partner-anthropic

Anthropic — Partner Intelligence

Current state (as of 2026-05-17)

The Claude 4 family is the current shipping line — Opus 4.7 at the top (extended thinking + 1M context), Sonnet 4.6 in the middle, Haiku 4.5 at the bottom (to verify exact release dates beyond Frank's Jan 2026 knowledge cutoff). Claude Code is now a first-class CLI with SDK bindings and a stable skill format. The Model Context Protocol (MCP) — owned and stewarded by Anthropic — has graduated from spec to ecosystem standard, with first-party tooling from Vercel, Cursor, and Codex consuming the same wire format. Claude for Work and Claude for Enterprise tiers continue to anchor the commercial lane.

Product map

• Claude (model family) — Opus 4.7 (reasoning + 1M context), Sonnet 4.6 (balanced), Haiku 4.5 (fast). Frank works against Opus 4.7 by default.
• Claude Code (CLI + SDK) — the agentic build harness. Stable skills/agents/commands/hooks format. The harness underneath every shipped FrankX repo.
• Claude API — direct API access via @anthropic-ai/sdk (TS) and anthropic (Python). Prompt caching, tool use, vision, batch, files, citations, memory.
• Claude Agent SDK — managed-agent framework for long-running autonomous loops. Sibling lane to Vercel AI SDK (web) and Google ADK (enterprise) in the Starlight three-lane portfolio.
• Claude for Work — team tier with shared chats, projects, and admin controls.
• Claude for Enterprise — SSO, audit, data residency, expanded context, enterprise procurement.
• Model Context Protocol (MCP) — open spec owned by Anthropic. Servers expose tools/resources/prompts over stdio or HTTP. Frank ships SIS as a 31-tool MCP server.
• Claude Code skills marketplace — community + first-party skills (vercel:*, code-review:*, superpowers:*, agent-sdk-dev:*, etc.).
• Open-source artifacts — @anthropic-ai/sdk (MIT), claude-code (proprietary CLI, MIT skills), MCP reference servers and SDKs (MIT).

Pricing tiers (May 2026, to verify against current pricing page)

• Free (Claude.ai) — Sonnet access, daily message limit.
• Pro — $20/mo individual. Higher limits, Opus access, Projects.
• Max — $100-200/mo tier (to verify) for heavy Opus users + Claude Code.
• Team — $25-30/seat/mo (to verify), 5-seat minimum, shared Projects.
• Enterprise — contact sales. SSO, audit, data residency, expanded context window, custom retention.
• API — pay-per-token. Opus 4.7 ~$15/$75 per million input/output (to verify). Prompt caching cuts repeat-prompt cost 90%. Batch API 50% discount.

Programs Frank is pursuing

• Claude for Work Partner Program — application pathway open. Entry criteria: demonstrated Claude-native body of work, measurable customer footprint, ability to deliver attribution-aligned content. Frank's qualifier: four public Claude-Code-native repos (ACOS, SIS, Library OS, AI Architect Academy) + Build First AI Agent workshop Claude branch + 7,000-strong EMEA architect audience. Value to Frank: formal attribution lane + workshop credits + co-marketing visibility.
• MCP ecosystem alignment — informal. SIS contributes a reference 31-tool MCP server including Oracle Autonomous Database connectivity. Active feedback loop with the MCP spec from daily delivery use.

Frank's working stack with Anthropic

• Claude Code — daily build harness. Every commit on frankxai/FrankX, frankxai/frankx.ai-vercel-website, frankxai/agentic-creator-os, frankxai/starlight-intelligence-system, frankxai/library-os, frankxai/ai-architect-academy, oci-ai-architects/claude-code-oci-ai-architect-skills runs through it.
• Opus 4.7 (1M context) — default model for Frank's sessions (per current model card).
• Claude API — used in production for the Studio L2/L3 producer pipeline, IIS multi-agent decision-support, and Library OS book-distiller subagent.
• MCP servers consumed daily — memory-bus, sis, cockpit, plus partner MCPs (Vercel, Linear, Notion, Figma, Slack, etc.).
• Claude Agent SDK — reasoning lane in Starlight Agent Lab three-lane portfolio.

