The thread today was money moving toward proven results, not promises. I started there and left the rest of what passed the filter below.

One thing I learned from working with credit products over the past few years is that serious money only moves toward technology when it proves results, not promises.

That is what caught my attention in a story this week. Vertical AI-agent startups raised US$1.8 billion in July, across more than 12 rounds.

What stands out is not the volume, but where the money went. Harvey AI, focused on legal work, raised US$200 million with US$35 million in ARR. Glean, focused on enterprise search, raised US$180 million. And Hebbia, focused on financial analysis, raised US$130 million. It is already present in 15 of the world’s 20 largest banks.

In brief

  • Vertical AI-agent startups raised US$1.8 billion in July, across more than 12 rounds.
  • Harvey AI, Glean and Hebbia raised US$200 million, US$180 million and US$130 million respectively; Hebbia is already present in 15 of the world’s 20 largest banks.
  • A well-designed AI agent with governance and an audit trail can accelerate credit workflows without taking the decision away from the person accountable for it.
  • The signal for people building products in finance is to target one specific problem and prove results, not launch another generic assistant.

This says a lot about where AI is moving from beautiful pilots to real operations. When 15 of the 20 largest banks on the planet run an agent for day-to-day financial analysis, it is not hype. It is because the result showed up and the customer paid again.

On the credit-product side, this maps directly to what we see every day. Receivables analysis, deal structuring and collateral reconciliation involve scattered information, large volumes of documents and a lot of manual cross-checking. That is exactly the kind of task where a well-designed agent, with AI governance and an audit trail, delivers a real speed gain without taking the decision away from the person accountable for it.

The signal for the coming months is clear. Investors are prioritizing specialized agents with proven ARR, not another generic assistant that promises to do a little bit of everything. For people building AI products in finance, it is time to target one specific problem and show results. That is part of AI product management, where technology needs to connect to an operation and a value metric.

If you are curious and want to read the full story, here is the analysis of AI-agent startup funding.

The rest of the radar

Anthropic launches Claude Opus 5 — a new cost-performance model changes the price-quality equation for products built on Anthropic’s API. Read more

Claude Cowork expands to web and mobile — Anthropic is targeting non-technical users with multi-step automation, increasing competition with productivity tools. Read more

Google expands Gemini Spark, a 24/7 personal agent — a reference case for an “always-on” agent with transaction-approval safeguards and a UX pattern worth studying. Read more

DeepSeek-V4-Flash-0731 moves into production — a cheap open model with agentic gains puts more pressure on the price of proprietary models used in products. Read more

OpenAI cuts GPT-5.6 Luna pricing by 80% and surpasses 1 billion users — the aggressive cut changes the cost calculation for features using OpenAI’s API and signals pressure on AI margins. Read more

n8n updates the LangChain Agent Node and AI nodes — improvements in agent orchestration reduce friction for PMs prototyping AI workflows without relying only on engineering. Read more

TrustScale launches Argus to detect AI hallucinations — an LLM verification tool addresses the number-one risk in enterprise adoption: output reliability. Read more

Backflip AI launches a generative CAD copilot — a vertical AI product with measurable ROI; cost per part fell by more than 100x, a reference for positioning around delivered value. Read more

Product-adoption metrics get confusing with AI agents — traditional DAU/MAU can mask agent-dominated usage and distort roadmap decisions. Read more

That is what made it through today’s filter.