting with OpenClaw, building internal systems, and moving meaningfully through the seven stages of organisational AI transformation (Dixon, 2025). The organisations that have not had that kind of intervention are mostly still where they were, talking about AI rather than building with it.

The Dixon AI approach is working. And the place where it works is the place where people pick up laptops and start making things.

Training is education. It is not transformation.

The most useful thing I can say to anyone planning their next intervention is this. Training, on its own, is not enough. Talking about AI projects is not enough. Slide decks about AI strategy are not enough.

What works is purpose-led, hands-on building. People upskilling with a clear view to making something specific. Iterating in the room. Solving a real problem with the technology in front of them. Literacy is acquired through the act of building, not before it. This is the principle underneath the GEFIE Programme, and it is the reason it produces movement where conventional training does not.

The shift I want every leader reading this to make is small in language and large in consequence:

Stop scheduling meetings about doing things with AI. Start scheduling meetings of people doing things with AI.

Get the laptops out. Bring the stakeholders who can make the decisions. Build collaboratively, in the session, on a real use case. An hour spent making something together is worth a day spent discussing whether to.

The IT department is where AI goes to die.

Here is the second observation from this past month, and it is the one I am going to keep saying until it sticks. If the doing of AI is concentrated in IT, the transformation stalls. I have seen it happen often enough now to call it a pattern.

It is not a comment on technology teams. It is a comment on where the centre of gravity for value creation sits. AI value is created on the value chain, in the commercial functions, where customers, products and revenue live. Organisations that have kept the drive out there, with senior leaders and commercial people in the driving seat, are seeing results. Organisations that have quietly outsourced the entire question to IT are getting stuck.

One observation, twice.

The two points are the same point. Building beats talking. Commercial beats IT. Both come back to a single principle: AI capability is built where value is created, by the people who create it, in the act of creating it. That is how the capability overhang closes (Dixon, 2024).

So this month's challenge for senior leaders is straightforward. Stop talking about AI. Stop analysing it into committees. Stop waiting for IT to hand you a strategy. Pick a real commercial problem, get the right people in a room with their laptops, and build something. Then build the next thing.

If this resonates, or if you would like to talk about how a purpose-led intervention could work in your organisation, you can reach me at rob@dixonai.com or visit dixonai.com.

Until next month.

Rob.

Robert J. Dixon is Founder and CEO of Dixon AI Ltd, an AI consultancy and training provider based in Manchester, and a Lecturer at Manchester Metropolitan University Business School. He is the author of The AI Transformation Playbook (Dixon AI Press, 2025).

References

Dixon, R., 2024. Defining AI Capability: A Holistic Framework for Organisational Readiness in an AI-Driven World. [online] LinkedIn. Available at: https://www.linkedin.com/pulse/defining-ai-capability-holistic-framework-readiness-ai-driven-dixon-omrle [Accessed 27 April 2026].

Dixon, R., 2025. The AI Transformation Playbook: A Step-by-Step Guide to Becoming an AI-Enabled Organisation. Dixon AI Press. Available at: https://www.amazon.co.uk/Transformation-Playbook-Step-Step-Organisation/dp/1919400508