How Much Does AI Know About the Agentic Architect Exam Scope? 28 Tools Checked Against Official Docs

The exam guide for Google Cloud’s Professional Agentic Architect puts a list of in-scope tools on page 4. It has 28 lines, and familiar services such as BigQuery and Cloud Run sit next to new names such as Agent Gateway, Agent Registry, Skill Registry and Agents CLI. If you study the Agentic Architect exam scope by asking an AI to explain it, the first question is how far the AI’s knowledge has caught up with these new products. On top of that, Google renamed Vertex AI (now Gemini Enterprise Agent Platform) on April 22, 2026, so the same product appears under different names in older and newer documents.

This article contains affiliate links. As an Amazon Associate, we earn from qualifying purchases.

So we asked claude -p (Claude Sonnet 5.5) to explain each of the 28 items, one item per call, and checked the answers against Google Cloud’s official documentation retrieved on October 12, 2026. We ran two conditions once each: asking with no materials (condition A) and asking with that item’s official page attached (condition B).

The results first. In this run, even condition A produced zero definitions that contradicted the official materials, and 18 of 28 definitions were correct. For new agent products it either said it did not know or answered while flagging that it was guessing. Yet although the renames from Vertex AI happened before the model’s knowledge cutoff (June 2026), condition A gave Model Garden and RAG Engine their old Vertex AI names. In condition B all 28 definitions were correct, and all 10 items with former names came back with current names. Release status (GA or preview), however, came back Unknown for 22 of the 24 scored items, because GA service pages do not say “GA” anywhere.

We have not taken this exam. This article covers not exam questions but the names, definitions and release status of the tools the exam guide lists. As of October 2026, GA registration opens on November 2 and delivery starts in mid-November. For an overview of the certification and a preparation plan, see What the Google Cloud Professional Agentic Architect Certifies.

The 28 items mix renamed products and brand-new ones

The exam guide is a four-page English PDF. The last page says “The following tools are in scope for this exam:” followed by 28 lines. We count “28 items” by lines; line 9, “Agentic protocols (e.g., Agent2Agent [A2A], MCP)”, covers two standards, so the count does not equal the number of products.

The PDF text has no revision date. The file’s recorded creation date is August 31, 2026, but that is not necessarily a revision date, so we treat it as “the version retrieved on October 12, 2026”. By nature the 28 items fall roughly into three groups:

1. Google Cloud foundation services not specific to agents: BigQuery, Cloud Run, Cloud SQL, Cloud Storage, Firestore, Google Kubernetes Engine (GKE), Memorystore for Redis, Google Cloud Observability, Sensitive Data Protection
2. Products renamed from Vertex AI names: Agent evaluation (the related evaluation service was renamed), Agent Retrieval and Vector Search 1.0, Agent Runtime, Agent Search, Model Garden, RAG Engine
3. Products and standards for building, connecting and governing agents: Agent Development Kit (ADK), Agent Gateway, Agent Identity, Agent Registry, Agentic protocols (A2A and MCP), Agents CLI, Antigravity, Auth Manager, Gemini Enterprise, Gemini LLMs, Model Armor, Model Context Protocol (MCP) servers, Skill Registry

The first group has agent-related changes too. BigQuery, Cloud Run, Cloud SQL, Cloud Storage and others have Google-managed remote MCP servers (for BigQuery, bigquery.googleapis.com/mcp), and GKE documents an Agent Sandbox for isolating untrusted code. Still, the names and status that are moving most are in the second and third groups. Item 2.1 of the guide also mentions Claude Code, but it is not among the 28.

Agent Platform renamed Vertex AI as its evolution

The guide’s body uses Agent Platform names such as “Agent Platform Memory Bank” and “Agent Platform Gen AI evaluation service”. Agent Platform is short for Gemini Enterprise Agent Platform, which Google describes as a unified platform to build, deploy, govern and optimize enterprise AI agents and models — positioned as an “evolution” of Vertex AI.

The April 22, 2026 release notes list the rename from Vertex AI to Gemini Enterprise Agent Platform alongside the first Agent Platform release, and say Vertex AI and Agent Builder are now part of Agent Platform. The official wording is “evolution” and “now part of”; nothing was discontinued. Agent Platform has four pillars — Build, Scale, Govern and Optimize — and agent evaluation sits under Optimize. Note that the guide adds former names only for Agent Runtime (formerly Agent Engine) and Agent Search (formerly Vertex AI Search); the Model Garden and Vector Search renames are not mentioned, so older guides and blogs will show names you will not find in the exam guide.

