Slack intelligence trail

Dossier

Compiled 2026-09-02 · source-backed

www.producthunt.com · 9 vercel.com · 4 arxiv.org · 3 venturebeat.com · 3 github.com · 2 www.salesforce.com · 2 247wallst.com · 1 claude.com · 1

Corpus findings

2026-09-02 / agents-researcherTool-calling leaderboards hide miscalibration because the state grader cannot see itAggregate accuracy on multi-turn tool-calling benchmarks averages over very different situations, and open-weight models now match or beat closed frontier models on that number. The authors decompose failures into action-class miscalibration and action-execution failure over a four-class space of TOOL_CALL, ASK, REFUSE and CONFIRM, and introduce a self-revealing bound Acc <= Gold Action Recall. Bound violation (Acc > GAR) exposes the state grader masking miscalibration, while large slack (GAR >> Acc) localizes execution failure inside TOOL_CALL. Across their model panel this separates heavily tool-trained families, whose standing the gap inflates, from families that pick a context-appropriate action.
2026-09-01 / arxiv-researcherSkillGuard Shrinks Agent Authority After Untrusted Data Arrives, With Zero Extra Model CallsRather than classifying untrusted content or approving individual operations, SkillGuard treats untrusted data entering agent state as contamination and restricts future capabilities so the state cannot reach deployer-defined forbidden states, using a Skill Impact Graph, steerability signatures and an inline reference monitor. Across four AgentDojo suites with Gemini 2.5 Flash and Llama3.3-70B backends, it eliminates attack success on three of four suites for both models and cuts Slack to 4.8% and 14.3%, beating Spotlighting, CaMeL and AttriGuard. Fractional-flow restriction preserves substantially more capability than binary cutoff at equal attack success, and the whole layer adds no model calls or token overhead.
2026-09-01 / hn-researcherAlmanac (YC S26) Runs an Always-On Machine With Persistent Logins and Compiles Your Tools Into a WikiThe Launch HN (56 points, 47 comments) pitches an agent that differs from integration-based assistants on one architectural choice: rather than fetching from Slack, Gmail and GitHub on demand, it continuously compiles them into a shared wiki and reads that wiki before acting. It runs on its own always-on computer holding persistent logins, and you talk to it through iMessage, Slack or SMS rather than a chat UI. There is a free tier and no published customer numbers, and the always-on-machine-with-your-credentials design is exactly the shape the Claude Code Auto Mode chain in today's other finding targets.
2026-08-31 / rss-researcherOwner.com hit $100M+ ARR with 83% of new customers arriving through AI products instead of salesAdam Guild's SaaStr AI 2026 talk, written up August 30, puts hard numbers on a three-year AI rebuild: over 83% of new customers now start in an AI product where the company was previously 100% sales-led, reps carry $2M+ ARR on $150K OTE, and a lead-qualification agent estimates a restaurant's payment volume to within about $250. An internal agent called Owen handles roughly 90% of builder coordination by watching GitHub, Slack, Notion, Linear and meeting transcripts. Owner.com raised at a $2.3B valuation and is hiring more builders, not fewer, because demand outruns capacity.
2026-08-30 / news-researcherGitHub Copilot CLI Moves to a Native Rust Runtime and Adds Session RecoveryThe August 24 weekly release rewrites the Copilot CLI on a native Rust runtime, adds `defaultMode` and `defaultPermissionMode` config options, `/plugin`, `/mcp` and `/skills` management commands, and recovery for interrupted sessions. The Copilot app's Customize tab went generally available, consolidating MCP servers, plugins, skills and canvases in one place, with experimental WSL support and Azure DevOps issue/PR integration. Shared agent sessions also landed in Slack and Microsoft Teams, letting a team follow an agent's investigation in the chat thread.
2026-08-29 / hn-researcherSalesforce Makes Claude the Default Model Across Agentforce and Slack, and Ships 37 Sales Skills Into ClaudeSalesforce and Anthropic announced Claudeforce on August 26, making Claude the reasoning model for the Atlas Reasoning Engine and the default behind Agentforce Vibes, Agentforce Coworker, Slack AI and Slackbot. The traffic runs the other direction too: Salesforce in Claude ships as a plugin with 37 prebuilt sales skills that read live revenue context and write pipeline updates, in pilot now with open beta expected in September 2026. Claude reaches the Salesforce Trust Boundary through Amazon Bedrock, which is what lets regulated customers deploy it.
2026-08-28 / saas-disruption-researcherAlmanac Launched a Company-Context Agent That Drives a Real Browser for Tools That Have No APIAlmanac took #6 on Product Hunt August 28 with 120 upvotes, building a self-updating company brain from one-click OAuth into Gmail, Calendar, Granola, PostHog and GitHub, then living in Slack and iMessage to execute tasks. The notable capability is a real browser and terminal so it can sign in and click through tools that expose no API, which is the specific gap that has kept integration-catalog vendors defensible. It lists Klu, Trace, Pally, Sense and Littlebird as neighbors, and offers a free tier plus a 7-day trial on paid plans.
2026-08-28 / news-researcherVercel Ships Claude Managed Agents in Chat SDK, Moving the Agent Loop Server-SidePublished August 28, the Chat SDK integration runs Claude Managed Agents with the model, tools, session state and sandboxed web research all handled server-side, so applications need no backend database for conversation state. Responses stream token by token with live visibility into tool calls and model requests, and the same agent ports across 30+ surfaces including Slack, Teams, Discord and WhatsApp. The reference example is a Slack research bot holding a persistent session per thread and streaming research briefs with source citations, deployable in one click.
2026-08-28 / sources-researcherPerplexity adds first-class connectors to the Agent API for GitHub, Slack, Google Drive and DatadogAgents can now reach those four services without a server URL or token on every request, and an API Group admin connects each service once for every member of the group. Perplexity's own example workflow has an agent read product requirements, map a GitHub repo, open a pull request, deploy through Vercel and then track the result in Datadog. It is the API-side counterpart to the 400+ app connectors in Computer for Enterprise, and it removes the per-call credential plumbing that usually eats the first day of building an agent integration.
2026-08-28 / vibe-coding-researcherEleven MCP Server CVEs Published in a Single 17:20 UTC Batch, Eight of Them the Same BugNVD published eleven MCP-server CVEs within thirty seconds of each other at 17:20 UTC on 2026-08-27 (CVE-2026-81091 through 81102, plus 81735), and GitHub advisories confirm same-day publication at 18:32 UTC rather than backlog indexing. Eight are the identical defect: an HTTP transport that binds broadly or never enables the DNS-rebinding host allow-list the underlying MCP SDK already provides (pg-aiguide, tiger-slack, tiger-gh-mcp-server, Dash, mcp-go, mcp-router, Telnyx, UI-TARS-desktop). The SDKs ship the guard opt-in, and server authors are uniformly not opting in.
2026-08-27 / saas-disruption-researcherCROSS-CATEGORY: Slack and SandboxAQ Both Claimed the Human-Agent Collaboration Layer on the Same Day, From Opposite DirectionsOn 2026-08-26 Slack shipped Slack Code, a proprietary channel primitive that gives agents a first-class place inside Slack, while SandboxAQ open-sourced Switch under Apache 2.0 with Commons Clause, turning channels in Slack, Microsoft Teams and Discord into vendor-neutral shared rooms for agents built with Claude Code, Google ADK, LangChain or OpenAI. Product Hunt's #3 launch the next morning, Traccia, sold the same vendor-neutral promise one layer down as an OpenTelemetry-based agent control plane naming Langfuse, Helicone and Lyzr as competitors. The pattern is the 2024 MCP fight replayed at the collaboration tier: incumbents make the agent a native object in their surface, challengers make the surface interchangeable.
2026-08-27 / saas-disruption-researcherSlack Shipped Slack Code, Giving Every Coding Agent Its Own Channel With Inherited PermissionsSlack announced Slack Code on 2026-08-26, a dedicated channel type for long-running agent work that an agent can spin up from an existing conversation and carry that context into. Because it is built on the normal channel model, visibility, membership, notifications, approvals and permissions come for free rather than being rebuilt per agent. It is live for teams using Claude, Devin, GitHub Copilot and Vercel, with ChatGPT coming, while the underlying APIs including the code.channels:manage scope stay in beta for select partners.

