Sales reps spend 72% of their week not selling. They're updating CRM fields, chasing invoice statuses, logging activities nobody reads. That stat, from Salesforce's State of Sales report, has been circulating for years. What's changed in 2025: the tools to fix it actually exist, and they're collapsing into each other faster than most GTM teams can keep up.
The agent convergence nobody planned for
The original pitch for AI agents went something like this: you'd have 100 specialized bots, one per function, each doing its narrow thing. Neat rows. Clean boundaries. Jason Lemkin at SaaStr took a different path. His team runs with 3 humans and over 20 agents. But instead of spinning up a standalone AI VP of Finance, they folded it into their existing AI VP of Marketing (built on 10K), which already handled email, content, and go-to-market tasks.
The result is an agent that reads contracts, updates Salesforce, creates invoices in QuickBooks, and manages collections. All from a shared knowledge base. Contract-to-invoice went from hours to under 30 seconds. That's not a marginal improvement. It rewrites the workflow.
Integration timelines were surprisingly short: bill.com took under 10 minutes, Brex about 60 seconds, PandaDoc same-day after enabling API access. QuickBooks was the outlier at roughly an hour, mostly due to security requirements. The agent even discovered underutilized features in existing tools the team hadn't known about.
Outbound isn't dead. It's being re-wired.
Cold email remains one of the highest-return outbound channels for B2B SaaS. That hasn't changed. What's shifting is who (or what) orchestrates it. Agentic AI can now qualify leads, personalize touchpoints across buying committees, and sequence follow-ups by trigger event rather than generic cadence. Gartner projected 75% of B2B sales organizations would use AI-guided selling by 2025. HubSpot's data shows AI adoption in sales jumped from 24% to 43% in recent years, with 73% of reps on AI-powered CRMs reporting boosted productivity.
McKinsey's 2024 numbers back the revenue case: companies deploying AI in sales report 3–15% revenue increases, 10–20% ROI improvement, and up to 50% more leads. Cost reductions hit 60% in some cases. Those are real ranges, not projections.
But the execution bar is higher than the hype suggests. Cold email deliverability requires warmed-up secondary domains and centralized reply handling. Personalization without segmentation by role and buying stage is just spam with a name merge. And 79% of B2B buyers now use AI tools like ChatGPT and Perplexity for research, which means your outbound strategy needs to account for "reference optimization" — becoming the source AI answers cite, not just ranking on page one.
Collections on autopilot (with guardrails)
Here's where it gets interesting for finance-adjacent operators. Collections automation used to mean reminder emails on a timer. Now agentic platforms can expand past-due account coverage by up to 10X without adding headcount. SaaStr's experience illustrates the pain: collections had fallen behind because a part-time finance team couldn't keep pace. For lean operations, slow-paying customers don't just hurt cash flow — they threaten the whole model.
The fix wasn't more people. It was bounded agency: the AI handles retries, escalations, and suspension triggers within operator-defined rules, while humans step in for disputes and complex negotiations. Start with small, reversible steps before letting loops run autonomously. That's the expert consensus, and it maps to what actually worked at SaaStr.
The catch (and it's a real one): none of this works without clean data, well-defined workflows, and deep ERP integration. SAP, Oracle, NetSuite — pick your poison, but the reconciliation layer has to be accurate. Over 40% of enterprise AI projects are projected to be canceled by 2027 due to weak data foundations or misaligned purpose. Automation built on messy inputs just produces faster mistakes.
What this means for your GTM stack
The agents aren't staying in their lanes. Marketing automation, sales sequencing, intent routing, and now collections are converging into unified workflows where handoffs happen automatically and attribution runs end-to-end from first touch to cash collected. That's the real story behind "agents collapsing into each other."
The risk is clear: compliance exposure scales with automation. Brand-safe messaging, escalation paths, and cancellable vendor contracts (with data portability) aren't nice-to-haves when an agent is sending dunning emails at 2 a.m.
Lemkin's team found three failure modes worth noting: stale data propagating outdated information, excessive guardrails breaking functionality, and missing conversational layers in tools that limit self-service. Those aren't edge cases. They're the default outcome when teams deploy agents without RevOps-grade foundations underneath.
The 30-second contract-to-invoice pipeline and the 10X collections coverage number are compelling. But the SaaStr team got there by starting with their most painful problem, connecting to real APIs, and treating every agent integration as an experiment with a stop-loss. The tools are available to any team willing to do the same. The question worth sitting with: which of your agents should have merged six months ago?
