Claude Code Tips and "Wild" 2026 Predictions
And other news in AI Coding: January 6, 2026
Ok, it seems like over Christmas break, a lot of people tried Opus 4.5 and something fundamentally changed in dev society. People quickly went from reviewing every step of AI agent, to firing multiple agents without even looking at the final code result.
Also, now it’s pretty clear to me that devs focus on Claude Code as the main AI tool. Alternatives like Codex, Cursor and others receive like 10x less raving on social media combined. So, I will probably focus on that as well and dive deeper into best practices of using Claude Code in my articles.
Reading AI coding tweets, I do feel overwhelmed by the amount of tools and methodologies, but this is where my weekly newsletter should help the sanity for both you and me :)
So, I will try to re-cap the main “trending” things you need to know about AI Coding.
For Paid Substack Members
From this week, I wanna publish one EXTRA piece of content for Premium members.
So here’s one for this week: My AI Guidelines for Laravel/PHP and Filament v4
My YouTube Videos
Claude Code: Plan Mode “Trick” for Bigger Feature/Project Spec (7 min)
Claude Code: in Terminal, in Cursor IDE, or in VS Code Extension? (18 mins)
The Future (I think) of CLAUDE.md and Other Guidelines (8 mins)
[SHORT] Claude Code: Prompt Stashing to Quickly Run Another Prompt
AI Coding News, Tips, and Opinions
Boris, creator of Claude Code, shared his own setup in a tweet that went absolutely viral.
Ralph Wiggum Claude Code plugin is a new “hot” thing, getting increasingly viral on social media.
Community is suspicious: Did Anthropic use the 2× event to quietly reduce usage limits?
Reddit post with praise from the community: I condensed 8 years of product design experience into a Claude skill, the results are impressive
If you’re into yearly reviews, a good one from Simon Willison: 2025: The year in LLMs
Reminder: Claude Code isn’t just for coding. Pietro on X: “I fed it my raw DNA data from an ancestry test and used it to find health related genes I should keep an eye on.”
Nader on X posted that he re-created TypeForm in 35 minutes with Opus 4.5. This is another proof of possibilities for coding, but also it’s showing that code is NOT the reason for success of 100B companies like TypeForm. Otherwise, all vibe-coders would be millionaires by now. I’m planning to shoot a video about it.
Another similar story: Noam vibecoded an open-source poker river solver with Codex and Claude Code. It looked good on the surface, but AI would make mistakes and encounter bugs in domain knowlegde a lot. Another angle for successful product: you need to be a DOMAIN expert. That’s your moat against AI agents who can “just” write code.
Prediction by Phillipp on X: If 2025 was beginning of agents, 2026 will be around Agent Harnesses.
Another prediction by Burkov on X: The new software development paradigm: the purity of the codebase is no longer the holy cow protected by angry senior devs (so poetically phrased).
Jaana from Google shared that Claude Code helped her to achieve result, which kinda shocked the community and quickly became a “I’m not joking and it isn’t funny“ meme.


"Domain expertise as moat" - 100% this.
The observation about developers "firing multiple agents without even looking at the final code result" over Christmas - I saw this shift too. Something fundamental changed.
Your point about domain expertise being the moat is exactly right. I've been building with Claude Code for months now and the differentiator isn't prompting skill - it's knowing what you're building and why. The AI handles syntax, you handle strategy.
The Plan Mode tip is underrated. I use it constantly now for anything beyond trivial changes. Seeing Claude's thinking before execution has prevented so many "oops, that's not what I meant" moments.
On your prediction about "agent harnesses" - already seeing this. My personal agent system (Wiz) is essentially this: a framework for managing multiple specialized agents, not the agents themselves.
Wrote my honest assessment of Claude Code (what works, what doesn't) here: https://thoughts.jock.pl/p/claude-code-review-real-testing-vs-zapier-make-2026