SaaStr reports that Vivun CMO Greene, speaking at SaaStr AI, described common go-to-market stacks that now contain roughly 15 to 20 AI agents stitched together inside a sales organisation. Greene argued that the resulting "fragmentation tax" surfaces most painfully on live sales calls, where context handoffs slow reps and increase failures; SaaStr reports Greene estimated roughly 40% of first meetings end with no decision. The talk introduced Vivun's framing of the "third hop," noting foundation models perform well in a single context window but begin to degrade after about a third context transfer, producing drift and hallucinations in multi-hop sales reasoning.
What happened
SaaStr reports that Vivun CMO Greene, in a talk at SaaStr AI, described how many go-to-market teams have purchased an agent for each discrete sales job and now routinely find 15 to 20 agents stitched together inside a single sales organisation. The write-up attributes to Greene the argument that this "fragmentation tax" appears most clearly on the live sales call, where handoffs between agents lose context and slow execution. SaaStr reports Greene estimated roughly 40% of first meetings end with no decision, often recorded as "ghosted." The article reports Vivun frames multi-step sales reasoning as a sequence of "hops" and finds foundation models begin to degrade after about the third hop.
Editorial analysis - technical context
Industry-pattern observations: Foundation models perform strongly within a single, well-scoped context window, but chaining independent agents creates state fragmentation and repeated context shifts. This pattern increases the probability of semantic drift and hallucination when answers require multi-hop reasoning across persona, buying cycle, incumbent, and objection dimensions.
Context and significance
Editorial analysis: For GTM teams, the practical cost is speed and fidelity on live calls. Buyers arrive pre-researched and expect the rep to resolve the remaining, high-value questions instantly. When agent handoffs introduce latency or incorrect context, deals are more likely to stall.
What to watch
Observers should track vendor features that preserve a single coherent context across workflows, metrics on first-meeting drop-off, and whether multi-agent tool vendors publish mitigations for multi-hop drift.
Scoring Rationale #
The talk highlights a widely observed operational problem for AI-driven sales tooling that matters to practitioners integrating agents. It is notable for product and GTM teams but not a front-line model or infrastructure breakthrough.
Practice interview problems based on real data
1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with.