Build production agents with computer use, the Skills API, and the Files API
Computer use, the Skills API, and the Files API are now GA on the Claude Platform, including a new browser-use tool for agents that operate inside web applications.
AI Daily Recap
18 articles · 5 categories
The finishable daily brief
What happened in AI — Aug 20, 2026
Thursday, Aug 20, 2026
18 articles · 5 categories
read top to bottom · then stop
In 30 seconds
- Anthropic pushed computer use, the Skills API, and the Files API to GA today, alongside a startup guide and an agent-first case study from monday.com.
- A fresh cluster of agent tooling shipped: LangSmith Preview Builds for testing PR branches, Pond for S3-backed session archives, and Do-over for undoing destructive agent shell commands.
- DeepSeek open-sourced its agent harness and shipped a 14-change post-beta update adding multimodal and subagent support.
- AWS and Google Cloud both published production-readiness playbooks for agentic systems, while IEEE Spectrum flags CPUs as agentic AI's new bottleneck.
- Coding-agent adoption kept showing up in formal channels: JetBrains research, a new InfoQ certification, and a Codex-powered launch case study at Stampli.
Anthropic had a coordinated platform push today: computer use, the Skills API, and the Files API all moved to general availability, backed by a startup playbook and a monday.com case study on building agent-first products. DeepSeek moved the same direction on the open-weight side, open-sourcing its agent harness and shipping its first major update since beta.
Elsewhere, the story was production friction: AWS and Google Cloud both published readiness checklists for running agentic systems at scale, and IEEE Spectrum makes the case that orchestration overhead is turning CPUs back into the bottleneck.
Claude's Agent Platform Reaches GA 3 items
Anthropic moved computer use, the Skills API, and the Files API from preview to general availability today, then backed the launch with a startup playbook and an agent-first case study.
The Claude Code Guide For Startups
Anthropic's guide distills five rules startups use with Claude Code, from an everyone-ships culture to AI-native SDLCs.
How monday.com transformed its platform into an agent-first product where humans and agents collaborate
monday.com rebuilt its platform around Claude so agents work alongside humans directly inside the product rather than through a bolted-on chatbot.
New Building Blocks for Agent Engineers 5 items
A cluster of new tools targets specific agent-engineering pain points today: testing changes safely, recovering from destructive actions, and decoupling coding agents from any single model.
Test Agent Changes with LangSmith Preview Builds
LangSmith Preview Builds spin up temporary, production-like deployments per pull request, so teams can test agent changes before merging instead of shipping straight to prod.
Show HN: Pond – lossless archive for agent sessions in your own S3
Pond stores agent sessions from multiple machines in one S3 bucket you own, filling the gap left by tools that can't preserve cross-machine session history without running a database.
Show HN: Do-over, undo for AI agent shell commands
Do-over hooks into Claude Code and snapshots what a shell command is about to touch right before it runs, giving builders an undo after a coding agent deletes the wrong files.
The /wayfinder Skill: Navigating the "Fog of War" of Planning
Matt Pocock's /wayfinder skill helps Claude Code plan through greenfield projects or moments when the way forward isn't clear yet.
Use your favorite coding agent with any model
Aivo decouples a coding agent from any single model provider, letting builders swap the underlying model without switching agents.
Open-Weight Labs Push Agent Infrastructure Forward 3 items
DeepSeek open-sourced its agent harness and shipped its first major post-beta update today, while Z.ai's CEO argued post-training technique, not parameter count, is now the real scaling lever.
The Open-Sourcing of DeepSeek Harness Opens the Door to Modular, Unbundled AI Agent Infrastructure
DeepSeek open-sourced its agent harness, opening the door to modular, unbundled agent infrastructure instead of monolithic, single-vendor frameworks.
DeepSeek Harness Ships First Major Post-Beta Update: 14 Changes Bring Multimodal and Subagent Upgrades
DeepSeek's harness update ships 14 changes including multimodal support and subagent upgrades, its first major revision since leaving beta.
[AINews] Death of Params: Z.ai CEO Jie Tang on GLM 5.3 and the new Post-training Scaling Law
Z.ai CEO Jie Tang argues post-training technique, not raw parameter count, is now the dominant scaling lever behind GLM 5.3.
Running Agentic Systems in Production 4 items
Today's enterprise posts converge on one theme: agentic workloads strain infrastructure and process differently than single-shot LLM calls, from CPU-bound orchestration to consent scopes for agent-driven OAuth flows.
Scaling agentic AI: Enterprise patterns without vendor lock-in
AWS's second post in its multi-agent series lays out enterprise patterns for running many agentic systems at once without locking into one vendor's stack.
10 questions every startup should answer before moving to production with their AI prototype
Google Cloud lists the production-readiness questions on reliability, cost, and ownership a startup should answer before shipping an AI prototype, not after.
Agentic AI has made CPUs the new performance bottleneck
Constant orchestration, tool-calling, and control-flow logic in agentic AI is shifting the performance bottleneck back to the CPU after years of GPU-centric optimization.
From all-or-nothing to task-based OAuth consent
Cloudflare's OAuth consent flow now supports task-based optional scopes instead of all-or-nothing grants, letting developers scope what an agent or app can actually touch.
Coding Agents Move From Pilot to Practice 3 items
Coding-agent adoption showed up in adoption research, formal certification, and a launch case study today, not just vendor announcements.
How ChatGPT Work helps Stampli move ideas to market
Stampli used Codex and ChatGPT Work to compress weeks of launch production into days after design resources got pulled elsewhere.
AI Coding Agents Adoption Trends
JetBrains' developer survey research tracks how AI coding agent adoption is shifting across its user base.
InfoQ Opens Enrollment for New AI-Assisted Engineering Online Certification Program
InfoQ opened enrollment for a five-week AI-Assisted Engineering certification aimed at senior engineers already running a coding agent against production code.
You are caught up for this edition