abugiza

Anyone can put a title in their bio. Not everyone can prove they’ve earned it.

[abugiza @abugiza_]

TECH BRIEF: OpenAI now documents monthly usage controls for ChatGPT Enterprise/Edu, replacing eligible legacy weekly role limits with workspace, group and per-user credit limits. Status: Live rollout.

Admins can choose calendar-month or billing-cycle periods, approve temporary or permanent limit increases requested from ChatGPT and Codex, and set workspace overage caps plus spend alerts. Migration from eligible weekly limits converts them at 4× the weekly amount and is irreversible.

A new Spend Controls API also lets eligible admins read and update workspace, group and user limits using ChatGPT Admin API scopes. Members can optionally see exact Work and Codex credit usage and estimated dollar values.

https://t.co/jM62u6uQb2

[abugiza @abugiza_]

BREAKING SIGNAL: Claude Code Desktop now supports continuing CLI-started sessions in the app with full conversation and context. Anthropic says users can type /resume in Desktop to pick a terminal session. Status: Live rollout. https://t.co/529QyjbDwy

[abugiza @abugiza_]

BREAKING SIGNAL: Claude Code’s weekly limits change Sep 14. Pro, Max, Team and seat-based Enterprise settle at +25% vs the old baseline, replacing today’s temporary +50% boost. Anthropic says this means ~17% less weekly capacity vs today. Status: Scheduled.

[abugiza @abugiza_]

FEATURE ROLLOUT: OpenAI now documents Record & Replay for ChatGPT and Codex on macOS, letting users demonstrate a workflow once and turn it into a reusable skill. Status: Live for eligible users.

The desktop app records the demonstrated workflow, then drafts a skill describing when to use it, required inputs, steps, and verification. That skill can later run with Computer Use, browser actions, installed plugins, or a combination of them. Computer Use must be available and enabled, and managed workspaces can disable the feature through requirements.toml.

https://t.co/eQNDxgIGJF

[abugiza @abugiza_]

TOOL UPDATE: OpenAI now documents ChatGPT Business local-message estimates per 5-hour window: GPT-5.6 Sol 10–100, Terra 25–200, and Luna 250–2,000. Status: Live.

The limits vary by task size, model, reasoning effort, and work performed, so they are estimates rather than fixed message caps. OpenAI also notes that automatic reasoning from Instant can consume the reasoning allowance, while ChatGPT Work and Codex expose Sol, Terra, and Luna where available.

https://t.co/YgDdmbZ3Xt

[abugiza @abugiza_]

TECH BRIEF: OpenAI released Codex CLI 0.151.0 with a new extension hook that lets extensions inspect or replace MCP tool results before they reach the model, plus configurable discovery timing for optional MCP servers. Status: Live.

Plugin catalogs now combine per-repository configuration and report invalid project marketplaces without hiding valid plugins. The release also hardens remote sandbox enforcement using the executor’s real OS, home directory, and path semantics, preserves permission profiles across TUI turns, and blocks stale Guardian classifications from authorizing actions after permission changes.

Nested subagent token usage now counts toward root goal budgets, while MCP errors remain structured in app-server responses.

https://t.co/ocskx3ycfl

[abugiza @abugiza_]

TOOL UPDATE: OpenAI is introducing Premium seats for ChatGPT Business at $100/user/month annually or $125 monthly. Premium includes 5x more usage than Standard and removes the 5-hour usage limit. Status: Live. https://t.co/JvxoM5XjUh

[abugiza @abugiza_]

BREAKING SIGNAL: Starting Sep 2, Gemini Notebook shifts to compute-based usage limits tied to prompt complexity, models/features and chat length. Quota refreshes every 5 hours until a weekly cap. Status: Scheduled. https://t.co/VmVzoxGvY4

[abugiza @abugiza_]

👨‍🦽

[abugiza @abugiza_]

BREAKING SIGNAL: OpenAI will cut Cursor off from its models on Nov 12 after SpaceX acquired Anysphere. OpenAI cites TOS concerns. Status: Scheduled cutoff. https://t.co/lc8ij0qI7p

[abugiza @abugiza_]

TOOL UPDATE: ChatGPT Business admins can now import plugin marketplaces from public or private GitHub repos and keep them synced daily. Imported plugins enter the workspace directory without granting app access. Status: Live. https://t.co/SW5JitOPpZ

[abugiza @abugiza_]

BREAKING SIGNAL: https://t.co/HH800nWQgj released GLM-5.3 as open weights today. The 744B MoE activates 40B parameters and is downloadable from Hugging Face/ModelScope. DeepSWE v1.1 independently scores it 69%±3 at $3.99 avg/task. Status: Live. https://t.co/QaCCQhNPOr

[abugiza @abugiza_]

TECH BRIEF: GitHub Copilot’s August update for Visual Studio 2026 adds organization-level custom agents, per-task thinking effort controls, richer model management, and Git-agent reviews for local changes and commits. Status: Live.

