2026-04-23
By Vadym · Generated with AI, curated by me
Google Cloud Next 2026 landed in Las Vegas on April 22 with a clear signal: the agentic era is no longer roadmap. Meanwhile, Anthropic’s revenue hit a $30B run rate—triple the year-end 2025 number—and a new industry survey is putting hard numbers on the security debt that AI-generated code is quietly accumulating.
At Google Cloud Next in Las Vegas on April 22, Google announced that its Agent2Agent protocol has 150 organizations running it in production—not pilot. A2A is now at version 1.2, with cryptographic signing for agent identity verification, and governance has moved to the Linux Foundation’s Agentic AI Foundation. Microsoft, AWS, Salesforce, SAP, and ServiceNow are all running it. Alongside the protocol news: fully managed MCP servers for Google Maps, BigQuery, Compute Engine, and Kubernetes went live, covering tool access and agent communication as a unified stack.Boba’s take: A2A moving to Linux Foundation is the governance milestone that actually matters. It signals this is no longer Google’s proprietary protocol—it’s becoming neutral infrastructure, the way HTTP and OAuth did. When Salesforce, Microsoft, and AWS are all running a protocol that Google invented, it’s time to treat it like a standard. The MCP server additions close the loop: tool access plus agent-to-agent communication is the full agentic stack.
Also announced at Google Cloud Next on April 22: Deep Research and Deep Research Max, two new research agents powered by Gemini 3.1 Pro. Both support MCP and can pull from public web data and a company’s internal systems through a single API call. Both natively generate charts and infographics, not just text reports. Deep Research optimizes for speed and cost; Deep Research Max optimizes for comprehensiveness at higher compute cost. Available in Gemini API public preview now.Boba’s take: The meaningful part isn’t the web search capability—that’s table stakes. It’s the enterprise data integration: a research agent that can pull from internal Confluence, BigQuery, or Salesforce alongside the open web changes what “research” means inside a company. That’s closer to a knowledge worker’s actual job than any chatbot has managed. MCP support means this plugs into the broader tool ecosystem rather than being a Google silo.
Anthropic disclosed that its annualized revenue run rate has surpassed $30 billion—up from $9 billion at the end of 2025, roughly a 3x increase in four months. More than 1,000 customers now spend over $1 million per year on Claude, a figure that more than doubled since February. Eight of the Fortune 10 are now Claude customers. Claude Code alone runs at $2.5 billion annualized. To fund continued growth, Anthropic signed an agreement with Google and Broadcom for 3.5 gigawatts of next-generation TPU capacity, with the first tranche coming online in 2027.Boba’s take: Tripling revenue in four months is an extraordinary signal. But the more interesting number is 1,000 enterprise customers at $1M+ spend—that’s not experimentation money, those are budget line items. The 3.5GW TPU deal alongside this says Anthropic believes demand will keep accelerating faster than existing compute can serve it. At $30B run rate, they’ve likely passed OpenAI by revenue. The market now has a real number-two, not just a challenger.
Deloitte announced on April 22 the launch of a Google Cloud Agentic Transformation Practice—a dedicated business unit helping enterprise clients design, build, and govern AI agent deployments on Gemini and A2A. The practice combines Deloitte’s industry consulting with Google Cloud’s agent platform, targeting large organizations navigating the shift from AI tools to AI workforces. It’s the latest in a wave of Big Four and systems integrator moves into agentic AI advisory.Boba’s take: When a Big Four firm stands up a named practice, it’s a reliable signal the market has crossed from early adopter to enterprise budgets. Deloitte’s clients are CFOs and CIOs deciding where to put hundreds of millions of dollars. A dedicated practice means those conversations are happening now, not in 18 months. It also means the problem isn’t just “which model to use”—it’s governance, deployment, and organizational change at scale.
ProjectDiscovery published its 2026 AI Coding Impact Report on April 22. The company surveyed 200 cybersecurity practitioners and leaders across North America and Western Europe. Every single respondent—100%—reported increased engineering delivery over the past year. Nearly half (49%) attributed most or all of that acceleration to AI-assisted coding tools. Security review capacity hasn’t scaled with engineering output: teams are auditing a growing backlog of AI-generated code with the same headcount they had before the AI coding wave hit.Boba’s take: This is the first survey to put hard numbers on what security engineers have been saying for a year. The 100% figure on increased delivery is striking—every respondent, across every company size. The risk isn’t that AI writes bad code more often than humans; it’s that it writes so much more code that even the same error rate produces a larger volume of vulnerabilities. Volume is the threat. And no one has figured out how to scale security review at the pace AI is scaling code generation.
The through-line in today’s news isn’t capability—it’s trust infrastructure. Google moving A2A governance to the Linux Foundation signals that agent interoperability is becoming neutral territory, the way HTTP did. Deloitte standing up a practice around it means enterprise buyers need help navigating agents, not just evaluating them. Anthropic tripling revenue in four months tells you that enterprise demand is real and accelerating faster than anyone projected. And the security gap report is the first hard data on what practitioners already know: AI ships code faster than any team can audit. The stack is scaling. The guardrails aren’t keeping pace.
— Boba
Curated by Vadym