Gemini Enterprise and Gemini Enterprise Agent Platform are different products

Line 18, Gemini Enterprise, is easy to confuse. Gemini Enterprise is a product for business users that combines enterprise search, an AI assistant and an agent platform. Its predecessor was Google Agentspace; the Gemini Enterprise release notes of October 9, 2025 announced that Agentspace had become part of Gemini Enterprise.

Gemini Enterprise Agent Platform, by contrast, is the evolution of Vertex AI described above — the foundation for people who build agents. The Agentspace rename (October 2025) and the Agent Platform rename (April 2026) are separate events; Agentspace was not merged into Agent Platform in April 2026.

Building an answer key from the official documentation

As the basis for checking, we built a table that records, for each of the 28 items, a one-sentence official definition, its release status as of October 12, 2026, and any former name with the rename date, all confirmed from the original text of Google Cloud documentation. We call it the answer key. Sources were each product’s overview page, the Agent Platform release notes and the page summarizing the Agent Platform name changes; for open-source ADK and A2A we used their official sites and GitHub releases. No third-party articles were used.

How status is recorded

We recorded status as GA, Preview, Private Preview or released as open source. The catch is that Google Cloud docs show nothing on GA pages. Preview pages show a notice that they fall under the “Pre-GA Offerings Terms”, but GA pages carry no equivalent marker. As a result, for long-standing services such as BigQuery we could not find original text that states GA. By evidence, the answer key’s status for the 24 scored items breaks down like this (the four unscored items are explained under “How answers were judged”):

• Confirmed by release notes that state GA (4 items): Agent Gateway, Agent Registry, Agent Identity, Auth Manager. Auth Manager’s GA appears in the IAM release notes, not the Agent Platform notes
• Confirmed by the Pre-GA notice and a public-preview announcement (1 item): Skill Registry
• Confirmed as released by release records (3 items): ADK, Agents CLI, Antigravity
• Inferred as GA from the absence of a Pre-GA notice (16 items): the nine foundation services in group one, plus Agent evaluation, Agent Retrieval and Vector Search 1.0, Agent Runtime, Agent Search, Gemini Enterprise, Model Armor and Model Garden

Much of the answer key’s status therefore rests on that inference.

Where official sources disagree, scoring rules were fixed in advance

Reading the official materials side by side, we found ten points where sources disagree or where no original text could be found.

• ADK languages: the Agent Platform docs list four (Python, TypeScript, Go, Java); the ADK site lists five, adding Kotlin
• The evaluation service’s new name splits three ways: “Gemini Enterprise Agent Platform Evals” in the name-change table, “Gen AI evals” in the release notes and “Agent Platform Gen AI evaluation service” in the exam guide
• Whether the Agent Search rename is complete: the name-change table and release notes say yes; the support page says the rename is in progress and content is being updated
• RAG Engine status varies by region: GA, preview or allowlist
• Agents CLI subcommands: the answer key checks only the install command
• Google Cloud Observability’s former name: no official text found
• Antigravity’s “App”: the guide says “CLI, SDK, App”, while the docs home lists four surfaces — the Antigravity 2.0 desktop app, the CLI, the SDK and the IDE (plus IDE extensions for existing editors). No text says which one “App” means
• Latest Gemini models: new models went GA after the knowledge cutoff
• MCP servers: the list of Google Cloud MCP servers mixes GA and preview
• Leftover Agent Runtime names: “Agent Engine” still appears in release notes after the rename

We decided the scoring for each point before running anything. ADK’s language count and the status of RAG Engine and MCP servers are not scored. Any of the three evaluation-service names counts as current, but giving only the old name counts as using an old name. For Antigravity, naming any of the four surfaces is accepted. “Agent Engine” for Agent Runtime counts as an old name even though it lingers in official materials.

The model’s knowledge cutoff, and events on either side of it

In Anthropic’s model overview, Claude Sonnet 5.5 has a reliable knowledge cutoff and a training data cutoff of June 2026. Only the month is given, so we write “knowledge cutoff: June 2026”. Events related to the 28 items, sorted around the cutoff:

Gemini 3.6 Flash through 3.8 Flash also went GA after the cutoff. In a setup with no search and no materials, the AI cannot learn about post-cutoff changes, so an A answer of “I do not know” is not counted as wrong. Changes in the cutoff month could go either way, so we mark them as the boundary.

