How much should you actually let an agent touch? We built a calculator.

We've spent three issues talking about bounded delegation in the abstract — tiers, identity, which tasks are safe. This week we built the thing that turns it into actual numbers for your situation.

This week in the agent economy

• Singapore's central bank published SAFR, a voluntary framework that checks an agent's identity, permissions, and risk limits before it executes an action. It was built with industry partners. What it means for you: the "Know Your Agent" idea from two issues ago now has a concrete reference design behind it, though it's an industry framework, not a regulation. Whether US rules end up looking similar is still an open question.
• PayPal is positioning itself as a "trust layer" for agent-initiated purchases, letting independent agents use its transaction graph while PayPal stays merchant-of-record. What it means for you: expect more of your existing payment tools to quietly add an "agent trust layer" rather than requiring you to adopt something new.
• A wave of bank-side agent pilots (Fiserv, FIS, Temenos) are moving from pilot to wider availability this month. All emphasize auditability and human-in-the-loop review as core, not optional. What it means for you: the auditable, reviewable pattern this newsletter keeps recommending isn't just our opinion — it's what the vendor side is actually building toward.

The deep dive: how much should you actually let an agent touch?

Every issue so far has argued for bounded limits without giving you a way to calculate your own. That's what this is: enter your income and current spending in a few categories, and get a suggested, bounded limit for each — using the same three-tier framework from issue one.

It's deliberately simple right now. No account, nothing saved, no image card to share yet — just the numbers. We'd rather ship something useful early and improve it than hold it back for a polished version.

Try it: finagenthub.com/simulator

Worth trying even if you're not ready to act on it — seeing your own numbers laid out by category tends to make the abstract "which tasks are safe" question much more concrete.

— FinAgentHub