Current State of AI April 2026: Agents Ship, Sora Dies, and Anthropic Takes the Lead

March 2026 was the most consequential month in AI since the original ChatGPT launch. One company shipped an open-source competitor killer in under four weeks. Another killed its flagship creative product to save its IPO. And the enterprise world quietly crossed the line from "experimenting with AI" to "running production agents."

Anthropic Is Taking the World by Storm

There's no other way to say it: Anthropic had the most impressive Q1 of any AI company in 2026. While others were trimming products and preparing for IPOs, Anthropic was shipping. (If you missed the full rundown of everything they released in March, check out my breakdown of Anthropic's madcap March.)

The headline move was Claude Code Channels — Anthropic's direct response to OpenClaw, the viral open-source autonomous agent that had lines around the block in Shenzhen. OpenClaw was gaining momentum fast: an open-source agent that could autonomously write code, manage projects, and integrate with anything through a messaging interface. The AI community was calling it a paradigm shift.

Anthropic's response? They shipped a better version in four weeks. Claude Code Channels lets developers message Claude Code directly through Telegram and Discord, turning the interaction model from synchronous "ask-and-wait" into asynchronous, autonomous partnership. But unlike OpenClaw, it comes with enterprise-grade security, admin oversight, and the full weight of Anthropic's Model Context Protocol (MCP) — the open standard they introduced in 2024 that's become the universal connector for AI tools.

The community reaction was immediate. "Claude just killed OpenClaw with this update" became the consensus take across developer forums. The speed of shipping — incorporating messaging integration, thousands of MCP skills, and autonomous bug-fixing in just one month — caught the entire industry off guard.

But Claude Code Channels wasn't even Anthropic's biggest move this quarter.

Cowork: The Agent That Spooked Wall Street

Anthropic's Cowork desktop agent, which launched in January and expanded to enterprise plans by late January, became the product that made the entire SaaS industry nervous. Cowork is a general-purpose AI agent that sits on your desktop and automates knowledge work — file management, document processing, email, scheduling, research, and complex multi-step workflows.

I've been using Cowork for all my business-side activities, and it's truly incredible. This isn't a chatbot with a file browser bolted on. It's an agent that connects to Google Drive, Gmail, DocuSign, FactSet, and dozens of other enterprise tools through plugins. You describe what you need done, and it handles the entire workflow autonomously.

The market took notice immediately. On February 3, Anthropic's release of 11 open-source Cowork plugins — including a legal workflow tool for contract review, NDA triage, and compliance — triggered a $285 billion selloff across software stocks. A Goldman Sachs basket of US software stocks sank 6% in a single day, its biggest decline since April 2025. Legal tech giants Thomson Reuters and RELX saw double-digit drops. The fear wasn't that AI is a productivity tool — it was that AI is a direct substitute for entire categories of the software and services value chain.

Then Microsoft validated the whole thesis. On March 9, they announced Copilot Cowork — powered by Anthropic's Claude — to handle long-running agentic workflows inside Microsoft 365. This wasn't a small partnership. It's built on a $30 billion Azure compute deal signed in November 2025. Despite a $13 billion investment in OpenAI, Microsoft built its newest flagship M365 feature on Anthropic's technology. When Microsoft bets that kind of money on your stack, enterprise buyers notice.

Anthropic's numbers reflect the momentum. Enterprise revenue accounts for roughly 80% of their business, with $14 billion in annualized run rate as of February 2026 — over 10x annual growth for three consecutive years. Eight of the Fortune 10 are Claude customers. And they just closed a $30 billion Series G at a $380 billion valuation. They're not just building models — they're building the infrastructure layer that enterprises are adopting out of the box for agentic functionality.

OpenAI's Strategic Pivot: Sora Dies, Codex Lives

On March 24, OpenAI made one of the most significant product decisions in its history: they killed Sora.

The numbers tell the story. Forbes estimated Sora's peak inference costs at approximately $15 million per day, based on analysis by Cantor Fitzgerald analyst Deepak Mathivanan. Total lifetime in-app revenue? $2.1 million. That gap was never going to close.