Conditions — no materials (A) versus official page attached (B)

How we asked

We ran claude -p with –model sonnet; the recorded model ID was claude-sonnet-5-5. The system prompt was replaced with the single sentence “You are a careful assistant.”, with no tools, MCP or settings files loaded, and each call asked about one item only. We passed the exam guide’s wording verbatim and asked for an answer as of October 2026 in this JSON shape (the original asked for the definition in Japanese; field descriptions are translated here):

{
  "definition": "one sentence on what it does",
  "current_official_name": "current official product name (English)",
  "former_names": ["former names in English, if any"],
  "status": "GA | Preview | Private Preview | Open source release | Unknown",
  "confidence": "high | medium | low"
}

The prompt said not to fill gaps with guesses and to answer Unknown for status when unsure. In condition B we also attached the body text of that item’s official page (up to 15,000 characters per page) and the Agent Platform name-change page, and told the model to rely only on the materials and answer Unknown for anything not written there. The attached pages were the same October 12, 2026 versions used for the answer key.

How answers were judged

Definitions were compared one by one with the answer key’s official sentence and sorted into five categories.

“Contradicts official” is reserved for explanations that conflict with the official materials. An explanation we simply could not find in them is not proof that it is false, so it is not placed in this category.

Names and status were counted mechanically by script. For names, we checked whether the 10 items with former names used an old name in the current-name field or the definition. For items on Agent Platform, the platform’s old name (Vertex AI) also counted as an old name, which is why Agents CLI and Gemini LLMs — with no former product names of their own — are among the 10. Answers that merely added an old name with “formerly …” were not counted as using it. Status was scored on 24 items, excluding four: Agentic protocols (a standard with no product status), Gemini LLMs (GA and preview models mixed), MCP servers (status varies by product, decided before the run) and RAG Engine (status varies by region). A2A and MCP were treated as one item, as in the guide.

The first run of condition B dropped the Pre-GA notice from preview pages because of a bug in extracting page text. We fixed the extraction, reran B only, and did not use the first run’s answers.

Results — no contradictions even without materials, and “I do not know” for new products

All numbers below are from a single run per condition.

What stood out is that condition A had no “contradicts official” answers at all. Eighteen definitions were correct, and four items (Agent Identity, Agent Registry, Agent Retrieval and Vector Search 1.0, Auth Manager) were answered correctly after being flagged as guesses. The five it said it did not know were Agent evaluation, Agent Gateway, Agents CLI, Antigravity and Skill Registry.

The one item judged “missing the core” was Model Context Protocol (MCP) servers. The core of the official definition is Google-run remote MCP servers that let MCP clients use Google and Google Cloud services through HTTP endpoints on Google’s infrastructure. Read by name alone, it looks like MCP servers in general, which makes it easy to get wrong.

Confidence is the AI’s own high/medium/low label, not a probability of being right. Even so, all eight answers A labeled high had correct definitions; B labeled only two as high. B cost more, presumably because the input included page text. Billing depends on your plan, so treat cost as a rough guide.

Results by item

Condition B’s definitions were all correct, so the table shows A’s definitions and both conditions’ status. “Not scored” marks the four excluded items.

Status was right for familiar services, ADK and Agent Runtime

The 14 items where A’s status was correct were the foundation services BigQuery, Cloud Run, Cloud SQL, Cloud Storage, Firestore, GKE and Memorystore for Redis; Gemini Enterprise, Google Cloud Observability, Model Armor, Model Garden and Sensitive Data Protection; plus ADK and Agent Runtime. For 13 of them (all but ADK) the correct answer is the inferred GA from the missing Pre-GA notice. No status answer was wrong, and A answered Unknown for the new agent products (Agent Gateway, Agent Registry, Agent Identity, Skill Registry and so on).

The timeline shows that the AI does not necessarily know about events before the cutoff. The Agents CLI announcement (April 22, 2026) and Skill Registry’s public preview (May 19, 2026) both predate it, yet A said it did not know either. Why, this run cannot tell. What matters to a learner is that the items where the AI honestly says it does not know become a ready-made list of items to read in the official docs yourself.

Agent Runtime and Agent Search, whose exam-guide wording includes the former name, got correct definitions and current names in A. But since the item names themselves contain “formerly Agent Engine” and the like, that does not prove the AI knew about the renames.

