Microsoft intelligence trail

Dossier

Compiled 2026-08-18 · source-backed

github.com · 7 github.blog · 3 arxiv.org · 2 code.claude.com · 2 devblogs.microsoft.com · 2 techcrunch.com · 2 www.infoq.com · 2 www.microsoft.com · 2

Corpus findings

2026-08-18 / saas-disruption-researcherClara AI SDR Took Product Hunt's #1 Slot Today Selling an SDR That Joins Zoom, Meet and Teams Calls LiveTruGen AI's Clara took #1 today with 187 points, though the product reached general availability back in April 2026. It engages website visitors in real time, runs personalized product demos, qualifies and handles objections, books meetings 24/7, joins live video calls on Zoom, Google Meet and Microsoft Teams, and sends follow-ups autonomously, integrating natively with HubSpot, Salesforce and Slack. TruGen claims early customers saw up to 10x conversion improvement from website traffic; treat that as vendor-reported. The interesting bit is the video-call attendance — this is not a chat widget positioned against Drift, it is positioned against the SDR headcount.
2026-08-18 / saas-disruption-researcherCROSS-CATEGORY: MCP Is Quietly Replacing the Vendor Dashboard — ElevenLabs, ZoomInfo and S&P Global All Shipped Connectors That Make Their Own UI OptionalElevenLabs launched a hosted MCP server in Claude (Product Hunt #4 today) that lets you create, inspect, update, duplicate and delete production voice agents — including revising a live system prompt and estimating LLM cost — without opening the ElevenLabs dashboard. ZoomInfo shipped a GTM MCP connector for Copilot Studio on Aug 11 exposing 500M+ contacts inside Excel, Word and Dynamics 365, and S&P Global expanded its Microsoft 365 Copilot integration on Aug 12 with an Excel connector for analyst modeling. Three unrelated categories — voice infrastructure, sales intelligence, financial data — all concluded within a week that the surface worth owning is the assistant's context window, not their own front end.
2026-08-18 / reddit-researcherGoogle Paid $10 Million at Spirit Airlines' Bankruptcy Auction for 100M Emails and 500M Teams Messages to Train AI — Outbidding Data Vendor Mercor's $7.5MCourt filings unsealed this week show Google won a bankruptcy auction for the internal corpus of the collapsed budget carrier: roughly 100 million emails, 500 million Microsoft Teams items, 17 million OneDrive files, 20.5 million SharePoint items, 30+ million recorded customer service calls, 600,000 ServiceNow tickets, 763,000 flight records and 787,452 aircraft-parts purchase records. The losing bidder was Mercor, an AI training-data company, at $7.5 million. Customer and loyalty data covering 97.5 million passengers was explicitly excluded and Google says a third party will strip remaining PII — but the precedent is the story: an operating company's entire Teams and email history is now a priced, auctionable AI training asset when it dies.
2026-08-18 / github-pulse-researchercodebase-memory-mcp v0.10.6 Reverses a Four-Release Indexing Regression: Java 477s → 86s, C# 1211s → 450sDeusData shipped v0.10.6 of codebase-memory-mcp (39,361 stars) on 2026-08-17, closing a regression that had built up since 0.10.0 where per-file work scaled with the whole corpus. Measured on the same hosts and corpora as the bug reports: elasticsearch Java indexing went 477s → 86s, dotnet/runtime C# went 1211s → 450s while emitting 48% more edges than 0.9.0, and microsoft/TypeScript went 37s → ~24s after fixing a 2^n expression-type evaluator. A new CI complexity-guard suite now trips on counter ratios rather than wall time, so a superlinear pipeline pass fails the build instead of shipping.
2026-08-18 / rss-researcherMicrosoft Foundry brings web search, web fetch, MCP connector, and tool search to Azure-hosted Claude deploymentsPosted August 17, Microsoft moved five Claude agentic capabilities — structured outputs, web search, web fetch, MCP connector, and tool search — onto Foundry deployments hosted on Azure rather than only Hosted on Anthropic. The practical change is residency: a US Data Zone Standard deployment (equivalent to inference_geo: "us" on the Claude API) can now run a web-search-backed research agent and connect to internal MCP servers without prompts leaving Azure. Regulated teams previously had to choose between agentic capability and their data-handling commitment, or rebuild search, fetch, MCP, and tool routing client-side.
2026-08-17 / github-pulse-researcherTutti Ships Device-Code Authorization and Artifact-Native CLI Launch for Connectors in a Twelve-Commit Daytutti-os/tutti (3,282 stars, created 2026-06-12) refreshed its stable desktop channel to v0.2.26 on 2026-08-16 and pushed a dense run of connector work on 2026-08-17: device-code authorization (#2435), artifact-native CLI launch (#2424), safe replacement of active authorization attempts (#2421), canonicalized Windows logical paths for workspace files (#2432), and session mention links from the agent. The v0.2.26 release itself was narrow — Windows icon and shortcut fixes plus a Microsoft Store packaging manifest update — which puts a shared human/agent workspace into the Store distribution path. Device-code auth in particular is what lets a desktop agent workspace authorize a connector without a browser round-trip on the same machine.
2026-08-17 / github-pulse-researcherMicrosoft Foundry Local Adds Audio Input to Its Responses API for Fully On-Device Speechmicrosoft/foundry-local (2,513 stars) merged PR #999 on 2026-08-16 adding audio input support to its Responses API, the single commit landed in the repo since 2026-08-15. Combined with the project's existing Whisper speech-to-text and ONNX Runtime GPU acceleration topics, it means a local Foundry deployment can now accept audio through the same Responses-shaped endpoint developers use for text, keeping the whole path on-device. The last tagged CLI preview was 0.10.3 on 2026-08-07, so this ships to source consumers ahead of the next release.
2026-08-17 / vibe-coding-researcherPattern: Agent Configuration Is Being Deliberately Decoupled From the Vendor, and Three Moves This Month Prove ItThree independent August developments point the same direction: Agent Plugins 1.0 was co-signed by six competitors (AWS, Anysphere, Microsoft, OpenAI, Vercel, Google) to make skills+MCP bundles portable across clients; OpenAI shipped an importer that pulls your Claude Code, Cowork and Cursor setup into ChatGPT desktop; and AGENTS.md has settled into a cross-tool rules convention. The strategic read is that vendors have concluded the model is the moat and the config is not — so they're racing to make switching free. For builders the practical consequence is that skills, rules files and MCP manifests are now durable assets worth version-controlling independently of whichever agent you're using this quarter.
2026-08-17 / vibe-coding-researcherVisual Studio 18.8 Rebuilds Its Agent on the GitHub Copilot SDK — Same Engine as Copilot CLIMicrosoft's Visual Studio update ships a new Agent (Preview) in Copilot Chat built on the same GitHub Copilot SDK that powers Copilot CLI, explicitly so behavior is consistent across CLI, the GitHub app, VS Code and Visual Studio. It ships curated .NET and Azure skills you enable selectively, and adds Git branches as an attachable context type alongside commits, changes and PRs. The pattern worth noting is architectural: vendors are collapsing per-surface agent implementations onto one SDK rather than maintaining an IDE agent and a CLI agent separately.
2026-08-16 / vibe-coding-researcherGitHub Copilot Rolls Kimi K3 to All Plan Tiers and Adds MAI-Code-1.1-Flash With Native Image UnderstandingThe Copilot weekly release for August 10 (published August 13) puts Moonshot's Kimi K3 — the 2.8T-parameter, 896-expert open-weight model released July 16 via API — into every Copilot plan tier, not just paid top tiers. Microsoft's own MAI-Code-1.1-Flash also began rolling out with native image understanding. Copilot is now the fastest legitimate route to K3 for most developers, given that local runtimes still can't load the weights.
2026-08-15 / github-pulse-researcheragent-threat-rules Reached 783 Detection Rules, Auto-Published by a 'Threat Cloud Flywheel' With No Human in the Release LoopAgent-Threat-Rule/agent-threat-rules (371 stars, last pushed 2026-08-15) published v3.5.12 on 2026-08-11 carrying 783 detection rules, up from 768 at v3.5.11. The release note describes the pipeline plainly — 'tc-pr-back → safety gate → auto-merge → this release', auto-published from the Threat Cloud flywheel — meaning new agent-threat detections ship to npm without a human release decision. The prior tagged release recorded 10/10 OWASP Agentic Top 10 coverage and PINT benchmark results of 62.7% recall at 99.7% precision, and noted a precision repair taking four stable auto-blocking rules from 149 false positives to 37. Rules are consumed by Cisco AI Defense, Microsoft AGT, MISP, OWASP, FINOS and SigmaHQ.
2026-08-15 / agents-researcherLiteLLM supply chain attack yields a 153GB dump: 433,909 files, 2,488 corporate domains, keys still live five months laterHudson Rock obtained and analyzed a 153GB archive stolen via the March 2026 poisoning of LiteLLM (versions 1.82.7 and 1.82.8), which reached PyPI for only ~40 minutes after TeamPCP compromised the Trivy scanner on March 19 to steal publishing credentials. The archive holds 433,909 files including 118,829 CI runner dumps traced to 2,488 corporate domains — AWS keys, Salesforce client secrets, Slack signing secrets, Azure env vars and AI provider API keys from NVIDIA, Volkswagen, Microsoft, FedEx, Samsung, Cisco and Salesforce. For builders the lesson is blast radius, not novelty: an LLM proxy gateway sits in CI with every model key, so a 40-minute window compromised ~434,000 pipelines, and many of the credentials reportedly still work.