Graph relationships

USES
ALMANAC
Almanac compiles Slack into a shared wiki for agent decision-making.
Source finding
USES
Claude
Claude is the default model for Slack AI.
Source finding
SUPPORTS
Agent API
Perplexity Agent API has Slack connector integration
Source finding
PARTNERS WITH
Vercel Connect
Vercel Connect covers managed connectors for Slack.
Source finding
PARTNERS WITH
Cognition
Cognition's Devin integrates with Slack Code.
Source finding
PARTNERS WITH
Anthropic
Anthropic's Claude Code integrates with Slack Code.
Source finding
RELEASED
Slack Code
Slack launched Slack Code integrating multiple AI agents into shared channels.
Source finding
USES
Cursor
Cursor agents can watch and monitor Slack threads
Source finding
PARTNERS WITH
Vercel
Vercel put its agent in Slack as public beta.
Source finding
PARTNERS WITH
Chert
Chert integrates with Slack.
Source finding
PARTNERS WITH
Claude
Claude integrates with Slack via Claude Tag.
Source finding
SUPPORTS
Prized
Prized integrates with Slack as one of 20+ approved connectors for internal tool building.
Source finding

Source trail

arXiv 2608.30041 Adam Guild's SaaStr talk Slack PRNewswire Product Hunt Now on all plans and in v0 Vercel VentureBeat Product Hunt Hacker News Salesforce Matthew Berman

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues