You picked your team's coding agent last month. Maybe it was Claude Code, maybe Codex, maybe Cursor. You almost certainly did not pick anything from Google — even though Google's tokens (the chunks of text an AI reads and writes, billed per million) cost a fraction of Anthropic's.

That reflex is the whole story. Google has the cheapest inference in the industry — running models is what "inference" means, and Google does it on its own TPU chips instead of renting Nvidia GPUs like everyone else. Yet its developer agent surfaces — Vertex ADK, Jules, Gemini CLI, Antigravity — barely register in any 2026 adoption survey. An agent, in case the word has lost meaning, is an AI that decides what to do next on its own, not just answers your question.

Six days from today, on April 22–24, Google Cloud Next 2026 opens at Mandalay Bay in Las Vegas. The session library went live February 24, and the Workspace keynote teaser promises "platform advancements in agentic AI" plus a "sneak peek at the upcoming 2026 product roadmap." Translation: Google knows it has to relaunch the agent story after Anthropic shipped Managed Agents on April 8 and OpenAI open-sourced its Agents SDK on April 15. 😼

Here's the inventory Google walks in with. Vertex AI Agent Builder — an umbrella of three products: ADK (Agent Development Kit, the framework you write agents in), Agent Garden (templates), and Agent Engine (the runtime, billed per vCPU-second, meaning you pay for every second a virtual processor is busy). ADK now speaks Python, Java, TypeScript, and Go, and Google claims over 7 million downloads. Jules, the asynchronous coding agent — "asynchronous" meaning it works in the background while you do something else — runs on Gemini 2.5 Pro and left public beta in August 2025. Gemini CLI, a terminal tool, just got a Jules extension. Antigravity, the IDE.

Now the JetBrains AI Pulse from April 2026, surveying 10,000+ professional developers. GitHub Copilot: 29% at-work adoption. Claude Code: 18% globally, 24% in US/Canada, with the highest customer-satisfaction score of any tool — 91% CSAT, NPS 54. Cursor: 18%. Google Antigravity: 6%. Gemini-as-chatbot for coding: 8%. Jules and Gemini Code Assist? Not even ranked separately. 😹

The price gap makes it absurd. Gemini 3 Flash costs $0.50 input / $3.00 output per million tokens. Claude Sonnet 4.6 costs $3.00 / $15.00 — so Claude is roughly 5× more expensive per token, and Gemini 2.5 Flash-Lite is closer to 20× cheaper. Gemini API traffic hit 85 billion monthly requests in January 2026, up 142% year-over-year. Developers clearly use Google's models. They just don't build agents on Google's platform.

The structural gap isn't the model and isn't the price — it's product shape. Anthropic shipped one opinionated coding surface called Claude Code and let it become a verb. Google shipped a buffet: Vertex, ADK, Jules, Antigravity, Gemini CLI, Workspace Studio, plus the rumored "Project Jitro" Jules successor leaking through TestingCatalog. Each lives under a different VP. Each has its own roadmap. None of them is the answer when a tech lead asks "what do we use?" 🐈‍⬛

Radical simplification at Next would mean cannibalizing Workspace's, Cloud's, and DeepMind's agent surfaces in favor of one product with one name. The org chart fights this. Free Gemma 4 (the open-weights model line), paid Gemini API, Vertex Agent Engine, and Workspace Studio currently tell four different stories to four different buyers. Even Simon Willison, who tracks every Gemini release down to the point version, spent April writing about Gemini 3.1 Flash TTS and Gemma 4 on iPhones — not one post on Vertex, ADK, or Jules. That silence is the data.

If you're picking an agent platform in Q2 2026, watch three signals during the April 22–24 keynotes. First: a single named coding-agent product, not another Jules preview tier. Second: one default ADK-on-Vertex recipe a junior dev can follow in an afternoon. Third: a token-plus-orchestration price that visibly undercuts Anthropic's roughly $0.08/session-hour for Claude Code. Anything less and the cheap-tokens-no-mindshare loop continues.

Ship those three and the agent race is genuinely three-horse by May. Ship another buffet of overlapping betas and Gemini becomes the Intel Inside of AI — cheapest silicon in the box, zero brand on the lid. 😾