AI work is leaving the chat window. Slack, SuperApp, Inc., and DeepSeek each announced products in August 2026 that push software development and collaboration toward persistent workspaces, shared records, and agents that operate inside team workflows.
Slack launched Slack Code on August 20, giving teams dedicated channels for building software with AI coding agents. People can tag agents such as Anthropic’s Claude Code or Cognition’s Devin in a conversation, and the agent then opens a project-specific code channel for the task.
Slack described the feature as working with agents the way teams work with colleagues in shared channels. Its own explanation is direct: “With Slack Code, when you have an idea or need to build a new feature, update a web page, or fix a bug, you simply tag in a coding agent like Anthropic’s Claude or Cognition’s Devin, and that agent then spins up a code channel to tackle the task.”
Slack Turns the Channel Into the Unit of Work
Slack Code channels are open and project-specific, with dedicated tabs for each participant. They archive when the work finishes, which gives teams a defined place for each coding task instead of forcing everything into one endless chat history.
The system integrates with AI agents available through Slack’s marketplace, including Claude Code, Devin, Vercel Agent, and GitHub Copilot. Slack Code is available on every Slack plan, including free workspaces, from its launch date—a rare moment when a new AI feature does not begin behind an enterprise paywall.
Slack Code treats the channel, not the chat window, as the unit of work. Code does not ship until a person in the channel approves it, and Slack maintains an audit log covering requests, changes, and approvals.
The feature rests on agent infrastructure Slack and Salesforce have shipped over more than a year, including a research preview Anthropic introduced on December 9, 2025. The practical goal is less mysterious than the surrounding AI theater: keep the agent’s work visible, reviewable, and attached to the team’s existing conversation.
SuperApp Makes the Conversation the Workspace
On August 17, 2026, Instabase changed its name to SuperApp, Inc. and launched SuperApp, an AI collaboration super app. The service runs on the web, iOS, Android, and desktop, with free access plus paid options for higher usage and enterprise deployments.
SuperApp combines features similar to Slack, AI models, and Google Docs in one live thread. It brings together teams and models from Anthropic, OpenAI, Google, xAI, and others, allowing people and AI to share the same conversation while keeping sources, feedback, and decisions connected.
“The conversation becomes the workspace,” SuperApp says. Anant Bhardwaj, its CEO, puts the pitch more bluntly: “The prompt box was a breakthrough. It made AI personal. SuperApp makes it collaborative. Work is multiplayer. AI should be too.”
SuperApp routes work across different AI providers, preserving model choice instead of tying an organization to one vendor. Its durable record connects messages, source files, model contributions, tool calls, results, revisions, and approvals—the paperwork of AI work, finally admitted into the room.
The company grew from Instabase’s experience in enterprise AI and moved toward collaborative AI in 2025. SuperApp migrated its workforce off Slack while preserving conversation history, moving more than 20 million historical messages; before release, it had more than 1,000 weekly active users, with roughly 90% outside the company.
SuperApp is backed by Andreessen Horowitz (a16z), Greylock Partners, Index Ventures, New Enterprise Associates (NEA), Spark Capital, Tribe Capital, and QIA. Sampriti Bhattacharyya, CEO of Navier, described the product this way: “The answers are genuinely thoughtful and rigorous — not fluffy. I organize my work in channels, carry history forward, and collaborate with my team on documents without losing the thread. So much of what I need is finally in one place.”
DeepSeek Adds Models and an Agent Harness
DeepSeek launched DeepSeek-V4-Pro and DeepSeek Harness v0.1 on August 13, 2026. V4-Pro targets agentic workloads and runs across DeepSeek’s web interface, mobile app, and API, with native support for the OpenAI Responses API and integration with Codex.
DeepSeek-V4-Pro has 1.6 trillion parameters, with 49 billion activated per token, and supports context windows of up to one million tokens. DeepSeek-V4-Flash has 284 billion parameters, with 13 billion activated, and supports the same maximum context size.
DeepSeek reports scores for V4-Pro-0813 of 87.9 on Terminal Bench 2.1, 74.1 on Toolathlon-Verified, 71.1 on DSBench-FullStack, and 67.2 on DSBench-Hard. The company says it tested V4-Pro-0813 using its upcoming Harness in “minimal mode,” a useful detail because benchmarks without test conditions are just decorative numerals.
DeepSeek Harness is an open-source developer preview under the MIT license and can be downloaded from GitHub. Its modular plugin system is designed for agent development, but DeepSeek warns that compatibility-breaking changes may arrive and that the harness is not yet a stable production platform.
DeepSeek also changed API pricing on August 16, moving to peak and off-peak rates. V4-Pro and V4-Flash cost more during peak times, adding a familiar constraint to the promise of unlimited machine labor: the agents may be ready, but the meter is still running.
Together, these launches point to a common shift. AI products are moving away from isolated prompts and toward shared channels, persistent records, model routing, review controls, and agents that participate in the same workspace as humans.
Based on
