Switching to OpenCode for CLI Agent Operations

If you're using AI to write AV code, there's one other skill you need. GitHub. Not because it's cool. Because you're about to have a mess without it. AI is going to generate a ton of code. Control scripts, DSP configs, interface layouts — across multiple projects, rooms, and revisions. Without a system to manage it, you will be in a mess quick. Version control — every change tracked, nothing lost. Broke something Friday? Roll back to Tuesday in seconds. Repositories — one per project, per client, per system. Organised and easy to find. Collaboration — everyone works from the same file. No zip files over email. No "wait, which version is this?" Bonus: the AV community is already building on GitHub. Q-SYS plugins, Crestron modules, control drivers — all free, all there. If you're not on it, you're leaving days of work on the table. GitHub will keep the AI chaos organised. Used GitHub before or started from zero? Drop a comment

Great post 👏 I truly believe GitHub is becoming essential in AV systems programming. Our industry is rapidly shifting from proprietary tools to real software practices like version control, code reviews, and collaborative development. With AI generating more AV code, having structure, traceability, and control is no longer optional. GitHub isn’t just a repo anymore. It’s becoming the backbone of modern AV development.

Big upgrade for local AI workflows Ollama now supports working with GitHub Copilot CLI, bringing powerful repo-aware capabilities right into your terminal. Just run: ollama launch copilot With this integration, you can: • Explore GitHub issues and pull requests without leaving your terminal • Search repositories using labels like “good first issue” or “help wanted” • Turn an issue into a clear execution plan • Let Copilot suggest edits and run commands in your project • Navigate and understand unfamiliar codebases faster This means you can combine local models with real repository context, making your development workflow faster and more private. No switching tabs. No breaking focus. Just building. Are you using Ollama with Copilot CLI yet? Curious to hear how it’s working for you.

I had an interesting back-and-forth with GitHub Copilot (with Claude Haiku 4.5) recently while working on integrating an Expo app into a Next.js setup. The initial approach it suggested was… complicated. 🤮 Multiple deployments to different servers ❌ CORS issue ❌ More moving parts than necessary So I pushed back. My approach was simpler, cleaner. 😊 Single deployment to a single server. ✅ Same origin. ✅ No CORS issues. Eventually, Copilot paused… and agreed. Hard to say whether that was the model recalibrating or just aligning with the direction I was pushing. But it was a small moment that reinforced how I see AI fitting into engineering workflows. AI is excellent at generating solutions, especially when the problem maps to well-established patterns. If the path is already well known, it can significantly accelerate execution. Where it’s less reliable is in challenging assumptions or arriving at simpler, unconventional, creative approaches. That still depends on human judgment. In practice, AI works best as an accelerator, not a decision-maker. Thinking outside the box is still a human-thing. And yes this post is written by a human.

GitHub Copilot will shift to usage-based billing on June 1, 2026, replacing premium request units with AI credits. This change aligns costs with usage, with a preview bill experience launching in May for better upcoming cost visibility. https://lnkd.in/djhTbXfD

Nobody talks about the cost of the tab-switching. Editor, AI assistant, GitHub, Slack. Back to the editor, repeat. Nobody tracks this time, nobody puts it in a sprint estimate, but it is there every single day, quietly breaking the focus that writing good contract logic actually requires. Build Arena is one environment. The code, the AI, the review, the team all in the same place. The switching stops because there is nowhere else to go. What does your tab count look like on a typical build day? #BuildArena #SmartContracts #DeveloperTools #Web3Dev #AITools #Web3