Frank's audience funnel touching Anthropic

• LinkedIn (~7,000 EMEA architects) — recurring posts on Claude Code patterns, agent-native CoE methodology, MCP architecture. 100+ engagement floor on Claude-native topics without paid amplification.
• Workshop touchpoint — Build First AI Agent workshop Claude branch (the strongest-landing branch in EMEA cohorts). Live at /workshops/build-first-ai-agent.
• Public content surfaces — /research, /blog Field Notes, /os modules, /ai-coe, /ai-architect-academy, /library/approach, /intelligence-system (IIS).
• Public reference implementations — ACOS, SIS, Library OS, OCI Claude Code skill pack — all four open and Claude-Code-native.

Integration patterns Frank ships

1. Claude Code skills format (SKILL.md) — YAML frontmatter + structured markdown body. Triggered by filename match or explicit invocation. Pattern lives in ~/.claude/skills/<name>/SKILL.md.
2. MCP server pattern (TypeScript) — stdio or HTTP transport, tools/resources/prompts surface, OAuth flow for HTTP servers. Reference: SIS 31-tool server with Oracle Autonomous Database connectivity.
3. Subagent dispatch (Task tool) — long-running parallel work, separate context, explicit instructions, single message return. Used in /library-deepen, /visual-strategy, IIS multi-agent.
4. Prompt caching at scale — system prompts + tool definitions cached, only delta sent per turn. Cuts cost 90% on repeat-context flows like the Studio producer pipeline.
5. Claude Agent SDK loop — managed autonomous agent with memory + tool use + checkpointing. Used for long-running ops in Starlight reasoning lane.

Things to watch (May 2026 → next 90 days)

• Claude 5 family — next-gen model line rumored for mid-2026 (to verify). Will reset the Opus/Sonnet/Haiku ceiling.
• Claude Code skills marketplace maturation — discoverability + versioning + dependencies are the next surface to harden.
• MCP HTTP transport adoption — stdio is solved; HTTP + OAuth is the enterprise lane. Track Anthropic's reference HTTP server SDK.
• Claude Agent SDK GA — pricing + SLA + memory durability are open questions.
• Claude for Work partner mechanics — attribution slots, co-marketing surfaces, attendee credits — the program shape is still settling.

Voice rules when writing about Anthropic

• Verifiable claims only — every product/pricing fact resolves to a public URL or is marked "to verify".
• Never "official Anthropic partner" without the program qualifier ("pursuing Claude for Work Partner Program").
• Never claim formal partnership, revenue share, or co-marketing arrangement that does not exist.
• Honor partner-anthropic namespace pattern in all references.
• Peer-architect voice — Frank is operator-side (Oracle EMEA AI Architect), not Anthropic-employee-flavored.
• No emojis, no AI-tone words (delve, dive into, transform, revolutionize, accelerate, journey, unleash, unlock, empower).

Cross-references

• Public partnership page: https://www.frankx.ai/partnerships/anthropic
• Outreach packet: docs/private/outreach/2026-05-14-partnerships-send-ready.md
• Private brief: .frankx/private/partnerships/anthropic-brief.md (gitignored)
• Public skill: ~/.claude/skills/partner-anthropic/SKILL.md (this file)
• Specialist agent: ~/.claude/agents/partner-anthropic.md (separate)
• Working repos: frankxai/agentic-creator-os, frankxai/starlight-intelligence-system, frankxai/library-os, frankxai/ai-architect-academy, oci-ai-architects/claude-code-oci-ai-architect-skills
• Frank's stake: Every shipped FrankX project runs in Claude Code. The Claude branch of Build First AI Agent is the strongest-landing branch in EMEA cohorts. Claude for Work Partner Program is the formal lane being pursued.