CTO / VP Engineering / Head of AI

"Six quarters into AI and the demo still looks like the demo — and her devs already started using the plugin."

Type: Exec sponsor · Function: Engineering · Size: mid-market + enterprise Lead with: AI-Native interface layer · Also carries: Build-in-codebase (expansion) · Motion: top-down + land & expand

Why them

Runs a 60–400-person eng org shipping agents. Owns the build-vs-buy call, the architecture, and increasingly the AI cost line; Anthropic Managed Agents (Apr 2026) looms. The Optibus pattern is the clean version — they had the orchestration loop, they bought us as the interface: build the brain, buy the operator. Two ways in:

• Top-down: evaluates the interface layer as part of the company's declared AI initiative.
• Expansion: her developers already adopted the Developer plugin bottoms-up; now she decides whether to standardize and take it to production.

Angles — what to say (lead AI-Native; pick by situation)

• Has agentic loops but no interface (the Optibus case): you built the brain; you need the operator. The interface layer is converging across the industry — owning it isn't your differentiator. Your loops and your data are. Hours, not quarters. Own Layer 2; keep your Layer 1 IP. Real-time context selects only the relevant knowledge + actions; the system prompt is never exposed; Trust Lab de-risks switching models and launching complex actions (early access). Proof: Optibus (4 products), 20–35% lower token cost, SOC 2 Type II / GDPR / ISO 27001.
• AI cost spiraling, CFO is asking questions: inference gateway. Model routing sends simple tasks to cost-effective models and optimizes context in real time — AI flips from a variable-cost nightmare into a predictable, high-margin asset.
• Build-in-codebase (expansion): your devs already chose it — standardize. One governed layer vs. N bespoke wrappers: shared evals (Trust Lab), compliance certified once, a data flywheel across products.

Hooks

• "You spent a quarter on the harness. Anthropic shipped one in April."
• "You don't need to own the interface layer. You need to own what competitors can't replicate: your loops, your data, your domain."
• "Three teams, three bespoke agent wrappers. Or one interface layer." (expansion)

Objections → responses

• "We'll just use Claude Managed Agents directly." → That's Layer 1 / the harness. We're Layer 2 — interface, analytics, evals, and flywheel above it, bound to your permissions.
• "This is a thin wrapper." → We own the context layer (system prompt not exposed) and the data flywheel. The moat is the signal, not the model.
• "We're in a regulated / chain-of-custody environment." → The deterministic backend stays the system of record. Every agent action flows through your existing APIs and respects the same permissions — auditable, reversible, compliant. SOC 2 Type II, ISO 27001, GDPR, EU + US residency.
• "I don't want a black box in front of my product." → It's not a replacement — it sits in front, bound by your APIs, with action-scoped JWT/HMAC identity on every action. You keep the control boundary.

Targeting & channels

• Technographics: LangChain / LangGraph · Braintrust · Arize · Vercel AI · Anthropic Managed Agents
• Paid: LinkedIn (CTO / VP Eng / Head of AI) · founder-to-CTO · the "agent generated in hours" developer demo · warm outbound (the Optibus motion)
• Voices: Simon Willison · Swyx / Latent Space · Hamel Husain · The Pragmatic Engineer · AI Engineer confs · A2A/MCP technical content

Avoid

"agentic transformation" · "production-ready" · "reliable" · "revolutionary" · "end-to-end agent platform" · vague "days not months"

Source: icp-b-ainative.md · matrix positioning/gtm-messaging-matrix.md · core/overview.md, core/architecture.md. (Absorbs the former Eng-leader expansion brief — same title, the Developer-motion expansion play.)