GitHub Copilot Six Month Review: Lessons Learned

GitHub Copilot Launches Repository Memory to Generate Organic Pull Requests 📌 GitHub Copilot’s new Repository Memory lets coding agents learn from a project’s evolution-not just its final state-so they generate pull requests that feel organic, not alien. This shift turns code generation into a continuous learning process, mirroring how human engineers study history before contributing. The result? Less redundant code, fewer rejections, and smarter, more realistic AI contributions. 🔗 Read more: https://lnkd.in/dvFfCA8d #Githubcopilot #Learningtocommit #Tsinghuauniversity #Llmcodingagents #Repositorymemory

🚨 If you use GitHub Copilot, read this before June 1. I have suggested and evangelised that people should look at CoPilot Plus subscriptions to help them with their work, and help to plug gaps in their knowledge. Only fair that I warn every one of this upcoming change. GitHub just announced Copilot is moving to usage-based billing. From June 1, every plan Free, Pro, Business, Enterprise switches from flat-rate premium requests to token-based "AI Credits." What that means in plain English: → Every line of code AI generates for you now costs real, metered money → The fallback to cheaper models when you hit your limit? Gone → Agentic coding sessions that used to be "included"? Now billed per token → Your subscription price stays the same, but your actual spend won't This affects everyone: SME founders, solo developers, teams of any size, and anyone using Copilot to learn to code. The free AI coding era is officially over. But here's the thing I actually think this is a good thing. And I've written about why. Because this change is going to force a conversation most developers have been avoiding: do you actually understand what your AI-generated code does? Or does it just "look like what we want"? 👉 Full article: https://lnkd.in/eiGVXRw2 #GitHubCopilot #AI #SoftwareEngineering #SME #AIStrategy

Anatomy of a Custom Agent Skill for your GitHub Copilot Agent 🚀 Stop treating your AI agent like a generalist! 🛑 Give it the specific expert knowledge it needs to master your project’s unique architecture. 🧠✨ #CustomSkills allow you to extend GitHub Copilot’s capabilities using nothing more than simple Markdown files. 📝 No complex backends, no heavy lifting, just clear instructions and high-quality context. ⚡️ Why this is a game-changer: 🎯 Precision: Guide the LLM to use specific libraries and internal patterns. ⚡ Efficiency: Trigger the right "tool" automatically via YAML metadata. 🛠️ Low Code: If you can write a README, you can build a Copilot Skill! I’ve open-sourced the full breakdown and integration guides in my #promptingblueprints repository. 📂⭐ Explore the tutorials here: 🔹 The Anatomy: https://lnkd.in/dsAcwMcU 🔍 🔹 The Integration: https://lnkd.in/dHNb3EsT ⚙️ Within our #DAiTA Platform at Österreichische Post AG Business Solutions, we utilize specialized skills to streamline AI-driven development and provide a robust skills layer for Agentic Frameworks. 📄🚀 📖 Read more about this approach in our blog: https://lnkd.in/dEc2ijrd Check out the image to see how simple the SKILL.md structure really is! #GitHubCopilot #AISDLC #VSCode #PromptEngineering #SoftwareDevelopment #DeveloperExperience #Coding

When I said we lived in an AI 𝗯𝘂𝗯𝗯𝗹𝗲, nobody believed me. The 𝗱𝗲𝗺𝗼𝗰𝗿𝗮𝘁𝗶𝗰 manifesto was saying: “everybody can code, everybody should code.” Yesterday, GitHub paused new sign-ups for GitHub Copilot Pro, Pro+, and Student plans.  Also planned to increase the fee on AI consumption. We’re using resources out of a marketing program and an 𝗶𝗱𝗲𝗼𝗹𝗼𝗴𝗶𝗰𝗮𝗹 one. It’s a subsidized marketing with 𝗰𝗵𝗲𝗮𝗽 tokens to 𝗵𝗼𝗼𝗸 users moving to a sustainable infrastructure or real-world 𝗽𝗿𝗶𝗰𝗶𝗻𝗴. The only way to survive is using local models as a "bridge" or a "pre-processor" for giants like Gemini and Claude, and the only 𝗹𝗼𝗴𝗶𝗰𝗮𝗹 move to avoid going 𝗯𝗿𝗼𝗸𝗲 while staying productive. For instance, models like Context7, an open-source Model Context Protocol (MCP) server developed by Upstash, provide AI coding 𝗮𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁𝘀 with real-time, up-to-date documentation for programming libraries and frameworks. It addresses a critical problem: AI models often have outdated knowledge about software libraries or hallucinate deprecated APIs, leading to incorrect code suggestions. See how to wire into 𝗚𝗶𝘁𝗛𝘂𝗯 𝗖𝗼𝗽𝗶𝗹𝗼𝘁 𝗖𝗟I, with the container booting autonomously every time the terminal opens. https://lnkd.in/dEGvAauq #AICoding #MCP #GitHubCopilot #Upstash #Context7 #LocalAI #DevTools #OpenSource #SoftwareArchitecture #CodingBubble #LLM #TechStrategy

