Vertical packages, CRM-inside-Claude, and a Copilot credit cut — procurement this week is corpus plus permissions, with metering snapping back September 1

2026-08-24 · Thomson Reuters · Legal · Incumbent Saas · Hybrid

Thomson LLM + CoCounsel Legal Tabular Analysis

Closed-deployment Thomson LLM (Qwen3.5-397B derivative) on ~$40M spend; Thomson-1.0-Small Qwen3.6 derivative on Hugging Face; CoCounsel retains Claude Agent SDK for agentic workflows

Legal procurement should score corpus access and workflow governance before model branding — Thomson's closed-deployment model handles tabular analysis while the agentic layer stays multi-model, and both Thomson tiers are open-weight derivatives per the technical report. Editorial forecast: lineage disclosure may become standard in legal AI RFPs by mid-2027; the falsifier lives in the Weekly synthesis (two top-tier vertical platforms go fully proprietary at the frontier with no open-weight base disclosed, by Q2 2027).

Sources Thomson Reuters

2026-08-25 · Google · Finance · Incumbent Saas · Seat Based

Gemini Enterprise for Financial Services

Google-managed Financial Research agent with 50+ foundational skills and MCP connectors to licensed market data

Finance leaders on Google Cloud should pilot vertical package before assembling custom MCP stacks — packaged skills and connectors are the procurement unit, not raw Gemini API. Compare connector list against internal data entitlements before GA commitment.

Sources Google Cloud

2026-08-25 · Google · Legal · Incumbent Saas · Seat Based

Gemini Enterprise for Legal

Domain skills for contract review, DSAR fulfillment, regulatory horizon scanning; connectors to iManage, NetDocuments, RelativityOne, Harvey, Legora

Legal ops on Google stack get vertical connectors without Harvey/Legora displacement — evaluate whether packaged skills cover your matter types before buying point solutions. Closed orchestration on Gemini, not open-weight routing.

Sources Google Cloud

2026-08-26 · Salesforce + Anthropic · Sales · Incumbent Saas · Seat Based

Salesforce in Claude (Claudeforce)

37 prebuilt sales skills with live CRM read/write through AIforce MCP harness; admin connects once with authentication and permissions managed centrally and actions routed through Salesforce business rules; open beta September 2026

RevOps should pilot Claudeforce as centrally administered CRM access with Salesforce enforcing business rules on every write — the press release describes one admin connection and central permission management, not per-user identity inheritance, so confirm seller-level attribution in audit logs during the beta. Vertical-AI sales tools must route actions through the system of record or lose auditability.

Sources Salesforce

2026-08-24 · Caddi · Operations · Startup · Seat Based

Caddi agent-that-builds-agents

Discovers repetitive back-office processes across 100+ apps; Loop Studio captures workflows; targets law, wealth, insurance back offices

Ops leaders in regulated back offices should evaluate agent discovery before greenfield automation — deterministic steps with AI only at judgment points reduce compliance risk versus full LLM loops. Team $250/user/month; Business $6,500/month.

Sources Caddi

2026-08-24 · DeepCura · Support · Startup · Hybrid

DeepCura Electronic Health Workforce (EHW)

Supervised multi-agent clinical ops — reception, scribing, scheduling — from $129/provider/month with usage credits; 6,000+ clinicians

Healthcare CIOs should compare EHW all-in pricing against point-solution stacks — vendor claims up to 10× software cost reduction versus full stack. Metered credits on top of subscription require spend caps like Copilot.

Sources DeepCura

2026-08-28 · Owner · Commerce · Startup · Outcome Based

Owner AI-native local business platform

$240M raise at $2.3B valuation past $100M ARR — autonomous agents as AI CMO/CTO for restaurants and local businesses

SaaS investors should track local-commerce agent capture as diffuse returns against hyperscaler capex — Owner projects $1B+ restaurant sales through platform in 2026. Incumbent POS vendors face agent-native replacement pressure.

