LoginSign Up
← Back to CRM
Weekly · CRM · Week of May 25, 2026

CRM splits in two: Salesforce makes Informatica the agent data layer while Streak adds citations to its inbox AI.

agentic aidata layerai citationsgmail-native crmbilling meterscomparison seo
O
By Onur Öztürk
LinkedIn ↗
Generated 1h agoDrawn from 11 products

The week in CRM

The CRM category split into two distinct postures this week. At the top, Salesforce relaunched Informatica as a headless, cross-cloud data foundation for AI agents — on AWS, Microsoft Foundry/Fabric, and Google Cloud simultaneously — and folded in autonomous data management agents plus what it bills as the industry's first unified agent and context catalog. The move is structural: Informatica stops being a captive Salesforce tool and becomes a horizontal substrate beneath every agentic workflow, including ones running inside competitor clouds. The MCP servers landing in AWS Agent Registry are the tell.

Underneath that, the mid-market and SMB CRMs continued to either ship AI primitives that close the trust gap (Streak's citations, Twenty's billing-v2 metering) or publish comparison content while the product surface stays quiet. Six of the eleven products this week produced no shipped releases at all — only editorial — which is itself a signal about where the category's energy actually concentrates.

Leaders

  • Salesforce News — The Informatica relaunch is the week's defining move in enterprise CRM. Salesforce is telling the market that the agent era's bottleneck is trusted data, and that it intends to own that layer regardless of where the agent runs. Agentforce Life Sciences crossing 140 customers (Chiesi, Moderna, Merck Animal Health) on the same week is the vertical proof point — regulated industries are where Salesforce's compliance posture beats horizontal AI platforms.
  • Streak — Citations land on every AI output: inline chips next to specific claims, a clickable source list pointing back to the originating email, meeting, note, or web page. Ask a Question and AI Autofill both get the treatment. Separately, full deal summaries and Q&A move into the Gmail sidebar at one credit per call. For a Gmail-native CRM, the bet is that auditability is what makes AI claims defensible inside a sales workflow — and that reps who never leave the inbox are reps who stay on the platform.
  • Twenty — Seven minor versions in roughly four weeks, with v2.7.3 shipping May 22 as the latest hotfix to the cross-version upgrade runner. The architectural commit of the cycle is v2.1.1, which moved AI credit-cap enforcement from per-workflow-step gating to the AI entry points themselves — agent execution, the generate-text REST handler, thread-title generation. Non-AI workflow steps now run when credits are exhausted; only the AI calls stop. Billing v2 is being metered where the cost lives.
  • Thryv — No engineering release notes, but a coherent narrative push: ImageAI for visual content, AI content optimization, and AI lead flow are now being named as first-class features inside an "invisible marketing" frame. The Great Training case study — 200 new clients, 10 hours saved per week — anchors the time-saved value prop. Vertical guides (HVAC, training/coaching) signal a deliberate move toward industry-tailored playbooks.

Wildcards

  • Recruiterflow — Heavy thought-leadership cadence around AIRA, the AI agent layer, anchored on original research (a 97-firm AI survey and the 2,100-firm Economics of Recruiting benchmark). The framing — separating AI experimenters from AI infrastructure builders, with Recruiterflow on the right side of the line — is interesting because it inverts the usual order: the thesis is being laid before the product proof. Worth watching whether AIRA-specific case studies follow.
  • Cognism — All editorial: CRM hygiene, deduplication, governance, plus a ZoomInfo head-to-head. The interesting move is the real-time customer data primer covering M&A, job moves, funding events, and team growth as trigger signals — reads as scaffolding for an upcoming feature rather than the feature itself. Wildcard for the data-quality wedge being pre-warmed.

Themes that compounded

  • The agent data layer becomes a category. Salesforce's Informatica relaunch makes "trusted data foundation for AI agents" a named market position. Cognism's data-quality content drumbeat — deduplication, governance, real-time signals — is the mid-market echo of the same thesis. Both are betting that agents without clean data are liabilities.
  • AI trust through provenance. Streak's citation chips are the operational version of what Salesforce sells as governance. The shared insight: a CRM AI that hallucinates a quote is a liability, while one that points to the exact email is an asset. Expect this pattern to spread.
  • Inbox as CRM surface. Streak puts deal summaries and Q&A inside Gmail sidebar; the AI Q&A now also reaches iOS/Android with voice dictation. The workflow keeps collapsing into where the work already happens, which for SMB sales is mostly email.
  • Billing as a real constraint. Twenty's v2.1.1 hotfix moved credit metering to AI entry points, and v2.4.2 fixed a workspaceId crash in the agent node — both fallout from billing v2. Streak prices Q&A and summaries at one credit per call. AI features are starting to look like utilities with meters, not subscriptions.
  • Editorial without ships. Six of eleven products this week published only marketing content. EngageBay returned from a 2.5-month publishing gap with HubSpot-targeted comparison pieces; Salesflare batch-published ten comparison pages on April 29 and went quiet; Onpipeline, Bitrix24, Insightly, and Cognism all produced positioning content with no product surface attached. For the mid-market, the comparison page is the product update.

Watch this week

The Informatica MCP servers in AWS Agent Registry are the load-bearing detail — they signal Salesforce is willing to be useful inside competitor platforms, which is the kind of move that either accelerates the agent-data-layer category or invites an immediate response from SAP-side competitors. Watch whether the next vertical Agentforce announcement lands in financial services or retail, and whether Streak's citation pattern starts showing up on competitor inboxes within weeks. Twenty's upgrade-path patches should also slow if v2.0's metadata cascade architecture is finally settling; another two or three hotfixes in the coming days would suggest it isn't.