After a flashy launch, Sora's worldwide user count peaked around one million before collapsing to fewer than 500,000. Downloads dropped 66% between November 2025 and February 2026. Impressive AI technology alone doesn't sustain user engagement — especially when the output requires expensive compute for every generation.

The Sora web and app experiences will be fully discontinued by April 26, with the API following on September 24.

But here's the part most people miss: Sora dying isn't a failure story. It's a strategic pivot story.

The OpenAI Super App Strategy

OpenAI isn't shrinking — they're consolidating. By killing Sora and redirecting that compute, they're betting everything on three products that actually make money:

Codex is the winner of this reallocation. OpenAI's refreshed coding agent now serves more than 2 million weekly users, up fivefold over the past three months. That's real enterprise traction.

The Super App is coming. On March 19, OpenAI's CEO of Applications Fidji Simo outlined a plan to merge ChatGPT, Codex, and the Atlas browser into a unified desktop application. The strategy: first equip Codex with agent-based features beyond just coding (data analysis, general productivity), then fold in ChatGPT and Atlas. The mobile ChatGPT app stays standalone.

The $122 billion war chest funds all of it. OpenAI closed the largest funding round in Silicon Valley history at a post-money valuation of $852 billion. The round was anchored by Amazon, Nvidia, and SoftBank, with continued participation from Microsoft. SoftBank co-led alongside Andreessen Horowitz, D.E. Shaw Ventures, MGX, and TPG.

This is IPO preparation. By shutting down the cash-burning creative product and doubling down on enterprise-revenue products like Codex and ChatGPT, OpenAI is cleaning up its portfolio for public market scrutiny. The message to investors: we know which products make money, and we're not sentimental about the rest.

The Enterprise AI Adoption Tipping Point

Zoom out from the Anthropic-OpenAI horse race and the macro picture is just as dramatic. Enterprise AI adoption has crossed from experimentation into deployment at scale.

The numbers from Q1 2026 paint a clear picture. 86% of enterprise respondents said their AI budget will increase this year. Total worldwide AI spending is projected to surpass $2 trillion in 2026, up from $1.5 trillion in 2025. The agentic AI market specifically is expected to hit $10.86 billion this year, growing at over 45% CAGR.

But the adoption gap is real. Almost four in five enterprises have adopted AI agents in some form, yet only one in nine runs them in production. That's the opportunity and the problem at the same time. By the end of 2026, Gartner projects 40% of enterprise applications will include task-specific AI agents.

The industries moving fastest are telecommunications (48% agentic AI adoption) and retail/CPG (47%). The biggest barrier isn't technology — it's the AI skills gap, with education identified as the number one way companies are adjusting their talent strategies.

What This Means for You Right Now

If you're building a business, running automations, or just trying to stay ahead of the curve, here's what the April 2026 landscape actually means in practice.

The agentic era isn't coming — it arrived. Cowork, Codex, and Claude Code Channels aren't demos. They're production tools with millions of users. If you're still treating AI as a chatbot you paste questions into, you're two generations behind.

The winner of the AI race right now is Anthropic. Not because their models are definitively better on every benchmark, but because they're shipping the infrastructure — MCP, Cowork, Claude Code Channels, enterprise plugins — that makes agentic AI actually usable in real businesses. They shipped a credible OpenClaw competitor in a month. That's execution speed that matters.

OpenAI isn't losing — they're repositioning. The Sora shutdown and Super App consolidation are smart moves. Codex at 2 million weekly users proves they have enterprise traction. But the IPO pressure means every product decision now gets filtered through "does this make the financials look better for public markets?"

The skills gap is your moat. With only one in nine enterprises running AI agents in production despite 80% having adopted them in some form, the people who actually know how to deploy and manage these systems are in massive demand. That's the opportunity. (For a deeper look at how agents are reshaping work, read my analysis on the rise of AI agents in 2026.)