How to build Claude analytics agent in Anthropic Cloud
Today's top 15 insights for PM Builders, ranked by relevance from Blogs, YouTube, X, and LinkedIn.
How to build Claude analytics agent in Anthropic Cloud
#1 đź“ť OpenAI News
How HP is scaling early AI wins across the enterprise - HP announced it will scale activation of its OpenAI Frontier strategic partnership (tested since February 2026) to deploy agents and AI workflows across customer- and partner-facing experiences, telemetry, employee productivity, and software development. Early pilots reportedly let one engineer clear 122 pull requests across 43 projects in weeks, enabled a security team to remediate several bugs in a day instead of up to a month, unlocked roughly 82 hours/week of security-team capacity, and aim to serve 100,000+ partners (with ~80% of HP’s business flowing through partners) via AI-enabled portals and support.
#2 ▶️
Inside Anthropic’s Bet on Claude Agents that Work While You Sleep | Jess Yan
Peter Yang
Jess Yan demonstrates how to build a Claude analytics agent from scratch in the Anthropic Cloud console using Cloud Managed Agents, granting it Python tool access to analyze a multi-million line 'Just in Time' grocery store dataset and generate three structured HTML reports including a predictive customer return model.
- Configured the agent with model selection (Claude), system prompt detailing data schemas, built-in file system and Python tool access, and automatic tool permissions in the Anthropic Cloud console.
- Uploaded a multi-million line CSV of a fictitious grocery store named 'Just in Time' and ran sequential analysis via Python packages to produce three HTML outputs: order pattern summary, shopper heat maps with radar charts, and a predictive return-probability model.
- Leveraged Cloud Managed Agents' Traces feature to visualize each agent action for debugging and built-in Evals loops—using binary pass/fail and LLM-as-judge scoring—to self-grade and refine outputs within the same session.
#3 ▶️
OpenAI Codex lead on the new shape of product work | Andrew Ambrosino
Lennys Podcast
Andrew Ambrosino demonstrates using Codex’s computer use feature to automate navigating the Google Cloud Console UI and configure Pub/Sub API triggers for an inbox spam filter.
- Codex’s computer use feature executed UI clicks in Chrome to set up Google Cloud Pub/Sub topics and subscription triggers without manual intervention.
- Since January, Codex usage has grown sixfold to over 5 million weekly active users.
- Nearly 100 percent of OpenAI employees across all disciplines use Codex weekly, not just engineers.
#4 đť•Ź
Harrison Chase announces Harbor, a LangChain/​LangSmith integration for running sandboxed evaluations, with self-hosted sandboxes coming soon.
#5 đť•Ź
clem 🤗 argues it makes sense to regulate closed-source frontier API models to ensure government transparency while leaving open-source AI unregulated, since closed APIs pose higher risks than open weights.
#6 đť•Ź
Santiago: China’s AI firms each race to release their own state-of-the-art models while open-sourcing them, creating a blend of competition and collaboration. He wishes the US would adopt a similar open approach.
#7 đť•Ź
Santiago open-sourced a browser automation system that records your web actions, prunes retries and dead ends, and abstracts the core logic into reusable skills. It then organizes these skills into a skill graph so agents can retrieve and apply them to new, related tasks.
#8 đť•Ź
Teresa Torres highlights Override Labs—a philanthropy-backed incubator founded by Priya that builds prevention-first AI to combat harms to women and children, launching a flagship tool to prevent teen sexual assault.
#9 đť•Ź
Boris Cherny sees five key archetypes—Prototyper, Builder, Sweeper, Grower, and Maintainer—in the Claude Code team. He argues that as engineering, product, and design meld, individuals will fluidly span these roles rather than traditional job functions.
#10 in
Marc Baselga highlights Fiona Fung’s “latent demand” insight from Anthropic’s Claude Code and Cowork teams. He argues that when users hack AI tools into unintended tasks—like non-engineers stretching Claude Code—it reveals hidden demand and guides product improvements.
#11 đź“ť Simon Willison
Agent in the loop - Simon highlights a Jon Udell piece arguing we should stop saying "human in the loop" and instead view agents as teammates we invite into our established workflows, avoiding opaque agent-only processes. The piece reframes agent-assisted development as collaborative rather than ceding authority to machines.
#12 đť•Ź
Harrison Chase announced that Fleet agents can now be embedded directly into existing work channels like Slack and Teams. Caspar’s demo video shows exactly how to set them up in Slack.
#13 đť•Ź
Lenny Rachitsky: Andrew Ambrosino’s Codex desktop app at OpenAI now has over 5 million weekly active users (6× growth since February) and near-100% employee adoption.
#14 đť•Ź
Lenny Rachitsky shares @ajambrosino’s insight that when anyone can build anything, the real competitive edge shifts to having the taste to know what to build.
#15 đť•Ź
Peter Yang says Anthropic PMs use Claude Managed Agents with direct codebase access to track PR merges and deployments, streamlining state management and giving PMs deeper, hands-on product insights.