Sources Owner

2026-08-25 · Google

Gemini Enterprise verticals (FinServ + Legal)

First purpose-built vertical industry packages atop Gemini Enterprise with 50+ skills and domain connectors

Incumbent counter-attack is packaged orchestration plus data connectors, not new base models. CIOs on Google should consolidate vertical agent procurement under Gemini Enterprise rather than parallel Harvey/Legora contracts where connectors overlap.

Sources Google Cloud

2026-08-26 · Salesforce

Claudeforce / Salesforce in Claude

Bidirectional partnership — Claude as default across Agentforce; CRM embedded in Claude via 37 sales skills and MCP harness

Salesforce bets users will work in Claude for CRM tasks while Agentforce runs on Claude internally — dual surface, one centrally administered permission model with business rules enforced in Salesforce. Require system-of-record rule enforcement and per-action attribution in any competing sales agent RFP.

Sources Salesforce

2026-08-25 · Microsoft

AI at Work Roadmap

Dynamics 365, Power Platform, Dataverse join continuous AI at Work Roadmap; Release Planner retires by November 15

Enterprise planning cycles must shift from twice-yearly release waves to continuous agent feature disclosure — procurement SLAs need rolling acceptance criteria, not wave-based UAT gates.

Sources Microsoft

2026-08-24 · Thomson Reuters

Thomson LLM + CoCounsel

~$40M proprietary legal LLM plus Qwen3.6 derivative open weights; multi-model agentic layer via Claude Agent SDK

Legal incumbent thesis: continual learning on proprietary corpora beats renting frontier APIs for high-volume structured work — but foundation is semi-open. Legal AI buyers should audit lineage and license terms on every 'sovereign' claim.

Sources Thomson Reuters

2026-08-28 · GitHub

Copilot billing

Promotional credit pools expire Sep 1 — included credits cut 37–44% at unchanged seat prices; overage enabled by default

Largest developer incumbent reasserts metering after promo period — engineering leaders must set spend caps today or route high-volume agent loops to BYOK alternatives before September overage bills land.

Sources GitHub

2026-08-28 · Commerce · $240M at $2.3B valuation

Owner

$240M raise at $2.3B valuation past $100M ARR — AI-native ops platform for local businesses (restaurants first)

Diffuse application-layer returns while hyperscaler capex concentrates — see weekly capitalFlow Hyperscaler-Hosted. Local-commerce agent platforms can scale ARR without GPU queue position. SaaS investors should track agent-native ops as a category distinct from horizontal copilots; incumbents in POS and local marketing face replacement pressure at outcome-based pricing.

Sources Owner

2026-08-24 · Operations

Caddi

Launch of agent-that-builds-agents platform for law, wealth, and insurance back offices across 100+ app connectors

Back-office agent factories compete with UiPath Maestro step-type orchestration (W34 established versioned step-type orchestration as enterprise default) — buyers should pilot discovery-first automation where repetitive workflows are undocumented before buying horizontal copilots.

Sources Caddi

2026-08-24 · Support

DeepCura

Electronic Health Workforce GA at $129/provider/month with usage credits — 6,000+ clinicians on platform

Clinical multi-agent ops at seat-plus-metered pricing is the healthcare analog to GitHub's Copilot credit cut — HIPAA buyers should model usage credits and subscription separately and cap agent spend before scaling reception and scribing agents across a health system.

Sources DeepCura

2026-08-25 · Finance

Fuse

Fuse Terminal GA — Bloomberg-style commercial insurance workspace with 65+ API endpoints; Fuse Pro $999/month for three seats

Vertical intelligence workspaces ship with never-gated APIs — insurance ops teams can build agents on canonical data without a separate API tier procurement. Competes with horizontal agent layers by owning the data model first; evaluate Terminal before assembling MCP connectors to fragmented insurance feeds.

Sources Fuse

2026-08-28 · Hybrid → Hybrid

GitHub

Promotional Copilot credit pools expire September 1: Business included credits fall 37% to 1,900 per user per month and Enterprise 44% to 3,900 at unchanged $19 and $39 seat prices, with additional usage enabled by default. Effective agent allowance per seat drops sharply — engineering leaders must set spend caps or route high-volume loops to BYOK stacks before Q4 planning locks copilot-dependent workflows.