The era of "all you can consume" AI for developers is officially ending. Woke up to the news yesterday that GitHub Copilot starting June 1, 2026... is moving to usage-based billing. While Claude Code, Cursor and other tools have also followed. It's a fundamental shift in how we build with agents. I posed about this last year that the subsidization of LLM costs was not going to last too long. Here we are now, the compute demands have become unsustainable. A single agentic loop can burn more tokens than a developer used in an entire month under the old flat-rate model. For copilot this is what it will look like from June: - "Unlimited" is replaced by credits: Your $10/mo plan now gives you exactly $10 in "GitHub AI Credits." (Personal observation, I consume $10 easily in a 6-8 hours of use with Sonnet on Copilot) - Token-based billing: You’re paying for every input, output, and cached token you consume. - Code reviews will take from that budget and will also consume github runner minutes. Double whammy there. Why does this matter? Because it forces a move toward what I call "Efficient Agency." The old model, a good agent was one that eventually found the answer, regardless of how many tokens it burned. The new eval benchmark for the future will be solving the problem with the absolute minimum number of tokens. However I dont think this is a bad thing. This shift will finally flush out the "wasteful" agents that just loop until they hit a context limit. It's going to reward engineering craftsmanship over "vibe coding" loops. P.S. At Optimal AI, we’ve been obsessing over this for a while. We use smart model routing and multi-model techniques to keep quality high while keeping costs drastically lower. This is how we can continue to provide unlimited-style value in a usage-based world. #GitHubCopilot #AIEfficiency #EngineeringLeadership #LLMOps #OptimalAI

Who's winning instead The market is crystallizing around three leaders who hold 70%+ market share: GitHub Copilot, Claude Code (Anthropic), and Anysphere (Cursor) — all have crossed $1B ARR. CB Insights Cursor is the most prominent challenger. It captured 18% market share within 18 months of launch. Quantumrun Cursor hit $1 billion in annualized revenue in under 24 months and commands a $29.3 billion valuation. Gene Dai It feels less like autocomplete and more like a smart collaborator inside your editor. Claude Code is arguably the most technically impressive. It scores 80.8% on SWE-bench Verified — the gold standard benchmark for real-world code editing tasks — and its 1M token context window can analyze 25,000–30,000 lines of code in a single prompt. It had a 46% "most loved" rating among developers in early 2026, compared to Cursor at 19% and Copilot at just 9%. Nxcode Even Microsoft's own platform acknowledged this shift — in February 2026, GitHub added Claude Code to its new Agent HQ multi-agent platform. Gene Dai

