Persona
Engineering Owners: CTO, VP Eng (+ AI Engineer, Developer champions)
Hears the AI-Native interface-layer story: build the brain, buy the operator. The most skeptical audience in the deal and the one that verifies everything. Simon Willison test on every asset.
Who they are (three seats, one page)
- CTO / VP Eng (economic buyer): owns the AI bet and its risk; fears the stalled build and the unbounded agent. Signs top-down deals.
- AI Engineer (champion): owns agents/loops/evals; evaluates routing, protocols, eval depth. The person who can kill us on a technicality or champion us to the CTO.
- Developer (user → champion): wants CLI/SDK, one command, drop-in; time-to-first-agent is the felt metric (no public benchmark yet: don't invent one).
Where they live, what they trust (external research)
- Discovery is peer-led and dark: private Slack groups, invite-only roundtables, conference hallways (KubeCon, re:Invent, QCon); engineers trust peers, not vendors (daily.dev).
- Artifacts that work: architecture deep-dives, benchmarks, open technical content, and video walkthroughs of the real interface over decks (Uplift GTM). 71% rely on whitepaper-depth content during research (LeadHaste).
- Peer relevance beats product relevance: a named outcome at a recognizable company in their sector beats ten thought-leadership posts (Somebody Digital). Getting Optibus-class logos cleared is a build-motion growth lever.
- Register: Simon Willison / Swyx / Pragmatic Engineer. If it reads like a marketing site, it failed.
Key messages (sheet) and how to say them
- Agentic Interface, FE not IP: their domain agents and data are the moat; the interface layer isn't. Don't spend the roadmap rebuilding chat UX, context engines, and eval plumbing.
- Build with Foldspace: A2A, MCP, SDK, deployment, testing. It's engineering on their stack, not configuration of ours.
- AI Score = latency, cost, quality, per action, per model. Cost control and model control as first-class features.
"You built the brain. We're the operator."
"Your agents stay yours. We front them via A2A."
"A prompt change is a code change. Test it like one."
Features that matter to them
| Feature | Frame it as |
|---|
| Developer plugin + SDK | Snippet, identify(), actions on existing endpoints. Show code in minute one |
| A2A + MCP | Standard protocols; the lock-in objection dies here |
| Per-response metrics + Optimizer | Model, tokens, cache, price per response; routing = 20–35% token cost cut |
| Trust Lab (early access, always) | Golden sets, LLM-judge + deterministic + tool-use evals, import-from-production regression |
| Bounded execution | Agent acts only through existing endpoints and the user's permissions; blast radius is a design property |
Proof pack
Optibus: 4 products live, 1–2 business days each, A2A into their LangChain loops, 20–35% token cost reduction. The strongest artifact we own for this persona; push for logo clearance. Anonymize until then.
Vocabulary
Say- Concrete specifics: protocols, endpoints, eval types, latencies, prices
- "Layer 2," "FE not IP," "bounded to your backend and permissions"
- Concede LangChain/Anthropic strengths first, then differentiate
Never- "Reliable" (most overused word in the category), "production-ready," "agents that work," "ship at scale"
- Marketing-speak of any kind; unverifiable metrics
- "End-to-end agent platform," "democratize AI"
Motion note: this persona is the heart of the BUILD motion (
mechanics here). In the TOOL motion they appear as the security/architecture
veto: answer bounded-action, blast-radius, observability, and lock-in questions in peer-engineer tone, pass the review, and leave the build seed planted.