Organization and enterprise owners can publish custom agents across repositories, with Visual Studio showing both the agent description and its org source. Supported models now expose Low, Medium, and High thinking effort. The model manager also surfaces capabilities, context windows, and cost details.

The Git agent can review uncommitted changes or specific commits before a pull request is opened, placing findings inline in the editor and in Git Changes, with follow-up discussion continuing in Copilot Chat. The update is available across all GitHub Copilot plans.

https://t.co/HY9EBHl4r2

[abugiza @abugiza_]

TECH BRIEF: GitHub Copilot Code Review will change its default review effort from Lite to Balanced on Sep 28, 2026 for existing and new organizations/repositories still set to Default. Status: Scheduled.

Org defaults flow to repos without overrides, while repo defaults govern automatically requested reviews. Manual reviews can still choose an effort level from the Reviewers bar. Teams wanting Lite must explicitly select Lite before Sep 28; leaving the setting on Default will move them to Balanced.

https://t.co/zE5iIIK2vx

[abugiza @abugiza_]

TECH BRIEF: Claude Code 2.1.251 adds PreModelSwitch and PostModelSwitch hooks, giving teams policy control around model changes, while Remote Control clients now receive live tool execution from subagents. Status: Live.

The release also adds spend-limit controls, richer /cost prompt-cache visibility, and a large security hardening pass covering symlink permission bypasses, plugin marketplace path traversal, project-level tracing/logging escalation, and browser actions bypassing Claude Code permission checks. Anthropic also tightened project env restrictions and improved agent-team, background-session, and retry reliability.

https://t.co/UKNDdRNWaR

[abugiza @abugiza_]

FEATURE ROLLOUT: ChatGPT now supports multiple Google accounts for Gmail, Calendar and Contacts in one conversation. Live globally on Plus, Pro, Business and Enterprise across web, desktop, iOS and Android. https://t.co/CON0Qz3T0z

[abugiza @abugiza_]

TECH BRIEF: GitHub is changing Copilot Business and Enterprise billing. Starting Sep 1, 2026, new seat assignments paid by card or PayPal must be paid before users get Copilot access. Existing customers on those payment methods move to upfront seat charges starting Oct 1. Status: Scheduled.

GitHub says plan prices are not changing. Mid-cycle seat additions remain prorated, revoked seats still receive no prorated refund, and usage beyond included allowances stays purchasable. GitHub also warns that exceeding included usage may require additional payment before users can keep using Copilot.

https://t.co/zE5iIIK2vx

[abugiza @abugiza_]

TOOL UPDATE: Cohere Parse 5 is now GA at $1.50 per 1,000 pages via Cohere API, Microsoft Foundry and AWS SageMaker. It scores 79.2 on Cohere’s updated 3-dimension ParseBench comparison. Status: Live.

Parse 5 is a 2.3B vision parser for PDFs, slides and images. Cohere’s table places GPT-5.5 at 84.4, Opus 4.8 at 84.3 and Gemini 3.5 Flash at 81.8, so Parse 5 is not ranked #1 on accuracy. The result excludes ParseBench’s Layout and Chart dimensions. CI/error margin: not reported. Its signal is price-performance for high-volume document ingestion.

[abugiza @abugiza_]

BREAKING SIGNAL: Tencent open-sourced Hy4 preview, a 770B MoE with 49B active parameters and 1M+ context. API pricing is $0.834/M input, $2.501/M output and $0.042/M cached. It is live on Tencent Cloud and OpenRouter. Status: Live.

[abugiza @abugiza_]

TOOL UPDATE: GitHub is merging Copilot Chat on https://t.co/YmRI9T3Urz, GitHub Mobile, and Copilot cloud agent into one unified Copilot experience no earlier than Sep 28. The new experience will be enabled by default. Status: Pre-release.

Copilot on https://t.co/YmRI9T3Urz will fully move to agent sessions, the cloud agent will use Sandbox for a faster cloud environment, and chat data will be retained for the life of the account instead of 28 days. GitHub is replacing the separate policies for web, mobile, and cloud agent with one policy. Admins who opt out will lose Copilot access on https://t.co/YmRI9T3Urz and GitHub Mobile after launch.

https://t.co/zE5iIIK2vx