AI Dev Pulse – 2026-09-05

September 5 is a distribution morning. Astra left the limited Daybreak lane and showed up where most developers already work, while OpenAI made the Cursor wind-down concrete with migration paths instead of another press release.

Treat today as a routing and policy day. Pick where Astra is allowed (Copilot policy, Work admin toggle, API spend), decide your Cursor fallback before November, and turn on content exclusions wherever agents can see the repo.

Top Stories

GPT-6 Astra is generally available in GitHub Copilot
Practical dev impact: If you are on Copilot Pro+, Max, Business, or Enterprise, select GPT-6 Astra in the model picker for long-horizon coding and agent runs without leaving Copilot. GitHub's September 4 changelog lists VS Code, Visual Studio, Copilot CLI, coding agent, Copilot app, github.com, GitHub Mobile, JetBrains, Xcode, and Eclipse, with gradual rollout. GitHub says internal testing favored how Astra plans, validates, and confirms results before finishing, with fewer steps on long tasks than prior OpenAI models. Billing is provider list pricing under usage-based billing. Business and Enterprise admins still gate access through the Copilot model policy; default enablement turns new models on unless that default is off or Astra is explicitly disabled.

OpenAI expands GPT-6 Astra to Pro and enterprise Work users, with API access live
Practical dev impact: Pro, Enterprise, and Business Premium seats can use Astra in ChatGPT Work and Codex now, and API callers can target gpt-6-astra without waiting on Cursor-managed routing. OpenAI's community update on September 4 confirmed the wider Work and Codex rollout and API availability, with Plus and standard Business still in the staged queue. Codex needs CLI 0.153.0 or newer. API docs price standard traffic at $10 / $50 per million tokens (cached input $1, cache writes $12.50), with a 1.05M context window, 128K max output, and tools including web search, file search, code execution, computer use, and MCP. Enterprise admins still control workspace enablement; Astra started off by default. Budget for faster allowance burn than GPT-5.6 Sol, especially with high reasoning effort or Fast mode.

OpenAI documents how to keep OpenAI models in Cursor after managed access ends
Practical dev impact: Teams that still rely on OpenAI inside Cursor should pick a path now, not in November. OpenAI's Help Center article (updated for the wind-down) says managed Cursor access is proposed to end November 12, 2026, and lists three local options: bring your own OpenAI API key for Chat and Agent, install the Codex IDE extension (ChatGPT sign-in or API key), or point Cursor at Bedrock, Azure, or another OpenAI-compatible gateway. BYOK and gateway credentials do not cover Cursor Tab, Auto, Cloud or Background Agents, Automations, the Cursor CLI, or Cursor's API and SDK. Those surfaces stay on Cursor-routed models. Pair that with OpenAI's earlier decision post: future models such as Astra were already withheld from Cursor-managed access during the transition.

Copilot JetBrains harness goes GA, and content exclusions reach the app and CLI
Practical dev impact: JetBrains users get the shared Copilot harness generally available, so agent workflows line up more closely with VS Code and Copilot CLI, while Business and Enterprise admins can finally trust content exclusion policies in the Copilot app and CLI, not only classic IDE chat. GitHub's September 4 weekly release notes call out the JetBrains harness GA and restate that the app and CLI now honor exclusion policies so excluded files stay out of agent context. The September 2 GA note covers Copilot Business and Enterprise. Same week, GitHub also flagged October 2 retirements for Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7, so map defaults while you enable Astra and Gemini 3.8 Flash.

Practical Impact Analysis

The through-line today is where Astra actually runs, and who still controls the path to OpenAI. Copilot GA and the Work or Codex expansion matter more than another benchmark table. If your org already pays for Copilot Business or Enterprise, the immediate move is policy: enable Astra where long agent jobs are allowed, keep spend alerts on, and compare it against Gemini 3.8 Flash and Claude Fable 5.1 on one shared task before you change the team default.

API callers get a clean id: gpt-6-astra, $10/$50, tools that match computer-use and MCP workflows. That is also the escape hatch for Cursor users who need OpenAI after November 12. Help Center is explicit that BYOK is local Chat and Agent only. Cloud agents, Tab, Auto, CLI, and Cursor's own API stay on Cursor routing. So the real decision is whether OpenAI work moves to Codex (extension or CLI 0.153.0+), stays in Cursor via BYOK for interactive sessions, or shifts to Copilot for the harness plus Astra picker.

IDC's September 3 read frames the same split as model-plus-harness competition: OpenAI deepening Codex, SpaceX pairing Cursor with Grok, Anthropic keeping Claude on Cursor. For an engineering manager, that is a procurement sentence, not a vibe. Write down which harness owns which repo class by November, enable Copilot content exclusions on every agent surface that can see secrets, and retire the October 2 Copilot models before CI pins them.

If you only do three things this morning: enable Astra in Copilot where policy allows, smoke-test gpt-6-astra with a real API key, and choose a Cursor fallback (BYOK, Codex extension, or Claude or Grok native) before the managed shutoff.

Tutorial

Smoke-test gpt-6-astra on the OpenAI API before you rewire agents. You need an OPENAI_API_KEY from the OpenAI dashboard. Confirm auth, model id, and latency with one short chat call. Do not paste a placeholder key.

1. Fail closed if OPENAI_API_KEY is unset.
2. Install the current openai package.
3. Call Chat Completions against gpt-6-astra with a short migration-planning prompt.
4. If the call succeeds, pin gpt-6-astra in one staging agent and compare token cost against your current default on the same prompt. If it 404s on the model name, your org is still waiting on API rollout.

: "${OPENAI_API_KEY:?export OPENAI_API_KEY from the OpenAI dashboard first}"
python -m pip install -U openai

python - <<'EOF'
import os
from openai import OpenAI

client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
resp = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[{
        "role": "user",
        "content": (
            "In 5 bullet points, outline a plan to migrate a Cursor team "
            "off Cursor-managed OpenAI models before 2026-11-12. "
            "Cover BYOK limits, Codex CLI 0.153.0+, and Copilot Astra."
        ),
    }],
    max_completion_tokens=400,
)
print(resp.choices[0].message.content)
print("model:", resp.model)
print("usage:", resp.usage)
EOF

Recommended AI prompt

Copy this paragraph into ChatGPT, Claude, Gemini, Grok, or whatever you use.

You are my staff engineer for AI tooling on 2026-09-05. GPT-6 Astra is GA in GitHub Copilot (Pro+, Max, Business, Enterprise) and wider in ChatGPT Work/Codex, with API id gpt-6-astra at $10/$50 per MTok and a 1.05M context window. Cursor managed OpenAI access is proposed to end 2026-11-12; Help Center options are BYOK for local Chat/Agent, the Codex IDE extension, or an OpenAI-compatible gateway, while Tab, Auto, Cloud Agents, Automations, Cursor CLI, and Cursor API/SDK stay on Cursor-routed models. Copilot's JetBrains harness is GA, content exclusions apply in the Copilot app and CLI, and four Copilot models retire 2026-10-02. Ask which harnesses we use, then produce (1) a one-page migration matrix (surface, OpenAI path after Nov 12, owner, risk), (2) a Copilot admin checklist for Astra plus content exclusions without surprise spend, and (3) a 30-minute eval comparing gpt-6-astra, Gemini 3.8 Flash, and Claude Fable 5.1 on one real repo task. Keep it concrete and copy-paste ready.