TFD-0015: R&D Stage 2 Decisions — Computer Use & MCP Channels

TFD-0015: R&D Stage 2 Decisions — Computer Use & MCP Channels

Status: Accepted Date: 2026-03-29 Decision maker: CTO (Clara)

Context

R&D Analyst (Riley) completed two evaluations on 2026-03-29:

  • RD-0016 — Computer Use in Claude Code (Score: 3.0)
  • RD-0017 — MCP Channels: Bidirectional Messaging (Score: 4.0)

Both require a Stage 2 decision per the R&D pipeline (TFD-0012).

Decisions

RD-0016 — Computer Use in Claude Code

Decision: Hold

Riley's evaluation is well-grounded. The strategic value is obvious — desktop app control without APIs opens the legacy SMB segment we can't reach today. But the numbers don't justify production investment yet:

  • 56% success rate on web tasks is below our quality bar
  • macOS-only blocks most SMB clients (Windows-dominant)
  • Research preview = no stability guarantees

Actions:

  1. Watch list — re-evaluate on GA release or in 60 days (2026-05-28), whichever comes first
  2. No internal prototype yet — the effort isn't worth it at 56% reliability. When success rate hits 75%+, Ada should design a fallback-to-screen pattern
  3. Hybrid strategy approved — Playwright MCP for browser automation now, computer use for desktop-only scenarios later

RD-0017 — MCP Channels: Bidirectional Messaging

Decision: Go

This is a no-brainer. Zero-effort activation (plugin already installed), immediate operational value (mobile factory control), and failure mode is simply falling back to what we have today. The 4.0 score is justified.

Actions:

  1. Ivan (Infrastructure) — activate --channels plugin:telegram@claude-plugins-official in standard launch configuration
  2. CEO — complete Telegram pairing flow, test permission relay round-trip
  3. Do NOT treat as critical infrastructure — this is an opportunistic upgrade. If Anthropic breaks it, we fall back to terminal
  4. Monitor — track Anthropic changelog for breaking changes. Known issue: rate limit drain bug (reported 2026-03-26)

Rationale

Two very different maturity levels. Computer Use is a future capability to watch; MCP Channels is a present capability to use. Resource allocation follows accordingly: zero investment in Computer Use, minimal investment in Channels activation.

References

  • RD-0016 (Computer Use evaluation)
  • RD-0017 (MCP Channels evaluation)
  • TFD-0012 (R&D pipeline)