R&D Scan Digest
Source: departments/executive/rd-analyst-riley/scans/2026-03-29-scan.md
R&D Scan Digest
Date: 2026-03-29 Analyst: R&D Analyst (Riley)
Discoveries
Evaluate
| # | Discovery | Type | Why It Matters |
|---|---|---|---|
| 1 | Computer Use in Claude Code (March 23) | Feature | Claude can now point, click, and navigate the screen when it lacks a direct tool — no setup required. Research preview for Pro/Max on macOS. Enables digital talents that interact with client desktop apps (ERP, legacy systems) without API integration. Also expands CEO workflow automation. Source |
| 2 | Claude Dispatch — Phone-to-Desktop Agent Control (March 17) | Feature | Assign tasks from iPhone, Claude executes on Mac desktop. QR-code pairing, persistent conversation. Research preview (Max first, then Pro). Enables mobile factory operations — CEO can dispatch production tasks while away from desk. Source |
| 3 | MCP Channels — Bidirectional Messaging (March 20) | Feature | --channels flag allows MCP servers to push events into a live Claude Code session. Official support for Telegram, Discord, iMessage. Security: sender allowlist + explicit channel activation. Upgrades our Telegram plugin from "receive-only" to full bidirectional agent control. Source |
Watch
| # | Discovery | Type | Re-eval Date | Note |
|---|---|---|---|---|
| 1 | Agent Teams (Opus 4.6 experimental) | Feature | 2026-04-28 | Multi-agent orchestration with inter-agent communication. Each teammate gets its own context window. Factory has decided against agent teams internally (user preference for subagents), but the pattern matters for client delivery — digital talents that coordinate multiple agents could use this. Source |
| 2 | Official Plugin Marketplace (claude-plugins-official) | Platform | 2026-04-28 | Anthropic-curated plugin directory with verification badges. /plugin install <n>@claude-plugins-official. Could become distribution channel for digital talent delivery — package and publish client solutions as verified plugins. Source |
| 3 | Effort Frontmatter for Skills | Feature | 2026-04-28 | Skills/commands can now override model effort level via frontmatter. Minor optimization for factory — high-effort for brainstorming/planning, low-effort for simple lookups. Not urgent but useful for skill tuning. Source |
Skip
| # | Discovery | Type | Reason Skipped |
|---|---|---|---|
| 1 | X-Claude-Code-Session-Id header | Feature | API proxy infrastructure detail. No factory impact. |
| 2 | Scroll performance (WASM → TypeScript yoga-layout) | Feature | Internal UX improvement. No action needed. |
| 3 | TP-Link Tapo MCP Server | MCP Server | Smart home IoT — outside factory scope. |
| 4 | Datadog MCP Server | MCP Server | Enterprise observability — no current need. Revisit if client requires monitoring integration. |
| 5 | Autodesk 3ds Max MCP Server | MCP Server | 3D creative — outside factory scope. |
| 6 | CKAN Open Data MCP Server | MCP Server | Open data portals — outside factory scope. |
| 7 | 9,000+ plugin ecosystem milestone | Market | Volume metric, not actionable for R&D. |
| 8 | Blackbaud Development Agent for Nonprofits | Product | Sector-specific agent launch. Interesting as market signal (domain-specific AI agents going GA) but not directly actionable. |
| 9 | --resume bug fix (v2.1.85) | Bugfix | Stability fix, no evaluation needed. |
| 10 | shift+enter typeahead fix | Bugfix | UX fix, no evaluation needed. |
Watch List Re-evaluation
All 5 existing watch items have re-eval date 2026-04-25 — none are due yet. No promotions this cycle.
Update on Watch #5 (Claude Channels): Originally added 2026-03-28 from video scan as "Watch". Now officially launched as --channels research preview (March 20) with Telegram/Discord/iMessage support. Promoted to Evaluate (row #3 above) based on new evidence — it's no longer speculative.
Radar Cross-Reference
| Radar Item | Current Ring | Suggested Change | Rationale |
|---|---|---|---|
| Claude Code (Opus 4.6) | Adopt | No change | Computer Use and Dispatch are additive capabilities, not a ring change. |
| Paperclip | Assess | No change | No new evidence since RD-0015 evaluation. |
Potential new radar entries (pending evaluation):
- Computer Use → could enter Assess ring if evaluation confirms delivery relevance
- Claude Dispatch → could enter Assess ring for mobile factory operations
- MCP Channels → could enter Trial ring given we already use Telegram plugin
Recommended Next Steps
/rd-evaluate "Computer Use in Claude Code"— Highest priority. Direct impact on digital talent delivery model: agents that can interact with client desktop apps (legacy ERP, Excel, web portals) without API integration. Also relevant for factory CEO workflow. Assess reliability (~50% success rate on complex tasks per Anthropic) and SMB use cases./rd-evaluate "MCP Channels (--channels bidirectional messaging)"— High priority. We already have the Telegram plugin. Channels could upgrade it from a simple relay to full bidirectional agent control. Quick win if it works with existing setup./rd-evaluate "Claude Dispatch mobile factory access"— Medium priority. Already flagged in 2026-03-28 video scan. CEO workflow enhancement — dispatch production tasks from phone. Research preview limitations (Mac-only, must stay awake) may limit immediate adoption.Continue pending evaluations from prior scans:
- Claude Agent SDK + Harness Design Patterns (intake log, 2026-03-29)
- MCP Agent-to-Agent Communication (2026-03-26 scan)
- Anthropic Agentic Coding Trends Report (2026-03-26 scan)