DEV INFRASTRUCTURE / APIS · ESSENTIAL
Google Cloud
Google Cloud's technical account: how-tos, demos and product news.
What Google Cloud does
Google Cloud's technical account: how-tos, demos and product news. Google Cloud provides compute, data and AI services (incl. Vertex AI/Gemini for enterprise). @GoogleCloudTech posts technical content; company news is on @GoogleCloud.
Good for
- Host apps and data with enterprise-grade infrastructure
- Use Gemini via enterprise cloud contracts
- Learn new cloud/AI features from demos
Try this in 10 minutes
- Score the trajectory (which steps and tools the agent used), not just the final answer. (shown at 26:15 in the video)
- Build a small golden dataset: inputs mapped to expected outputs and tool sequences. (shown at 32:41 in the video)
- Use LLM-as-judge for fuzzy qualities like tone, but expect drift and extra cost. (shown at 33:53 in the video)
From our fact-or-hype test list for “Annie Wang (Google Cloud): Agent102 Table for N lab, PART 2 (24:30 to end) – evals, checkers and the self-improving loop (re-upload, video notes)”.
Honest take
No red flags in our notes so far
- Fact or hype verdict on Annie Wang (Google Cloud): Agent102 Table for N lab, PART 2 (24:30 to end) – evals, checkers and the self-improving loop (re-upload, video notes): the most useful free walkthrough we've seen of evals plus an "improve the prompt, re-test" loop. Skip the post and go to the codelab. Source video
- Fact or hype verdict on Google Cloud's free Graph RAG agent codelab (Spanner Graph + ADK), presented by Annie from Google, reposted by @ryouxbt as a course (re-upload, video notes): a solid, free, hands-on way to see what "Graph RAG" actually means (relationships in a graph + meaning-based search) on Google's stack; skip the post and go straight to goo.gle/graphrag. Source video
- Fact or hype verdict on Annie Wang (Google Cloud): Agent102 'Table for N' self-improving agent lab, sold as 'Google's free harness engineering course' (re-upload, PARTIAL video notes): a solid, free, hands-on lab for learning agent evaluation and self-improvement loops; ignore the 'course' framing and go straight to the codelab. Source video
Pricing
Pay-as-you-go usage-based; specific plan details Unknown here.
Prices as listed by the maker when we last checked (2026-10-30); usually in US dollars. Check the website for current prices.
Latest update
- 2026-10-30: Knowledge Catalog semantic model ingestion is now GA: pulls business metric definitions from Looker, BigQuery, dbt, Snowflake, Databricks and SAP into one governed context graph, so agents answer with the definitions analysts already use.
- : Gemini Robotics Embodied Reasoning 2 (ER 2) now available on Google Cloud (27 Oct), for building robots that reason, act and adapt in the physical world.
- : Agent-evaluation playbook (26 Oct): 3 building blocks (critical user journeys, ADK simulated users, four-layer checks) and 4 traps (moving baselines, fuzzy 1-5 judge scores instead of pass/fail, ignoring cost/latency, framework-locked evals), with a Day 0-2 rollout to CI merge gates using the open-source agent-eval toolkit.
- : Google open-sourced TorchTPU at PyTorch Conference 2026: native PyTorch on Cloud TPUs, new vLLM and SGLang serving backends (with Inferact and RadixArk) and TorchTitan on TPU for large-scale training with Meta, plus a TPU Developer Hub of quickstarts and code samples.
Is Google Cloud worth it? Quick answers
Is Google Cloud worth it?
Google Cloud is one of the tools we rate as essential to know about in the Dev infrastructure / APIs category. It is a paid tool, so test it on one real task before committing.
How much does Google Cloud cost?
Pay-as-you-go usage-based; specific plan details Unknown here.
What is Google Cloud used for?
Google Cloud's technical account: how-tos, demos and product news. Typical uses: host apps and data with enterprise-grade infrastructure; use Gemini via enterprise cloud contracts; learn new cloud/AI features from demos.