Gerald Versluis’ Post

GitHub has paused new sign-ups for Copilot Pro, Pro+, and Student, tightened usage limits on individual plans, and pulled Opus access from Pro. The reason is blunt and familiar to anyone who has run shared platforms at scale: agentic workflows are chewing through far more compute than the original plan model was built for. This matters because it is another reminder that AI pricing, quotas, and service design are still catching up to real-world usage. Once users start running long sessions, parallel tasks, and subagents, the nice tidy commercial model meets production reality. Usually at 2 a.m. and with a status page open. The practical takeaway is simple. If your teams are leaning on AI coding tools, treat them like constrained infrastructure, not magic. Put guardrails around model choice, parallel execution, and cost visibility. Build fallback workflows for when quotas hit. And do not base delivery promises on "the assistant will handle it" unless you know exactly where the limits are. The interesting part is not that GitHub changed a plan. It is that we are now seeing the operational bill for agentic AI arrive in public. #GitHub #Copilot #AICodeAssistants #DevOps #PlatformEngineering #CloudCost #SecurityOperations

Hitting GitHub Copilot Rate Limits? Here’s How to Keep Your Workflow Moving- As GitHub Copilot becomes more agentic and workloads become longer‑running and parallel, many developers are running into session and weekly rate limits. it’s a natural outcome of higher‑velocity, higher‑compute workflows introduced by agents and sub‑agents. The problem: Rate limits increasingly interrupt flow just when teams are scaling AI‑assisted development — especially with parallel tasks, heavy models, and long‑running sessions. GitHub has been explicit about why these guardrails exist: to protect reliability as usage patterns intensify. 3 practical ways to avoid hitting the limit: 1.Use smaller‑multiplier models for simpler tasks Not every task needs the most powerful model. Higher multipliers burn through limits faster — reserving heavyweight models for complex reasoning helps stretch usage efficiently. 2. Switch to plan mode for structured work Using plan mode in VS Code or Copilot CLI improves task efficiency and success rate. Better planning reduces wasted tokens and reruns. 3. Reduce parallel workflows Highly parallel tools (like /fleet) dramatically increase token consumption. Reserving them for high‑leverage moments — instead of default usage — helps avoid surprise limits. Closing thought: As Copilot evolves from “assistant” to autonomous collaborator, cost, limits, and efficiency become design concerns — not afterthoughts. The teams that win won’t just prompt better; they’ll architect their AI usage intentionally. 📖 Full GitHub announcement: https://lnkd.in/gz6j4DzY #GitHubCopilot #AIDevelopment #EngineeringLeadership #DevEx #AgenticAI #SoftwareStewardship

S08E17 - Context Is Everything: Getting the Most from GitHub Copilot with Joydip Kanjilal Software architect and Microsoft MVP Joydip Kanjilal joins Jamie to discuss GitHub Copilot and AI-assisted development — covering what Copilot actually is, why context is everything when prompting it, and the governance and training considerations for teams adopting it.

GitHub Copilot is changing their pricing model. The move to usage-based pricing is rational. But killing the annual plan may be a foot gun. I have been using GitHub Copilot for the past two years, subscribed to the annual Pro plan. That was just $100 a year, and included 300 "premium requests" each month. Every time I assigned a ticket or task to an agent, that was 1 to 3 premium requests. No token limits (that I ever hit). Obviously, that pricing model could not last. Having "add error handling to this function" and "implement this ten page specification" cost the same was totally unsustainable. Honestly, I was surprised when I was able to renew the plan last year. It was a lot of AI usage for comparatively little money. But GitHub admits the problem with this move in their own FAQ. "This just wiped GitHub’s value moat – why should I stay?" They have an answer. It just is not a convincing one. Now, I don't believe the GitHub Copilot value proposition is necessarily "bad" after the change. At ten dollars a month, it's only slightly more expensive than the old annual plan. The included usage credits are probably plenty to cover my real-life usage. I may even end up continuing on a monthly plan. But I will definitely be looking into both Claude Code and OpenAI's Codex over the next few weeks, before my Copilot subscription runs out. Getting access to premium AI chat features as well as coding agent features for $20 sounds like a good value. Copilot uses their models under the hood anyway. And Claude appears to be on the cutting edge when it comes to smart prompts for agents, with GitHub lagging well behind. Not sure where I will land, but it was GitHub that gave me the push to jump.

🚀 GitHub Copilot CLI is here — and it’s not just autocomplete anymore. This is agentic AI inside your terminal. It doesn’t just suggest code… it plans, executes, tests, and even ships. Here’s the simplest way to understand it 👇 ⸻ 💡 What it actually does Think of it as a coding assistant that can: • Read your repo • Plan changes • Write code • Run tests • Open PRs 👉 All from your terminal. ⸻ ⚙️ Get started quickly Install → Login → Run copilot That’s it. You’re in. ⸻ 🧠 Make it smarter (important) Add a file: .github/copilot-instructions.md Define your: ✔️ Tech stack ✔️ Naming conventions ✔️ Code standards Now it works your way, every time. ⸻ 🔄 Two powerful modes • Plan Mode → Think first, ask before acting • Autopilot Mode → Executes everything end-to-end ⚡ Pro tip: Start with Plan → Move to Autopilot ⸻ 🤖 Switch models anytime Use /model mid-session No restart. No friction. ⸻ 🔗 Works natively with GitHub You can literally say: 👉 “Review PR #42” And it will: • Analyze code • Detect issues • Suggest fixes ⸻ ⚡ Parallel execution with /fleet Run multiple tasks at once No waiting. Just results. ⸻ 🔌 Extend with MCP servers Connect: • APIs • Databases • Internal tools One config → unlimited capability ⸻ 🎯 Try this prompt: “CI is failing on integration tests. Find the root cause, fix it, run tests, and open a PR if it passes.” 👀 Then just watch what happens. ⸻ 💭 We’re moving from writing code… to managing outcomes. Are you ready to let AI handle your dev workflow?

