We skipped a beat. The last Stack Movements ran in late April, and the four weeks since have been the densest stretch of platform news we have tracked all year — so this edition is a catch-up that covers roughly May 1 through June 6. The honest question for this month: when the foundation-model vendors raise at three-comma valuations and ship their own agent layers in the same fortnight, is the AI-native GTM application layer still a business — or just a thin UI waiting to be absorbed? The rounds below suggest the money has already answered. We are less sure.

The foundation-model layer goes vertical

The headline number is Anthropic’s: a $65B Series H at a $965B valuation on May 28, followed by a confidential IPO S-1 filing on June 1 — and, on the same day as the raise, the ship of Claude Opus 4.8. Three events, four days. Read together, they describe a company that has stopped behaving like a model API and started behaving like a platform with public-market ambitions and a product cadence to match. Our Claude / Anthropic profile covers why Opus has become the default substrate for the agentic-GTM tools we track.

OpenAI is not standing still. The $852B valuation set by its $122B round on March 31 remains the reference point for the category, and in May the product moved: GPT-5.5 Instant became the new default on May 5, pushing latency and cost down at the exact layer where outbound and support agents live. The OpenAI profile has the model-by-model breakdown. The number that matters for buyers is not the valuation — it is that the cost floor for a competent reasoning call dropped again, which is the single biggest input to whether an AI SDR’s unit economics work.

Then Google. At I/O 2026 on May 19, it shipped three things at once — Gemini 3.5, Gemini Omni, and Gemini Spark. Omni is the multimodal play; Spark is the lightweight, embeddable tier aimed squarely at the high-volume, low-margin tasks — classification, routing, first-draft personalization — that make up most of a GTM agent’s actual token spend. Full Gemini profile here. Our read: Google is no longer competing on frontier benchmarks for GTM workloads — it is competing on the distribution it already owns through Workspace, and Spark is the wedge.

The pattern across all three: the model vendors are done being neutral infrastructure. Each is reaching up the stack toward the agent — the layer where the GTM application vendors live.

Platforms move into the agent layer

The clearest move of the month: OpenAI’s Workspace Agents, launched April 22 and gaining traction through May. This is the foundation-model layer stepping directly onto GTM turf — agents that read your documents, draft your outreach, and act across your tools without a vendor in the middle. It is the first credible answer to the question “why pay an application vendor when the model company ships the agent itself?”

On the incumbent side, Salesforce shipped Agentforce Operations to GA on April 29 — the operations-and-monitoring layer for its agent platform. This is the less glamorous, more telling release. Agentforce’s first wave was about spinning up agents; Operations is about running them at scale: observability, guardrails, lifecycle management. That is what a platform builds once it believes agents are production infrastructure, not a demo. For RevOps teams already on Salesforce, it materially raises the switching cost of going with a standalone agent vendor.

So the squeeze on pure-play GTM-agent vendors now comes from both ends — the model layer above (Workspace Agents) and the system-of-record below (Agentforce). That is the backdrop against which this month’s funding rounds have to be read.

GTM-native rounds: the money still believes in the application layer

Three rounds, three different bets that the application layer survives the squeeze:

• Sierra — $950M Series C at $15.8B (May 4). The largest GTM-adjacent round of the month, and a near-billion-dollar vote that a dedicated conversational-AI platform out-executes the model vendors’ own agents on the hard part: production reliability for customer-facing support and sales conversations.
• Hightouch — $150M Series D at $2.75B (April 29). The composable-CDP and data-activation layer. This is the un-sexy, load-bearing bet: agents are only as good as the data they act on, and Hightouch is wiring the warehouse directly into GTM execution.
• Monaco — $50M Series B, Benchmark-led (May 12). The smallest check, the sharpest signal. Benchmark leading a Series B is a partner staking conviction this early — not a spray-and-pray.

The throughline: capital is flowing to vendors who own something the model layer cannot trivially replicate — Sierra’s reliability engineering, Hightouch’s data gravity, and whatever defensible wedge earned Monaco a Benchmark term sheet. None of these are “thin UI over GPT” bets. That is the application layer’s whole defense, and investors are pricing it as real.

What it means for your stack

For most GTM teams, none of this requires action this quarter — but it should reset your default assumptions. First: stop treating your model choice as permanent. With Opus 4.8, GPT-5.5 Instant, and Gemini Spark all shipping inside a month, the cost-and-capability floor moves faster than your procurement cycle — architect for model-swappability and let your orchestration layer abstract the provider. Second: assume the agent layer is now contested by your foundation-model vendor and your system of record. Before signing a standalone GTM-agent vendor, ask the honest question — what does it do that Workspace Agents or Agentforce Operations will not do for free inside a tool you already pay for? Sierra-class reliability is a real answer; a thin outreach wrapper is not. Third — the buyer split: if you are already deep on Salesforce, the gravity now pulls toward Agentforce, and the burden of proof is on any agent vendor asking you to leave; if you are warehouse-native and data-led, Hightouch-style activation is the more durable bet than betting on whichever model is cheapest this month. The verdict for May–June: the foundation-model vendors got bigger, richer, and more vertical — but the funding rounds say the application layer is not dead, it is just being forced to prove it owns something. Buy from the vendors who clearly do. Wait on the ones who don’t.