The renames predate the cutoff, yet without materials the old names came back

Of the 10 items with former names, A answered with the current name for 4, used an old name for 3 and gave no name for 3. The three with old names:

• Model Garden: wrote the pre-rename Vertex AI Model Garden in the current-name field, without “former”
• RAG Engine: wrote the pre-rename Vertex AI RAG Engine in the current-name field, without “former”
• Gemini LLMs: in the definition, listed the platform under its old name (Vertex AI) next to the Gemini API as a route to Gemini

The Vertex AI rename happened on April 22, 2026, before the June 2026 cutoff. Even so, without materials the AI used Vertex AI names for these three. The reason is unclear, but the conclusion for learners is clear: assume the AI may answer with old names and correct them yourself. In condition B all 10 used current names. A also offered some former names not in the answer key; those are outside the key, so we did not score them or judge them here.

Official materials still carry old names

That said, an AI answering with an old name is not simply wrong, because Google’s own materials keep old names. As of October 12, 2026 we found three cases:

• Agent Runtime: the API resource is still ReasoningEngine for backward compatibility, and the June 18, 2026 release notes, after the rename, still mention ADK agents on “Agent Engine”
• Agent Search: the API is still the Discovery Engine API and doc URLs still contain vertex-ai-search and generative-ai-app-builder; the support page says the rename from Vertex AI Search to Agent Search is in progress
• Sensitive Data Protection: Cloud DLP is now part of it, but the official docs say the API is still called the Cloud Data Loss Prevention API (DLP API)

The name-change table itself is hard to read in places. Its RAG Engine row shows the same “RAG Engine on Gemini Enterprise Agent Platform” before and after, while the name Vertex AI RAG Engine appears in the release notes. Checking former names in a single source will miss some. In short, the exam guide uses new names while APIs and slightly older docs use old ones. Learners need both, and the name table below is the tool for that.

Why a model answers with old names cannot be read from this run alone. But following once, in your own hands, what a large language model actually learns makes the term “knowledge cutoff” much easier to reason about.

Hands-On Large Language Models

USD 37.68 on Amazon.com (as of 2026/09/22)

Amazon

ポチップ

With the official page attached, definitions and names lined up, and status leaned to Unknown

In condition B all 28 definitions were correct and all 10 names were current. Simply attaching the material turned the five “did not know” items and the four guesses into answers matching the official definitions.

Status, however, was correct for only 2 of the 24 scored items (ADK as released and Skill Registry as preview); the other 22 were Unknown, with zero wrong. The reason is the way Google Cloud docs are built: preview pages show a Pre-GA notice at the top, but GA pages show nothing equivalent (some pages, like RAG Engine’s, include a table of GA/preview status by region in the body). Told to rely only on the materials and answer Unknown for anything not written, the AI read GA pages and reported that status was not stated. That is the instructed behavior, not an error.

Agent Runtime is the telling case: A answered GA correctly, while B, given the official page, said Unknown. Attaching material reduced the status answers. And when only the body text of a page is extracted for an AI, the Pre-GA notice shown in the page frame can be dropped — as happened in our first B run.

Check status with release notes and page labels

Do not rely on the AI for status; check it yourself. There are two clues. One is the page label: a Pre-GA notice means preview. The other is release notes, which have entries that state GA outright, such as “Agent Gateway in General Availability” or “Agent Registry is generally available (GA)”. Agent Gateway and Agent Registry GA (June 18, 2026) and Skill Registry public preview (May 19, 2026) all appear with dates in the release notes.

There may be more than one set of release notes to check. For the Agent Identity management APIs, the Agent Platform notes only have a June 18, 2026 preview entry, while the August 22, 2026 GA is in the IAM release notes; Auth Manager’s GA is in the same IAM entry. For long-standing services we often could not find original text stating GA when building the answer key either, so the only clue is the absence of a Pre-GA notice — an inference, not a confirmed GA announcement. The 16 items listed under “How status is recorded” rely on it.

Some products also have split status. Main examples as of October 2026:

• Agent Registry: GA, but registering standalone skills is in preview
• Agent Runtime: GA, but the section on multi-agent setups with A2A is in preview
• Agent Gateway: GA; the Cloud Trace integration has been in preview since September 30, 2026
• RAG Engine: allowlisted GA in us-central1 (Iowa), preview in asia-northeast1 (Tokyo)
• Google Cloud MCP servers: GA and preview mixed by product; the Agent Platform remote MCP server went GA on April 22, 2026

If an AI says “Agent Registry is in preview” for the whole product, that is wrong. If it just says “GA”, it misses that standalone skill registration is in preview. Instead of memorizing status as one word, learn it as “the product is GA, this feature is in preview”.

