I'm excited to share my first new book (or e-book) in over five years! Understanding JavaScript Promises explains not just how promises work but also how to use them in the real world.
And the best part? It's free. Grab yours now.
bit.ly/promises-ebook
Today's experiment was with DeepSeek v4 Pro and Copilot. I had it write a tech spec and then used Gemini 3.1 Pro to evaluate it. Made some edits, then used DeepSeek to come up with a plan and implement. Everything worked, just one edge case missed.
If you want to use OpenCode Go in VS Code, this extension makes it easy:
marketplace.visualstudio.com…
Add your API key once and it pulls in all of the available models for you.
It seems a lot of folks don't realize that Cursor is heavily subsidized by Anthropic and that both Anthropic and OpenAI are subsidizing their own access. GitHub is just the first to switch to a usage-based pricing model. They won't be the last.
newsletter.humanwhocodes.com…
MiMo v2.5 Pro has also proven to be a quite capable coding alternative to Copilot's selection. As a bonus, works really well with Copilot BYOK (wheras MiniMax had some trouble with tool calling).
Before the GitHub Copilot pricing changes, I used it every day and typically ended up using only 40% of my premium requests each month.
Five days of *very* light usage this week and I've already used 33% of my monthly token allotment.
I've found Gemini 3.1 Pro to be one of the least token-hungry models for Copilot. Not as good with tool calling but on debugging operations it's quite a bit more efficient than GPT-5.4 or Claude 4.6 Sonnet.
Nice reflective post from @slicknet :
"What we lose when we stop coding"
newsletter.humanwhocodes.com…
He cautions against over-reliance on AI coding agents for 100% of tasks and suggests spending a short daily session on unassisted coding each day.
Spent an hour using OpenCode with MiniMax m2.5 to implement a feature from a techspec Copilot created. The result was excellent and cost me $0.65.
Don't sleep on open source models, especially the cheap ones.
At this point, I have to say Copilot Pro plan is not a plan for everyday coding needs. If you really want to use it, you'll need to get the Copilot Max plan...even then I'm not sure the credits will last a month.
I'm currently experimenting with OpenCode Go to fill in the void.
Things I am now too scared to use on my Copilot Pro plan:
- GitHub coding agent
- GitHub code review
- GitHub desktop app
- Agentic workflows
- Autopilot
I know the subsidies had to end but the hit to the user experience is deep.
I feel like text-to-speech, speech-to-text, and summarization are going to be built into every web browser and operating system at some point soon. Implementing as an app-specific feature seems like a waste.
The first day of GitHub Copilot's usage-based billing starts with a bang: I'm one hour into work and have already spent 2% of my monthly tokens. 🤯 This is going to be challenging.
FWIW, GPT-5 mini is the cheapest model in Copilot.
Great post from @slicknet about writing agent hooks in VS Code:
humanwhocodes.com/blog/2026/…
For example, you can use a hook each time GitHub Copilot uses its edit tool, check which files were edited by that tool, then auto-run a validation step on those files.
Agent hooks in VS Code allow you to wire up deterministic actions at certain points in an agent session. In my latest post, I explain how to trigger actions when a specific file is edited by an agent:
humanwhocodes.com/blog/2026/…
I'm incredibly excited to share that Bredbox, my Pocket clone, is now in beta! This marks a big milestone as it transitions to a paid product.
Invites are still required and are sent out weekly. If you miss Pocket, I hope you'll give it a try.
bredbox.app