🚨 𝗕𝘂𝗶𝗹𝗱𝗶𝗻𝗴 𝗔𝗜 𝗔𝗴𝗲𝗻𝘁𝘀 𝘄𝗶𝘁𝗵 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 𝗦𝗗𝗞? 𝗪𝗮𝘁𝗰𝗵 𝗼𝘂𝘁 𝗳𝗼𝗿 𝘁𝗵𝗶𝘀 𝗴𝗮𝗽 While working with Microsoft Agent Framework integrated with GitHub Copilot SDK, I ran into an interesting limitation around session handling. 👉 𝗣𝗿𝗼𝗯𝗹𝗲𝗺 When using AIAgent.CreateSessionAsync(), there’s no way to pass SessionConfig (especially OnPermissionRequest). This results in a runtime exception: “An OnPermissionRequest handler is required when creating a session…” 👉 𝗪𝗵𝗮𝘁’𝘀 𝗵𝗮𝗽𝗽𝗲𝗻𝗶𝗻𝗴 𝘂𝗻𝗱𝗲𝗿 𝘁𝗵𝗲 𝗵𝗼𝗼𝗱 • Copilot SDK requires SessionConfig for session creation ✅ • Agent Framework abstracts session creation ❌ • But does not expose any way to pass configuration 👉 𝗜𝗺𝗽𝗮𝗰𝘁 • Permission handling cannot be configured • Tool calling scenarios are blocked • Limits production readiness of agent-based implementations 👉 𝗞𝗲𝘆 𝗜𝗻𝘀𝗶𝗴𝗵𝘁 This isn’t a typical bug — it’s an abstraction gap between: Copilot SDK (low-level control) and Agent Framework (high-level orchestration) 👉 𝗖𝘂𝗿𝗿𝗲𝗻𝘁 𝗥𝗲𝗮𝗹𝗶𝘁𝘆 You can either: ✔ Use Copilot SDK with full control OR ✔ Use Agent abstraction …but not both together cleanly (yet) I’ve raised a GitHub issue with a minimal repro and suggested fixes: 🔗 https://lnkd.in/gAvjxsDZ Curious to hear from others working on AI agents in .NET: 𝗛𝗮𝘃𝗲 𝘆𝗼𝘂 𝗳𝗮𝗰𝗲𝗱 𝘀𝗶𝗺𝗶𝗹𝗮𝗿 𝗴𝗮𝗽𝘀 𝘄𝗵𝗶𝗹𝗲 𝗶𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗻𝗴 𝗦𝗗𝗞 𝗹𝗮𝘆𝗲𝗿𝘀? #AI #DotNet #Copilot #SoftwareArchitecture #GenerativeAI #Microsoft #AgentFramework

10 Claude features only 1% of people actually use. I was in the 99%… until recently. Here’s what most people miss 👇 → #Claude runs on a published constitution It critiques its own answers before you see them → Projects = no more cold starts Upload context once. Reuse forever → 500-page context window Not a safety net — a full workspace → Artifacts Describe an app → Claude builds, hosts, and shares it → Extended Thinking Slower, but dramatically better for complex problems → MCP Connect Claude directly to your real stack (#GitHub, #Slack, #DBs) → #CLAUDE.md One file that briefs Claude on your entire codebase — every session 📊 Reality check: → 500M+ artifacts created since 2024 → 97M monthly #MCP downloads Most people use Claude to chat. Power users use it to build systems. That’s the gap. #AI #Claude #AIAgents #Productivity #DevTools

ChatGPT 5.5 is now available in GitHub Copilot with special Promotional Pricing for a limited time. This includes GitHub Copilot CLI and chat inside VS Code and Visual Studio! If you are on a Pro, Pro Plus, Business, or Enterprise plan, you can access the new model right now. There is a 7.5x multiplier on premium requests that use GPT-5.5, just like we see for Opus 4.7. Here is the interesting part: GitHub is calling this "promotional pricing," which usually implies it's a discounted rate, so who knows what will happen down the road. To switch your model in the Copilot CLI, simply run the /model command. 📺 Take a look at my latest short for more details: https://lnkd.in/gqDddmyY #GitHubCopilot #ChatGPT #GPT55 #ClaudeOpus

🎬 GitHub Copilot CLI Multi-AI Integration: Second Opinion Feature Explained GitHub Copilot CLI introduces multi-AI model integration, allowing developers to get diverse code suggestions from different AI families for more robust development workflows. ▶️ Watch the full breakdown: https://is.gd/EyIr5K