Sources GitHub Docs

2026-08-26 · Usage Based → Usage Based

Z.ai

GLM-5.3-Flash list API at $0.15/$0.50 per million tokens with 50% launch promo through September 9 follows a six-day free Ox Alpha stealth period — steady-state agent economics remain unproven after promo expiry. Renegotiate any agent contract priced off stealth-week throughput.

Sources Z.ai

2026-08-26 · Usage Based → Usage Based

Alibaba

Qwen3.8-Flash production API at $0.16 input / $0.47 output per million tokens with 1M default context per Alibaba's release — hosted million-context agent API at sub-frontier list rates, separate from the Flash-Next open checkpoint SKU.

Sources Alibaba Cloud

2026-08-26 · Seat Based → Usage Based

Workato

Otto for Everyone extends autonomous superagents to self-serve individuals on credit-metered pricing with 500-credit free trial (see startupSignals) — superagent category defaults to credits rather than seats outside enterprise Workato tenants, signaling metered autonomy as the SMB entry price.

Sources Otto by Workato

Vertical winners bundle corpus access, platform-governed orchestration, and open-weight foundation derivatives

Regulated vertical moats consolidate on proprietary corpus plus orchestration atop open-weight derivative foundations — procurement should ask foundation lineage and how permissions and attribution are enforced in every vertical RFP. Editorial forecast (not week-35 observable): procurement may require lineage disclosure by mid-2027 — same falsifier and polarity as the Weekly synthesis connection on vertical moats.

Examples

Thomson LLM on Qwen3.5-397B and Thomson-1.0-Small on Qwen3.6, Claudeforce centrally administered MCP with Salesforce business rules, Gemini Enterprise vertical connectors

Sources Thomson Reuters, Salesforce, Google Cloud

Metering reasserts after promotional agent credit periods

Seat price stability with shrinking included agent credits is the incumbent SaaS response to Jevons paradox (cheaper inputs lifting total consumption) on tokens — CFOs should model post-promo agent cost separately from seat renewals.

Examples

GitHub Copilot Sep 1 cut, GLM-5.3-Flash promo Sep 9, Workato credit Otto

Sources GitHub, Z.ai, Workato

Always-on roadmaps replace release-wave enterprise planning

Agent features ship weekly — procurement and change management must move from semiannual UAT gates to rolling acceptance with versioned skill artifacts.

Examples

Microsoft AI at Work Roadmap, Continuous Dynamics disclosure

Sources Microsoft

Sep 1

GitHub Copilot promotional credits expire

37–44% included credit cut at unchanged seat prices — configure spend caps or route to BYOK before overage defaults bite.

Sep 2026

Salesforce Claudeforce open beta

First live CRM-in-Claude under a centrally administered connection with Salesforce business-rule enforcement — RevOps pilot window to test per-seller attribution.

Sep 9

GLM-5.3-Flash API promo ends

Steady-state pricing for open-weight agent routing — rerun vertical agent business cases.

Nov 15

Microsoft AI at Work Roadmap becomes primary D365 disclosure

Release Planner retires — enterprise change management must adapt to continuous agent releases.

Q4 2026

Commerce BIS remote-GPU rule consultation

Southeast Asia colo operators serving Chinese tenants face compliance cliff — affects cross-border inference routing.

• Big read centers corpus-plus-permissions vertical thesis and Sep 1 Copilot metering cut against Hot Chips noise.
• Claudeforce and Gemini Enterprise verticals logged as incumbent platform-governed orchestration pattern.
• Pricing shifts section carries four rows — Copilot, GLM-5.3-Flash, Qwen3.8-Flash, Otto credits.
• Revision 3: Claudeforce restated as a centrally administered connection with Salesforce business-rule enforcement (the Aug 26 press release does not claim per-user identity inheritance); Thomson LLM lineage corrected to a closed deployment of a Qwen3.5-397B derivative; Thomson lineage falsifier aligned to the weekly polarity; Qwen3.8-Flash input price carried at $0.16.