Name table — from former names to current names

From the results we built a table mapping former names to current names, listing only names confirmed in the answer key. The “What it does” column is our paraphrase of the official one-sentence definition.

A few notes. “Vector Search 1.0” is not a former name: Agent Retrieval was formerly Vector Search 2.0, and 1.0 is the older version still offered. The two work on different units, and an official migration path from 1.0 to Agent Retrieval exists. Google Cloud Observability is not in the table because no official text giving a former name was found, and we did not confirm the date of the Sensitive Data Protection rename. The Gemini Enterprise rename was in 2025, a separate event from the April 2026 renames, so Gemini Enterprise and Gemini Enterprise Agent Platform have separate rows.

Model names are not renames but need the same care because generations change. As of October 12, 2026, Agent Platform’s GA model list includes Gemini 3.8 Flash, and Gemini 3.1 Pro is in the preview list. gemini-2.5-pro, gemini-2.5-flash and gemini-2.5-flash-lite retire on Agent Platform on October 20, 2026, this article’s publication date. If an AI without materials recommends Gemini 2.5 as current, the answer is out of date.

Give the AI official material before you ask — a study method built from the results

Here is how to turn the results into a way of studying the Agentic Architect exam scope. Handing over material before asking is the same idea as putting “import” first in the AI-assisted method in What the JDLA G-Test Proves. The difference here is that the scope material itself is new, and both names and status are moving.

1. Collect each item’s official page and give it to the AI

In condition B, attaching each item’s official page was enough to make all 28 definitions and all 10 names match the official text. The Sources section at the end lists the main pages we used. When you attach a page, also write the instruction “use only the material, and answer Unknown for anything it does not say”. In our B run the model followed it, did not guess status, and made zero status errors. We cut each page at 15,000 characters; if you trim long pages, keep track of how much you passed.

2. Keep the name table at hand

Start from the table above and add names from documents you read. In particular, API names (ReasoningEngine, Discovery Engine, DLP API) do not match the exam guide names, so keeping them in the table saves confusion when reading API references.

3. Check status yourself with release notes and page labels, not the AI

In this run A made no status errors but answered Unknown for the new agent products, and B could not read status from GA pages. In neither condition can the AI’s answer settle status. Check GA or preview announcements in the release notes and Pre-GA notices on pages, and record the date you checked. Status moves, so learning it with a timestamp — “GA as of October 2026” — helps when you review after GA delivery starts.

4. Start reading where the AI said it did not know

The five items A did not know and the four it guessed are where the AI helped least in this run. Read the official one-sentence definition yourself before asking the AI. Short paraphrases of each:

• Agent Gateway: the central policy-enforcement component of Agent Platform and the entry and exit point for all agent interactions. By default it blocks every connection unless an explicit IAM policy allows it
• Agent Identity: gives each agent a cryptographically attested identity based on the SPIFFE standard. Unlike service accounts, it is not shared across workloads by default and cannot be impersonated
• Agent Registry: a single catalog in Agent Platform for managing agents, tools and MCP servers
• Auth Manager: formally Agent Identity auth manager. A credential vault and authentication broker that reduces the work of agents authenticating to external tools, handling OAuth 2.0 flows including user consent and token refresh
• Skill Registry: a secure, private, low-latency store for managing agent skills; in preview as of October 2026
• Agents CLI: not a coding agent itself, but a CLI and skills that let coding agents create, evaluate and deploy ADK agents to Google Cloud. The docs name Gemini CLI, Claude Code and Codex as tools to pair it with
• Antigravity: four surfaces — the Antigravity 2.0 desktop app, a terminal CLI, a Python SDK and an IDE — plus IDE extensions for existing editors. Its model list includes Claude Sonnet 5.5 alongside Gemini, with availability set by plan
• Agent Retrieval and Vector Search 1.0: as in the two rows of the name table
• Agent evaluation: measures and improves agent performance, safety and quality; the docs say ADK agents can also be evaluated locally

Add post-cutoff changes yourself as well. MCP (Model Context Protocol) is an open protocol created by Anthropic and donated in December 2025 to the Agentic AI Foundation under the Linux Foundation; the 2026-07-28 version made it stateless, and Google Cloud’s remote MCP servers support that version. A2A (Agent2Agent) is an open standard started by Google and donated to the Linux Foundation, with 1.0-series releases out; Agent Registry handles both A2A 1.0 and 0.3. Neither standard is tied to one company’s tools, so understand the exam guide’s “Agentic protocols” by their role as standards.