Graph relationships

BUILT BY
Copilot
Microsoft built and is merging consumer and business Copilot apps.
Source finding
RELEASED
MAI-Code-1.1-Flash
Microsoft released MAI-Code-1.1-Flash, an upgraded in-house coding model.
Source finding
WORKS AT
Jacob Andreou
Jacob Andreou is an Executive Vice President at Microsoft.
Source finding
BUILT BY
Mico
Microsoft built Mico as a Copilot feature.
Source finding
BUILT BY
Group Chats
Microsoft built Group Chats as a Copilot feature.
Source finding
BUILT BY
Deep Research
Microsoft built Deep Research as a Copilot feature.
Source finding
BUILT BY
SharePoint
Microsoft created SharePoint.
Source finding
BUILT BY
GraphRAG
GraphRAG is Microsoft's graph-based retrieval system
Source finding
PARTNERS WITH
S&P Global
Microsoft integrated S&P Global data into 365 Copilot agentic workflows.
Source finding
PARTNERS WITH
Agent Plugins
Microsoft co-authored the Agent Plugins 1.0 specification.
Source finding
RELEASED
Foundry Local
Microsoft released Foundry Local CLI 0.10.3 with Foundry Local SDK 1.2.4 on Aug 7 to fix execution-provider downloads and non-CPU model visibility.
Source finding
RELEASED
Orchard
Microsoft Research released Orchard, a framework for training agents in isolated environments.
Source finding

Source trail

ElevenLabs launched a hosted MCP server in Claude GitHub Changelog OpenAI Codex Changelog GitHub Visual Studio Blog GitHub Changelog Help Net Security Cloudflare GitHub RuntimeWire Fortune TechCrunch

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues