Today’s radar brought a heavy package of new models — Gemini, transcription, multimodal capabilities — but what stayed with me was a launch outside the model world: an agent taking over an entire sales team’s CRM.

This week I saw a launch that captures where applied AI for business is heading, and I wanted to bring it here.

In brief

  • Salesforce and Anthropic launched Claudeforce, a plugin that brings the entire CRM into Claude.
  • It includes 37 ready-made sales skills, such as preparing a meeting, checking deal health and reviewing the pipeline, while querying and acting on live CRM data.
  • Salesforce did not create another parallel screen for AI: the agent enters the workflow the team already uses, without switching context. The market reacted quickly, and the stock rose nearly 20% after the announcement.
  • In credit products, the same logic points to automation inside origination, receivables management and portfolio monitoring.

Claudeforce: when the agent enters the CRM the team already uses

Salesforce and Anthropic launched Claudeforce, a plugin that brings the entire CRM into Claude. It is not a question-and-answer chatbot. It includes 37 ready-made sales skills: preparing a meeting, checking deal health, reviewing the pipeline—while querying and acting directly on live CRM data, without switching screens.

For anyone who wants to understand the launch in more detail, here is the analysis of Claudeforce.

What catches my attention is not the novelty of the agent. It is the model. Instead of creating another parallel system, AI enters the workflow the team already uses. The market reacted quickly: Salesforce shares rose nearly 20% after the announcement.

What this changes for credit products

On the credit product side, this is exactly the logic I defend. We do not need another screen for an analyst to open. We need automation that lives inside origination, receivables management and portfolio monitoring, querying real data and suggesting the next action.

This is the kind of case that helps move AI agents out of the demo category and into a process someone already needs to execute. Instead of asking only whether an agent can perform a task, it is worth looking at where that task happens, which data the agent queries and what action comes next.

That perspective is part of AI product management. AI agent examples become more useful when they help us see the complete workflow, not just the model’s isolated capability.

The standard for a good AI product

This changes the standard for what makes a good AI product in business. Less “look what the model can do” and more “how much time do I remove from the workflow of someone who makes decisions every day?”. It is a much harder standard to meet, but it is what actually moves results.

It will be worth watching how this type of agent embedded in CRM and other critical revenue systems matures over the next few months, because financial products will likely follow the same path.

The rest of the radar

Nvidia is negotiating to buy Hugging Face for $13 billion — changes control over the largest platform of open models and datasets used to build AI products. Read more

Gemini 3.5 Transcribe — Google’s new transcription model expands the options for speech-to-text features in products. Read more

Gemini Omni 1.1 Flash — a lighter, cheaper version of Google’s multimodal model and another cost-effective option for production. Read more

U.S. justice blocks the Pentagon’s ban on Anthropic — signals regulatory risk for AI suppliers in the public sector and could reopen government contracts for the company. Read more

“Small Models Have Arrived” — argues that small models already cover many real-world use cases, changing product architecture and cost decisions. Read more

Terminal-Bench-Science — a new benchmark for measuring the reliability of AI agents in multi-step scientific workflows. Read more

Show HN: Conduct, open-source guardrails for LLMs and MCP — addresses a growing product risk: agents calling tools outside their scope or in unsafe ways. Read more

Show HN: a model gateway that learns from usage — an open-source alternative to OpenRouter that routes model calls based on real usage to reduce cost. Read more

Bill Gates: “the turbulent era of AI has arrived” — a high-level view of investment pace, risks and strategic choices, useful for calibrating roadmaps and leadership messaging. Read more

That is all for today. Tomorrow, the radar returns with whatever changes by then.