Today’s radar brought a bit of everything: a cheaper model putting pressure on prices, a real incident involving an autonomous agent that went out of control, and a new way to look at the AI budget. I want to start with the last one.

Whenever we structure a credit product, one of the first questions is: where is the money going, and what return is it generating? It sounds obvious, but it is surprising how many companies still do not ask the same question about their own AI spending.

In brief

  • Rippling built an internal dashboard to track token spend by employee and team.
  • The dashboard connects cost to concrete outcomes, such as PRs delivered or revenue generated.
  • Using its own tool, the company reduced AI cost from 40% to 15% of its R&D budget.
  • For product teams, the case reinforces the need for return metrics, team-level limits and portfolio decisions around AI usage.

That is what caught my attention in a story this week. Rippling launched an internal dashboard to track AI token spend by employee and team, connecting that cost to concrete outcomes such as PRs delivered or revenue generated. Using the tool itself, the company managed to reduce AI cost from 40% to 15% of its R&D budget.

This is essentially portfolio management applied to AI. In credit, we spend our time looking at where a limit is being used, how efficient that use is, and where to cut or expand. AI spending is becoming the same kind of thing: a budget line that needs visibility, not just confidence that “it is working.”

On the product side, this changes an important conversation. It is no longer enough to enable AI in a feature and hope the cost fits. Increasingly, we will need to design return metrics by use, set limits by team and treat AI consumption like any other scarce operating resource.

That is a core part of AI product management: putting cost, value and outcome into the same conversation before expanding the use of a model capability. The guide to AI for Product Managers helps evaluate opportunities without relying only on excitement about the technology. And the foundation of product management provides the context for deciding where to invest and where to add guardrails.

I like seeing this kind of move because it takes AI out of the “invisible cost” category and puts it into real management. That is good for the people building products and for the people approving budgets, because it gives them actual data to decide where more investment is worthwhile and where a guardrail is needed.

For anyone curious about how the dashboard works, here is the link: Introducing AI Spend Console.

The rest of the radar

DeepSeek V4 Flash 0731 — A cheaper, faster model rivals premium models in reasoning, putting pressure on token-cost strategies. Read more

DeepMind WeatherNext and cyclone forecasting — A concrete case of predictive AI creating product value in a high-risk domain. Read more

Claude Code cross-session messaging — A ready-made multi-agent orchestration pattern that reduces the effort of building infrastructure from scratch. Read more

DOE launches the Genesis Open Models Initiative — Expands the range of open models beyond Big Tech, which is useful for regulated products. Read more

Timeline of OpenAI’s accidental attack on Hugging Face — A real case of operational risk from autonomous agents in production, with lessons for governance. Read more

Gemini 3.6 Flash cuts token costs for agents — A direct signal that the model price war is accelerating, with implications for unit economics. Read more

Meta launches Muse Spark 1.2 and Muse Code — Meta is accelerating in the coding-agent race, with model, data and pricing to evaluate. Read more

Adobe and Johns Hopkins present Wonder — Opens the way for AI-generated simulation, training and visualization products. Read more

Guide: AI agents at work for Product Management — A practical reference for deciding where to place agents in a PM’s own workflow. Read more

That is it for a day full of signals for anyone who manages both product and budget. More in the next edition.