SaaStr is consolidating its AI agents rather than building separate ones for each function, with its new AI VP of Finance running inside 10K, a Replit-built marketing agent that already handles email, content, and go-to-market work. The move challenges the industry consensus that companies will eventually run dozens of specialized agents, each with its own login and silo. Instead, SaaStr's agents are collapsing into each other, creating fewer, deeper tools that share a single body of knowledge about how the business runs.

Why Did SaaStr Build an AI Finance Agent?

The decision to build an AI VP of Finance came from a specific, expensive pain point: collections had fallen six figures behind. SaaStr's part-time finance team fell behind on invoicing, someone went on vacation, and the backlog never recovered. The company ended up six figures behind on money it had already earned. Collections work is also something people quietly avoid because it feels awkward, especially when asking customers to pay for sponsorships 60 days after an event. That discomfort meant follow-up reminders slipped to the bottom of the priority list, sometimes resulting in lost revenue that was already closed and delivered.

The finance team's challenge was straightforward but costly: a signed deal would go on a slow relay. Whoever closed it sent the contract to finance, which didn't have a Salesforce login, so nothing moved in the CRM. When finance next logged in, they read the contract and created invoices by hand. On a good day that happened same-day; on a normal day it lagged into the next day. Contract-to-invoice was measured in hours or a full day, and collections follow-up after that was inconsistent at best.

How Does the AI Finance Agent Actually Work?

The AI VP of Finance is not a standalone app. Instead, it sits on top of APIs across SaaStr's finance stack, allowing it to see the whole picture instead of one slice. The moment a deal closes, the agent runs the entire sequence itself: it reads the entire contract, flips the opportunity to Closed Won in Salesforce, updates contacts, and creates invoices in bill.com off the actual contract terms, including split invoices when the deal calls for them.

Connecting the finance stack required integrating four systems, each with different levels of effort and complexity:

• bill.com: Took under 10 minutes to integrate. The agent found features the team didn't even know existed and runs mostly read-only, with the ability to write only when explicitly asked and not for everything.
• QuickBooks: Required about an hour and was the hardest integration because it requires an Intuit developer account and passing a security questionnaire. The effort was justified immediately when the agent translated tax team questions into plain English and drafted answers using real numbers pulled from QuickBooks and bill.com.
• Brex: Took about 60 seconds to integrate. The agent needs it to see upcoming card bills and incoming cash in one place, not just the invoiced side of the business.
• PandaDoc: Took same-day once support turned on API access. This is how the agent reads the contract the instant a deal is signed.

The underlying pattern across all four integrations is straightforward: start by connecting agents to real APIs. Systems of record already have guardrails, permissions, and audit trails built in, so you get the leverage of an agent without giving up the controls. As SaaStr found repeatedly, the agent starts telling you about capabilities in those tools you never knew existed.

Steps to Building an AI Agent Into Your Existing Stack

• Start with the broken thing: Don't automate everything you could this quarter. Find the specific workflow that is actually broken and causing measurable pain, then start there. For SaaStr, that was collections falling six figures behind.
• Connect to real APIs, not workarounds: Wire your agent directly into systems of record like bill.com, QuickBooks, and Salesforce. These systems already have built-in guardrails, permissions, and audit trails, so you maintain control while gaining agent leverage.
• Let the agent discover capabilities: Once connected, the agent will often find features and workflows in your existing tools that your team didn't know existed. Use these discoveries to expand what the agent can do.
• Consolidate rather than silo: Instead of building a separate app for each function, consider running new agents inside existing ones that share a body of knowledge about how your business runs.

SaaStr runs with just three humans and more than 20 agents in production. The company's approach challenges the prevailing wisdom that the future of AI agents is a hundred narrow, specialized tools. Instead, what's actually happening inside SaaStr's stack is the opposite: agents are collapsing into each other, each going deeper and drawing on shared knowledge. This structure is closer to a monorepo than an app store, where multiple functions live inside a single, integrated system rather than scattered across separate applications.

The AI VP of Finance now handles what used to be a slow, manual process. Collections follow-up is no longer inconsistent; it's automated and immediate. The finance team no longer has to manually read contracts and create invoices. And the company recovered the six figures it had fallen behind on, proving that the specific, expensive pain point justified building the entire agent in the first place.