Pondero AI logo

Pondero AI

Archives
Log in
Subscribe
August 30, 2026

Pondero Brief: Claude Code boots before it loads; Sonnet 5 price hike canceled

Pondero Brief - AUGUST 30TH, 2026

Plus: Copilot ships org-wide custom agents, and Okta makes AI agents real identities.
pondero. BRIEF · AUG 30

Claude Code now starts before it finishes loading.

The August 29 update boots async, cuts the Linux binary from 338 MB to 75 MB per Anthropic's changelog, and adds per-subagent model and cost visibility.

Anthropic pushed a Claude Code update on August 29. The CLI now boots async: it takes prompts before the sandbox and MCP servers finish loading, so you type immediately instead of watching an init sequence.

Also in today's brief

  • Sonnet 5's September price hike is canceled - $2/$10 stays permanent
  • Copilot for Visual Studio now lets orgs publish custom agents to all repos
  • ChatGPT adds an effort slider and native Google Drive folder access
  • Okta makes AI agents first-class identities with short-lived token delegation
  • Cursor Cloud Agents no longer requires a GitHub account to start
 
Models & Releases

Sonnet 5 stays at $2/$10; the September hike is off.

Anthropic canceled the September 1 price increase for Claude Sonnet 5. At launch in late June, Anthropic set the model at $2 input and $10 output per million tokens and said rates would climb to $3/$15 on September 1. In August it confirmed the increase will not happen, and the $2/$10 rate holds for both the API and Claude Code sessions, per Anthropic's pricing docs.

Why it matters. Any team locking September budgets today (the hike was due tomorrow) keeps the launch rate with no contract scramble. It also reads as a direct price shot at GPT-5.6 Sol.

 
Tools & How-To

GitHub Copilot lets orgs publish custom agents to every repo.

In the August 28 Copilot for Visual Studio update, org and enterprise owners can publish custom agents across every repository under the account, and Visual Studio surfaces them in the agent picker with the org source attached. A new per-message effort setting (Low, Medium, High) sets how deeply a supported model reasons before it answers, a per-task cost-vs-accuracy dial. The Git agent can also review uncommitted changes and flag findings inline before you open a pull request.

Why it matters. Org-level publishing ships one custom agent to every developer without each person wiring it up. The effort dial is the same control OpenAI just added to ChatGPT, so it is becoming table stakes. Read the changelog or try Copilot.

 

ChatGPT adds a reasoning-effort slider and native Google Drive.

OpenAI shipped a per-message reasoning-effort slider for ChatGPT Plus and Pro, so you pick how much compute GPT-5.6 Sol spends before answering. It also added native Google Drive integration: keep Docs, Sheets, and Slides open alongside a chat, or point it at a whole folder for cross-file analysis. A new temporary chat mode carries per-session controls for memory and custom instructions.

Why it matters. The slider makes reasoning cost visible and tunable per message. Drive folder access is the request most Workspace users have had, since their documents already live there. See the release notes.

 
Quick Hits
• Okta Agent SSO is generally available. Agents become first-class identities in Universal Directory with short-lived tokens and the same access policies as employees, at no extra cost inside the core SSO product. Details →
• Cursor Cloud Agents drops the GitHub requirement. As of August 27 you can build in a standalone Cursor Origin repo, preview in the browser, and push to a Vercel URL with no source control connected, removing the main adoption blocker for GitLab and Bitbucket shops. Details →
• Anthropic's Compliance API exits beta. The API is now GA for Claude Code and Cowork, with transcript collection extended to Claude Science and Microsoft 365 sessions and admin SDKs in seven languages. Details →
• n8n August release targets the 2026-07-28 MCP spec. The update adds OAuth/OIDC auth, finer tool-call ordering, and parallel agent execution inside one workflow. Details →
 

How was today's brief?

★★★★★ Nailed it  |  ★★★ Solid  |  ★ Missed

Jonathan Hildebrandt Jonathan Hildebrandt
Co-founder and primary operator of Pondero. Writes the Pondero Brief.

Affiliate disclosure  ·  Unsubscribe  ·  Manage preferences

Pondero earns commissions on some links. This does not affect our editorial picks.

Don't miss what's next. Subscribe to Pondero AI:
← Newer Pondero Brief: OpenAI cuts Cursor off by November 12, migrate your model stack Older → Pondero Brief: AnyDoc parses 100x faster than Docling. Pick by format, not benchmark.
pondero.ai
Bluesky
LinkedIn
Twitter
LinkedIn
Powered by Buttondown, the easiest way to start and grow your newsletter.