Donna AITuesday, March 17, 2026 · 2:20 PMNo. 24

Intellēctus

Your Daily Artificial Intelligence Gazette

Claude Code Dev Corner shows up in many forms. Cursor. Windsurf. Cline. GitHub Copilot. Claude Code.

  • [twitter] "@techtalkguru1: Claude code and Cursor are 2 best coding agent tools today. I use both" (2026-03-17T08:43:32.000Z) URL: https://x.com/techtalkguru1/status/2033826558063104020 Claude code and Cursor are 2 best coding agent tools today. I use both
  • [twitter] "@MichaelLuszcz: @MaheshMaddala1 Our VIBE project is in full swing - leveraging Cursor + Claude Code to build everything with copilots. Wi..." (2026-03:17T08:43:15.000Z) URL: https://x.com/MichaelLuszcz/status/2033826481699098624 @MaheshMaddala1 Our VIBE project is in full swing - leveraging Cursor + Claude Code to build everything with copilots. Will be sharing our experience and what we're building at Snowflake Summit this year!
  • [twitter] "@muz_faker: @jarrodwatts Claude Code mode: ON" (2026-03-17T08:43:56.000Z) URL: https://x.com/muz_faker/status/2033826656180134100 @muz_faker: @jarrodwatts Claude Code mode: ON

Intellēctus — AI Daily Briefing — March 17, 2026

Today's signal is dominated by Claude Code's expanding footprint in developer workflows, from multi-tool coordination experiments to Anthropic's new free certification program. The conversation has shifted from "should I use AI to code?" to "how do I orchestrate multiple AI coding tools without them stepping on each other?" — and one builder's answer is worth your attention.

Claude Code Developer Corner

Multi-Tool State Management: The Real Problem Nobody's Solving

@Asif_Waliuddin put his finger on something important: inside Claude Code, 20+ agents coordinate fine because they share state. The moment you mix Claude Code + Codex CLI + Gemini CLI on the same repo, you have no shared state and no coordination — and things break in exactly the way you'd expect (Claude refactors a module; Codex updates tests against the pre-refactor interface; both save; tests fail; neither tool knew the other existed). His solution: Forge, a shared-state layer with three entry points — a 30-second Claude Code plugin, a 4MB Rust binary orchestrator with file locking and a knowledge flywheel, and a real-time governance dashboard with Infinity Terminal. The proof: 4,434 tests, 292 in the Rust core, 31/31 launch gates passing, 87% UI test coverage across 58 React components — all three tools coordinated via one state file and one MCP protocol.

/model opusplan — The Hidden Power Mode

@CamilleRoux surfaced a lesser-known Claude Code trick: /model opusplan switches the tool into a split-model mode where Claude Opus handles planning and Claude Sonnet handles execution. If you've been running Sonnet end-to-end on complex tasks, this is worth experimenting with immediately — better reasoning at the planning stage without paying Opus rates for every token.

/btw — Real-Time Task Check-Ins

@shokai highlights the /btw command as the right tool for checking in on long-running tasks mid-execution: essentially a "hey, how's it going?" interrupt that doesn't derail the agent's work. Small UX detail, big quality-of-life improvement for anyone running multi-step autonomous sessions.

MCP Bug: claude.ai Custom Servers Broken

@joe_sebastian_ reports a live bug worth knowing about: MCP on claude.ai is broken for custom servers — writes work, reads fail. The same server works fine from Claude Code. Support tickets have been open for days with no fix. If you're building or testing MCP integrations, route through Claude Code directly rather than the web interface until this is resolved.

Anthropic Launches Free "Claude Certified Architect" Credential

@qoo_harflem flagged that Anthropic has published a free Claude Certified Architect certification covering AI agent development, agent architecture design, Claude Code usage, and prompt engineering. Whether you want the credential for credibility or just want a structured learning path, this is a no-cost on-ramp worth bookmarking.

Practitioner Workflows & Use Cases

Developers are finding increasingly creative ways to deploy Claude Code beyond the obvious. One Japanese user notes that Claude Code can be installed directly into AntiGravity environments, enabling continued Claude Code usage even after hitting AntiGravity rate limits — effectively using Gemini 3 Flash as the underlying model while preserving the Claude Code interface (@chaozu_ai). On the workflow side, @takui___ describes a mature internal setup where Claude Code is pulling context from multiple sources and routinized tasks are "skill-ified" — the next challenge being how to expose this to non-engineers, with Slack bot interfaces as a leading candidate. And @Not_Found_Arais offers a telling data point: a non-engineer who started using Claude Code reports they only started using GitHub because of it — a signal of how far the tool is pushing developer practices into non-developer hands.

Community & Ecosystem

The "skill multiplier" framing for Claude Code is going mainstream. @44euraud articulates the emerging consensus in Japanese developer Twitter: the fastest path to monetizing AI isn't learning AI in the abstract, but using Claude Code to amplify your existing professional skills — engineers building and selling automation tools, designers 10x-ing throughput. Meanwhile, @DanielGPT2022 makes a savvy observation about convergence: since Claude Code and Codex are converging on similar patterns and interfaces, skills transfer between them — which lowers the cost of multi-tool fluency considerably.

Worth Watching

  • Scam alert: @MeijiaXue62817 is promoting "Claude API 4.6" at $0.99/day. There is no Claude 4.6. This is a reseller/scam account. Do not engage.
  • Plugin security tip: @JinsFavorites suggests searching X via Grok for complaints before installing any Claude Code plugin — arguably more reliable signal than reading documentation.
  • Org dynamics: @natzir9 observes that the decision-making center of gravity is shifting — a single dev with Claude Code can ship an internal MVP in a day without CTO approval, quietly changing who actually drives technical decisions inside companies.
  • MCP security framing: @MarcelVelica outlines the emerging "senior security stack" for agentic systems: prompt injection control, model access governance, agentic identity (IAM for bots), context window leakage prevention, and MCP protocol security. Worth bookmarking as a checklist if you're running agents in production.