💡 𝗦𝘁𝗼𝗽 𝗿𝗲𝗮𝗱𝗶𝗻𝗴 𝗰𝗼𝗱𝗲. 𝗦𝘁𝗮𝗿𝘁 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗮𝗻𝗱𝗶𝗻𝗴 𝗶𝘁 — 𝗶𝗻𝘀𝘁𝗮𝗻𝘁𝗹𝘆. One of the most underrated features in GitHub Copilot is the /explain command. As developers, we don’t always get shiny new projects. More often than not, we step into existing or legacy systems. And that’s where the real challenge begins: 👉 “𝘞𝘩𝘢𝘵 𝘪𝘴 𝘵𝘩𝘪𝘴 𝘤𝘰𝘥𝘦 𝘥𝘰𝘪𝘯𝘨?” 👉 “𝘞𝘩𝘺 𝘸𝘢𝘴 𝘵𝘩𝘪𝘴 𝘸𝘳𝘪𝘵𝘵𝘦𝘯 𝘭𝘪𝘬𝘦 𝘵𝘩𝘪𝘴?” 🔍 𝗘𝗻𝘁𝗲𝗿 /𝗲𝘅𝗽𝗹𝗮𝗶𝗻 Just highlight the code and type: 👉 /𝘦𝘹𝘱𝘭𝘢𝘪𝘯 And Copilot will: ✔ Break down logic in simple terms ✔ Explain complex conditions and flows ✔ Decode unfamiliar code instantly ✔ Help you ramp up faster on existing projects 🚀 𝗪𝗵𝘆 𝘁𝗵𝗶𝘀 𝗺𝗮𝘁𝘁𝗲𝗿𝘀 𝗺𝗼𝗿𝗲 𝘁𝗵𝗮𝗻 𝗲𝘃𝗲𝗿: • Not every assignment is greenfield • Legacy code is everywhere • Faster understanding = faster delivery • Better understanding = fewer production bugs ⚡ 𝗥𝗲𝗮𝗹 𝗶𝗺𝗽𝗮𝗰𝘁: Instead of spending hours reverse-engineering code, you get a clear explanation in seconds. Have you used /explain on legacy code yet? What was your experience? #GitHubCopilot #AI #DeveloperProductivity #LegacyCode #SoftwareDevelopment

A few months ago, I started using GitHub Copilot, and it’s made writing code feel smoother and faster. It speeds up boilerplate code and offers smarter suggestions for functions. I usually pair it with VSCode and Playwright to test and ship features more quickly. One thing I’ve learned: don’t accept every suggestion without a careful check. Balancing AI help with manual reviews keeps my code reliable and clean. If you want to speed up your workflow but stay in control, try Copilot alongside tools like Cypress or Claude Code. How are you bringing AI into your coding routine? 🚀 #GitHubCopilot #AI #CodingLife #SoftwareEngineering #AIDrivenDevelopment

GitHub just changed how Copilot is priced for individuals. New signups paused, tighter limits. Opus models removed from the base Pro tier. The stated reason: agentic workflows now regularly generate compute costs that exceed the plan price. A handful of requests could consume more than a month’s subscription. This is a direct consequence of the agent/subagent era. A lot of people read this as “AI is getting more expensive.” I don’t totally agree. What actually happened is that the unit of AI work changed - and the pricing model hadn’t caught up. Copilot was priced for chat. You send a message, you get a reply, that’s a request. Agents broke that model. A single well-specified session can now do what used to take 40 back-and-forth exchanges. The compute is real and the request count is not the right proxy for it anymore. The shift is straightforward: write the spec first. Give the agent the full picture upfront - what you’re building, the constraints, the acceptance criteria, what done looks like. One well-constructed session replaces 40 back-and-forth exchanges. That’s one request, not forty. This is exactly how Claude Opus 4.7 is designed to be used - and why the 7.5x premium request weight (introductory price) is justified. “More expensive” is the wrong lens. The real question is whether you’re interacting with it correctly for the agent era by using long-horizon, well-specified, context-rich sessions - not rapid-fire back-and-forth that burn requests. The price of getting AI wrong is going up. Expect more restructuring and usage-based pricing and tighter tiers across the industry. #GitHubCopilot #AIProductivity #DeveloperProductivity #SoftwareEngineering #AgenticAI #SpecDrivenDevelopment

I cancelled my GitHub Copilot subscription today. Not because it's bad. Because something better replaced it. A few weeks ago I disabled it to test other AI tools. Since then I've only turned it back on twice and immediately turned it off again. The only thing I actually missed was inline code completion. And honestly? I'm writing less and less code anyway. The tools I've picked up since are better at the work I'm actually doing now: architecture, product decisions, and prompting. Copilot was built for how I worked a year ago. The tools replacing it were built for how I work today. My OpenAI subscription is probably next. Not because ChatGPT is bad, but because other tools are catching up and pulling ahead in the areas I actually use daily. The AI tooling space is moving so fast that the best tool from six months ago might not even make your shortlist today. What tools have you swapped out recently? What replaced them? #aitools #githubcopilot #github #openai #claudecode #codex