GitHub Copilot just announced usage limits. I wrote about the solution two months ago. This week GitHub paused new sign-ups and tightened usage limits for Copilot — citing agentic workflows consuming far more resources than the original pricing model could handle. Here's the thing: this is exactly the problem I was trying to solve when I wrote my blog post back in March. I noticed I was defaulting to Opus models for everything — file listings, simple queries, complex architecture decisions — all treated the same. So I ran an experiment: I had all 17 GitHub Copilot CLI models evaluate each other anonymously using the LLM Council technique, to figure out which model is actually right for which task. The conclusion was straightforward: - Opus = deep reasoning, architecture, the expensive senior engineer - Sonnet = solid default for most coding tasks - Haiku / Mini = fast execution for simple, well-defined work - Codex = precision coding and terminal workflows Match model tier to task complexity. Use fast/cheap models for 80% of tasks, escalate for the 20% that actually need it. This can be fully automated with agent instructions. GitHub's new limits are essentially forcing that discipline. But you don't have to wait to get hit by a limit — you can set it up intentionally. GitHub's announcement: https://lnkd.in/dk3RZQQb My full writeup with the experiment, results, and model selection instructions: https://lnkd.in/dysdTKSJ Have you thought about which model you're defaulting to — and whether it's actually the right one for the job? #GitHubCopilot #AI #DeveloperTools #Microsoft

GitHub Copilot just changed how it charges, and I think it's worth paying attention to. Starting June 1, every Copilot plan moves to usage-based billing through GitHub AI Credits. The plan prices stay the same on paper, but your actual bill won't. Token usage; input, output, cached all counts now. Heavy users are going to notice this quickly, and teams will need to start thinking about spend controls in a way they never had to before. What's quietly buried in GitHub's blog post is the honest admission: Copilot got expensive to run, and GitHub was absorbing that cost. That arrangement is done. The fallback models are gone. The premium request buffer is gone. Even code review now pulls from both AI Credits and GitHub Actions minutes. This isn't a minor update. The economics of AI development tools are genuinely shifting. For a long time, AI tooling felt almost free, vendors were subsidizing access to win market share. That made sense as a growth strategy. It doesn't scale forever, and we're watching it unwind in real time. Copilot is just the most visible example right now. The question for developers and engineering teams isn't really whether to use AI anymore. It's whether you're using it in a way that's sustainable when the bill actually reflects the cost. Worth thinking about before June.

This is a great callout. The economics of AI tools are clearly shifting, and it’s going to force more discipline in how teams use them. It also feels like a return to an old reality: where you could only optimize for two of three - speed, quality, or cost. When AI felt “free,” teams could stretch across all three. That’s likely changing. This is why quality should be the focus. If AI is just speeding up coding, costs will add up quickly and it may be better to operate without it since many of the bottlenecks elsewhere in the process will still be there. If it’s improving quality, reducing rework, and enabling engineers to operate more end-to-end, you still come out ahead - and avoid paying for the same work twice. Teams will have to be more intentional about how they use AI. That’s where it starts to separate teams that are actually getting leverage from those that aren’t.

GitHub Copilot Changes from June 1. Premium request units, or PRUs, will be replaced by GitHub AI Credits which are a monthly allotment of credits consumed based on token consumption (input, output, and cached tokens) according to the listed API rates per model. This change aligns Copilot pricing with actual usage and is an important step toward a sustainable, reliable Copilot business and experience for all users. Copilot code review’s agentic architecture runs on GitHub Actions. Copilot code review will now consume GitHub Actions minutes, in addition to GitHub AI Credits. These minutes are billed at the same per-minute rates as other GitHub Actions workflows. Fallback experiences will no longer be available. Under the new model, usage will instead be governed by available credits. Your plan pricing is unchanged: Copilot Pro remains $10/month and Pro+ remains $39/month, and each includes $10 and $39 in monthly AI Credits, respectively. Code completions and next edit suggestions stay included in all plans and do not use AI Credits. For other features, you'll need to wait for your monthly included usage to reset, purchase additional usage, or upgrade your plan.

🧠 𝗔𝗴𝗲𝗻𝘁𝘀 𝗮𝗿𝗲 𝗯𝗿𝗲𝗮𝗸𝗶𝗻𝗴 𝗳𝗹𝗮𝘁-𝗿𝗮𝘁𝗲 𝗽𝗿𝗶𝗰𝗶𝗻𝗴 — 𝗮𝗻𝗱 𝗚𝗶𝘁𝗛𝘂𝗯 𝗷𝘂𝘀𝘁 𝘀𝗮𝗶𝗱 𝗶𝘁 𝗼𝘂𝘁 𝗹𝗼𝘂𝗱 GitHub announced major changes to Copilot Individual plans last week. New sign-ups paused, tighter usage limits, Opus 4.7 restricted to Pro+. But the why is the part worth reading 👇 Quote from the announcement: "It's now common for a handful of requests to incur costs that exceed the plan price." That's the AI-coding economy in one sentence. What it means in practice: ✅ Long-running, parallelized agent sessions = real infrastructure cost ✅ A single /fleet run can burn more tokens than a week of autocomplete ✅ Token-based weekly caps are the new reality, on top of premium request quotas ✅ Model access is becoming a tier differentiator (Opus → Pro+ only) 🔗 https://lnkd.in/dsKcnirg #GitHubCopilot #AIAgents #VibeCoding #SoftwareEngineering #DevTools #MVPBuzz