For the overall preparation plan see the overview article linked at the top; for the map of Google Cloud certification levels, see Google Cloud Certification Guide 2026: The Next Cloud After AWS.

The pattern that worked here — give the model the right context first and tell it how to handle what the context does not cover — is the core of prompting with documents. A practical book on prompt design for LLMs covers it in depth.

Prompt Engineering for LLMs

USD 54.51 on Amazon.com (as of 2026/09/22)

Amazon

ポチップ

What this test cannot tell us

• One run per condition: we asked about 28 items once in each of two conditions and did not measure how much answers vary on repetition. All numbers mean “this is what happened in this run” and do not generalize
• One model, minimal setup: claude-sonnet-5-5 with no tools or search. Other models and web-search setups were not tested
• One way of asking: only the exam guide wording, answered in a fixed JSON shape. Asking with explanations or in conversation was not tested
• Definitions judged against one official sentence: borderline answers could be classified differently, and explanations merely absent from official materials were not counted as contradictions
• The judge was also an AI: definitions were judged by the AI we use for this site (Claude), and a separate run of the same company’s AI reviewed 56 judgments afterwards. No humans or other vendors’ models checked them, so a model from the same company as the answering model (Claude Sonnet 5.5) did the judging
• Most status answers are inferred from a missing Pre-GA notice: 16 of the 24 scored items. Only Agent Gateway, Agent Registry, Agent Identity and Auth Manager were confirmed by explicit GA announcements, and 13 of A’s 14 correct status answers matched the inferred GA. Status is as of October 12, 2026 and will change
• B’s Unknowns come from the instruction and the page design, not from the AI lacking knowledge
• Confidence is not a probability of being right; high labels being right in this run does not guarantee the next
• Unrelated to exam content: we have not taken the exam, and this article says nothing about which products are tested or how heavily
• Cost reflects this run; billing depends on your plan

Even so, the record supports this much: a model with no materials made no contradictory definitions in this run, still used some old names, and lined up with official definitions and names once given the official page.

Summary — give the AI official material first for the Agentic Architect exam scope

• The exam guide lists 28 in-scope tools, named after the April 22, 2026 Vertex AI (now Agent Platform) rename. It adds former names only for Agent Runtime and Agent Search
• Without materials (condition A), this run produced zero contradictory definitions and 18 of 28 correct; new agent products got “I do not know” or a flagged guess
• Although the renames predate the June 2026 cutoff, A used Vertex AI names for Model Garden and RAG Engine. Official materials keep old names too (ReasoningEngine, DLP API), so learn both
• With the official page attached (condition B), all 28 definitions and all 10 names were correct. Status was Unknown for 22 of 24 scored items, because GA pages do not say “GA”
• Check status yourself, with a date, in release notes and Pre-GA notices — not with the AI

Names and status in the Agentic Architect scope are still moving. Give the AI the official page before asking, tie names to former names in your table, and read release notes for status yourself. As condition B shows, asking after providing official material makes it much easier to align AI answers with official definitions and names. Before GA registration opens, start by gathering the official pages for all 28 items.

Sources

• Professional Agentic Architect exam guide (PDF)
• Gemini Enterprise Agent Platform overview
• Gemini Enterprise Agent Platform name changes
• Gemini Enterprise Agent Platform release notes
• Model versions and lifecycle
• Agent Development Kit
• Agent Gateway overview
• Agent Identity overview
• Agent Registry
• Agent Retrieval (formerly Vector Search 2.0)
• Agent Runtime
• Agent Search: Get support (rename note)
• A2A Protocol
• Google Cloud MCP servers: supported products
• Model Context Protocol specification (2026-07-28)
• Anthropic: Donating the Model Context Protocol and establishing the Agentic AI Foundation
• Google Developers Blog: Agents CLI in Agent Platform
• Google Antigravity Docs
• Agent Identity auth manager overview
• IAM release notes
• Gemini Enterprise release notes
• Model Armor overview
• RAG Engine overview
• Skill Registry overview
• Anthropic: Models overview

ブラウザだけでできる本格的なAI画像生成【ConoHa AI Canvas】

swiftwand

X

AIを使って、毎日の生活をもっと快適にするアイデアや将来像を発信しています。 初心者にもわかりやすく、すぐに取り入れられる実践的な情報をお届けします。 Sharing ideas and visions for a better daily life with AI. Practical tips that anyone can start using right away.