Claude in Slack Retires Today — Say Hello to Multiplayer Claude
The old per-user Claude in Slack app goes offline today, August 3, closing the migration window Anthropic opened when it launched Claude Tag in public beta on June 23. Workspaces that didn’t opt in are migrated to default settings automatically as the legacy app shuts down.
What replaces it is a different animal. Claude Tag gives a team one shared @Claude per channel — a persistent, multiplayer teammate rather than a per-person bot. Running on Opus 4.8, it retains shared context across people and time, builds memory, can watch a channel proactively in ambient mode, and executes multi-step tasks asynchronously over hours or days — not just when it’s @-tagged. It’s the first properly multiplayer Claude: everyone on the team steers the same instance, and the whole thread sees the work.
Snowflake and Anthropic Deepen the Governed-AI Partnership
Snowflake and Anthropic are pressing their bet that enterprises want AI to run where governed data already lives. Building on the December 2025 tie-up — reported at the time as a $200M partnership — Claude now powers Snowflake Cortex Code (the fastest-growing product in Snowflake’s history) and Snowflake Intelligence, with Block, eSentire, Indeed, and Carvana running it in production.
The pitch is governance, not novelty: keep security, compliance, and lineage controls intact by moving reasoning to the data rather than shipping data out to a model. Snowflake is also one of six launch partners in Anthropic’s Claude Marketplace, letting customers apply existing Anthropic spend commitments toward Snowflake AI services — smoothing procurement for regulated buyers.
The MCP 2026-07-28 Spec Lands — Stateless Core, Tunnels, and Extensions
The newest Model Context Protocol spec is coming to Claude, and it’s the biggest structural change since MCP shipped. The headline: MCP moves from a bidirectional stateful protocol to a stateless request/response model, so connectors can finally deploy on serverless and edge infrastructure instead of holding long-lived connections.
Around that core: MCP Apps and Tasks now ship under a versioned extensions framework, giving a formal path to interactive UIs and long-running work without touching the core protocol. MCP Tunnels (research preview) connect Claude to servers inside a private network — no public endpoints, no inbound firewall rules, no IP allowlisting. And connector builders get a developer observability dashboard to track adoption, latency, and errors across Claude surfaces. With MCP now past 400M monthly SDK downloads, this is the plumbing hardening for scale.
Experimental Prompt Tools APIs and the Workbench Sunset August 17
A housekeeping deadline for builders: Anthropic is retiring the experimental prompt tools APIs on August 17, along with the Workbench. If any of your tooling calls the prompt-generation or prompt-improvement endpoints — or your workflow leans on the Workbench for prompt iteration — now is the time to migrate off before the cutoff.
It’s a small deprecation, but a reminder that the platform is pruning as fast as it ships. The experimental prompt helpers were always labeled as such; folding them away as the Console matures signals Anthropic pointing developers toward the stable, documented surfaces rather than the lab bench.
The Sonnet 5 Promo Clock: Four Weeks Left
A standing reminder as the calendar turns to August: Claude Sonnet 5’s promotional $2 / $10 per million tokens pricing ends August 31, with standard $3 / $15 rates starting September 1. On list price that’s a 50% jump on input and output.
The subtler cost is the tokenizer: Sonnet 5 maps the same content to 1.0–1.35× more tokens than earlier versions, so an identical workload can get more expensive than the headline change implies. With four weeks left, teams running Sonnet 5 in production should re-benchmark real token counts now rather than trust old estimates when the September bill arrives.
Claude Code’s 50% Weekly Usage Boost Runs Through August 19
Heavy users still have headroom: Anthropic’s temporary 50% weekly usage boost for Claude Code subscribers remains in effect through August 19. It’s extra runway on the weekly limits before the boost expires — worth planning bigger agentic sessions around while it lasts.
The pattern is familiar by now: agentic coding burns tokens far faster than chat ever did, and usage caps have become the real friction point. Anthropic keeps nudging the ceiling up rather than let limits be the reason a developer drifts to a competitor — a cheap way to keep the most demanding users inside the tent.
The Agent Grows Up: Shared, Persistent, and Close to the Data
Three of today’s stories point the same direction. Claude Tag turns Claude from a per-user bot into a shared teammate that remembers, watches, and works asynchronously. The MCP spec goes stateless so connectors can live on serverless infrastructure and reach into private networks through tunnels. And Snowflake’s pitch is to run reasoning where governed data already sits. Different surfaces, one theme: the agent is being re-architected to persist, to be multiplayer, and to sit next to the systems of record.
That’s a meaningful shift from the 2025 mental model of an AI you prompt one turn at a time. A shared Claude in a channel is an organizational actor, not a personal tool — which is exactly why the enterprise plumbing (managed auth, tunnels, governed data, observability) matters more than another benchmark. The interesting question is governance: when one agent holds a team’s context and acts over days, who owns its memory, its permissions, and its mistakes? The capability is arriving faster than the answers — and the labs that ship the controls alongside the power will be the ones regulated buyers actually deploy.