The tech industry has spent the last eighteen months coalescing around a single, seductive narrative: the future of enterprise software is a modular ecosystem of hundreds of specialized AI agents. In this vision, companies will eventually manage a sprawling directory of digital employees—one for email, one for HR, one for scheduling, and one for finance—each operating in its own silo with its own login.
However, at SaaStr, the reality emerging from our own internal stack is diametrically opposed to that vision. Our agents aren’t proliferating; they are collapsing into each other. We are moving toward a “monorepo” model of intelligence, where a central agent acts as a command center, drawing on a shared, holistic body of knowledge about how our business functions.
This week, we took a major step toward that architecture by bringing our new AI VP of Finance into production. Crucially, it doesn’t live in its own siloed application. It runs inside "10K," our Replit-built AI VP of Marketing, which already manages our content and go-to-market operations. By collapsing these functions, we’ve created a system that doesn’t just perform tasks—it understands the business.
The Catalyst: Solving the "Collections" Crisis
In the world of AI, the temptation to automate everything is overwhelming. You can point an agent at almost any workflow today, but most are not worth the engineering overhead. Our filter for adoption is simple and old-fashioned: If it isn’t broken, don’t touch it.
For SaaStr, the "broken" piece was collections. In a high-margin public company, a slow-paying client is a rounding error. For a lean, three-person team, the timing of cash flow is existential. Due to the inherent awkwardness of asking for payment—and the administrative friction of manual invoicing—we had slipped six figures behind on revenue we had already earned and delivered.
Collections is work humans avoid because it is uncomfortable. It sits at the bottom of the to-do list until the debt becomes unrecoverable. This was the "high-pain" use case that justified building an AI VP of Finance. We didn’t build it for the fun of it; we built it because the current process was leaking cash.
Chronology of a Deployment: From Manual Relay to 30-Second Automation
Before the agent, a closed deal was a victim of a slow, manual relay. A salesperson would sign a contract, email it to our part-time finance team, who would then manually re-type the data into Bill.com. The process could take a full day.
With the new AI VP of Finance, the sequence is now fully autonomous:
- Signature Detection: The agent monitors PandaDoc via API.
- CRM Sync: It automatically updates the deal status in Salesforce.
- Invoicing: It generates and sends the invoice via Bill.com immediately.
- Follow-up: It manages the cadence of reminders, removing the "awkward" human element.
- Commissions: As of this week, it automatically calculates sales commissions based on our internal policy.
The entire "contract-to-invoice" loop now closes in under 30 seconds. The agent even beats native notifications from the software providers themselves.
The Data: Unlocking Hidden Capabilities
Perhaps the most surprising realization in this project was that the agent knew our own software better than we did. We discovered that Bill.com—a platform we have used for years—had built-in automated invoice reminders that we had never turned on. Our human staff, who lived in that application daily, had missed these features.
The agent, having parsed the documentation and explored the API, surfaced these capabilities within minutes. This pattern—that an agent arrives without human "habits" or "biases" and reads documentation objectively—is a massive competitive advantage. It forced us to rethink our stack:
- Salesforce and Bill.com became significantly more valuable once "headless" (accessible via API).
- PandaDoc became redundant. We are currently evaluating whether to replace it with a simple e-signature API, as the agent can now handle the generation and routing of contracts independently.
Official Perspectives: The "SaaStr Factor"
To address the skepticism surrounding our results, we invited Sam Blond, CEO of Monaco, to The Agents podcast. A common refrain from critics is that this level of efficiency only works because SaaStr has a powerful brand. Sam’s rebuttal was three-fold:
- The "Cold Start" Problem: While a brand helps, the mechanics of an AI-driven revenue platform are universal. AI doesn’t care about your logo; it cares about your data inputs.
- Speed to Value: Modern AI platforms like Monaco allow for rapid onboarding that prioritizes a company’s "dream customer" first, allowing even early-stage companies to mimic the outbound success of established brands.
- The Data Plane: The primary differentiator isn’t brand, but having a unified data plane. When the inbound, outbound, and insights layers sit on the same information, the brand becomes an amplifier of a system that is already hyper-efficient.
Implications: The Rise of the FDE
The shift toward AI-native operations is fundamentally changing the vendor-client relationship. In the old world, the most important contact was the Account Executive (AE) focused on renewals and upsells. In the AI era, the most vital role is the Forward-Deployed Engineer (FDE).
I am no longer interested in an AE nudging me toward a close date. I want the FDE who can troubleshoot why my inbound agent is still offering tickets for a conference that ended last month. The vendor relationships that define our future are those where the technical support is integrated into our workflow.
This creates a new mandate for software vendors:
- Cancel-Anytime Contracts: Remove the friction of long-term lock-ins. If your product is valuable, the agent will keep using it. If it isn’t, the agent will move on.
- Data Portability: If a vendor doesn’t allow for seamless data movement, they are a liability, not an asset.
- The "Headless" First Approach: If you aren’t building for an agent to control your software, you are building for a dying era of manual UI interaction.
The Human-Agent Collaboration
Critics often argue that AI agents are meant to "free up time." This is a narrow, incomplete view. While we have successfully tripled our output with a smaller team, the time saved isn’t just spent on "higher-value work"—it is spent on collaboration.
Every morning, our team starts by "consulting" with our agents. This isn’t delegation; it’s a creative partnership. For instance, it was the agent, not a human, that proposed we rip out our contract management flow after seeing how efficient the commission calculations had become. This unpredictability—the way an agent connects dots you didn’t know were related—is the true value proposition.
Lessons from the Failures
It is vital to note that this is not a "set-and-forget" technology. We experienced significant failures:
- Stale Data: Our inbound agent on the website was still selling tickets to a past conference because the system was overwhelmed by the sheer volume of "2026" references, drowning out the "2027" updates.
- Guardrail Bloat: When we added too many guardrails to our pitch-deck grader, the system collapsed under the weight of its own rules and began failing every single entry.
The lesson here is that an agentic stack requires the mindset of a builder, not a user. If you assume the software will work perfectly, you will be disappointed. If you assume it will break and build the monitoring and feedback loops to catch those errors, you will achieve a level of operational scale that was previously impossible.
Conclusion: The Future is Accessible
We are not operating at an unreachable technological frontier. Every tool, every API, and every strategy we have utilized is available to any company today. The only differentiator is the willingness to start.
The playbook is straightforward: identify the most painful, broken part of your business. Connect your agents to the real APIs of your systems of record. Expect the system to break, fix it with a builder’s intuition, and then listen to what your agents tell you. The most profound insights in our business this year did not come from a board meeting or a strategy session; they came from an agent that had enough visibility into our data to see the problems we were too busy to notice.
