AI Coding News: GitHub Copilot Lands in Slack & Teams, Claude Code Ships v2.1.239
Three tools moved fast enough this week to matter for anyone shipping code with an AI agent at their side. Between August 20 and August 23, 2026, Claude Code pushed one of its heaviest single releases of the month, GitHub Copilot broke out of the editor and into team chat for the first time, and OpenAI's Codex CLI shipped a proper dashboard for managing background agents. Here's what actually changed, and what to do about it.
Claude Code v2.1.239: cost overhauls and a Python SDK migrator
Anthropic pushed Claude Code to v2.1.239 on August 21, followed by two stability-focused releases on August 22 (v2.1.240) and August 23 (v2.1.241). The headline changes in v2.1.239:
- Cost estimates in /cost, the status line, and --max-budget-usd now factor in the 1.1x inference premium that applies to US-only data-residency workspaces, so budgets stop under-reporting spend for regulated teams.
- A new /claude-api upgrade command automatically migrates Python projects off the legacy anthropic 0.x SDK onto 1.x.
- Cloud sessions now sync plugins from claude.ai under a name@synced label instead of silently overriding a locally installed plugin with the same name.
- Windows finally gets cross-session messaging, matching what Mac and Linux users already had through ListAgents and SendMessage.
- A long tail of fixes: Bedrock streaming behind proxies that strip response headers, JetBrains terminal Edit/Write pauses, /resume losing custom titles past 64 KB of conversation, and MCP elicitation forms clipping in fullscreen mode.
If your team runs Claude Code on Bedrock, Vertex, or a data-residency workspace, both the cost-estimate fix and the SDK migrator are worth checking before your next billing cycle.
GitHub Copilot goes multiplayer: now live in Slack and Microsoft Teams
On August 21, GitHub moved Copilot's agentic features out of the IDE and into where teams actually talk, with two public previews landing the same day.
- Copilot in Slack: mention @GitHub in any channel to start an agent session that answers code questions, triages bugs, investigates failures in a sandbox, and opens a pull request linked back to the conversation. Dedicated Slack Code channels keep the work out of general chat.
- Copilot in Microsoft Teams: the same idea for Teams. @GitHub in a channel starts a cloud agent the whole team can watch and steer, including turning meeting action items straight into tracked work.
Both previews require a Copilot Business or Enterprise plan, draw on existing Copilot entitlements or AI credits, and let repository admins require extra approval before an agent-authored pull request merges, so human oversight stays intact even as the surface area expands. Worth testing on a low-stakes repo before enabling it org-wide.
OpenAI's Codex CLI adds an agents dashboard
OpenAI's Codex CLI hit rust-v0.149.0 on August 20, its most workflow-focused release yet:
- A new interactive agents dashboard lets you search, start, open, rename, and stop background tasks without leaving the terminal.
- /cd, /pwd, and /cwd manage the working directory mid-session; codex queue sends messages to already-running local or remote sessions.
- Vim mode picked up more real editing verbs (cw, c$, cc, character replacement).
- codex doctor now diagnoses endpoint protection, network/proxy failures, and update connectivity, useful for anyone running Codex behind a corporate proxy.
- On the security side, plugin isolation was hardened, and Codex now denies network access outright if a policy update fails to apply.
Quick hits: Cursor and OpenCode
Cursor and OpenCode stayed quieter this stretch. Cursor's most recent moves, the Origin code-hosting beta and faster cloud-agent builds, landed August 13-19 and are still rolling out to more paid plans. OpenCode's last confirmed release, v1.18.18, shipped August 13 with message-ordering and MCP/OAuth reliability fixes; no newer version had published as of this writing.
What it means for developers
The pattern across all three vendors this week is the same: agents are being pushed out of the terminal and into whatever surface developers already spend their day in, whether that's Slack, Teams, a dashboard, or a budget line. None of these are default-on changes yet, they're opt-in previews or point releases, but they signal where each vendor is placing its next bet: Anthropic on cost transparency and SDK continuity, GitHub on team-wide agent collaboration, and OpenAI on making background agents manageable at scale.
Sources & further reading
Comments
Share your thoughts and join the conversation
