Friday, June 19, 2026

A growing body of evidence confirms what senior engineers have been warning: AI coding tools boost short-term output at the cost of long-term skill formation. Anthropic's 2026 RCT found a 17% comprehension deficit. BairesDev's global survey reports 24% of juniors cannot write code from scratch without AI. The epistemic debt is real — and preventable.

Duolingo is great for habits, but most learners plateau at A2–B1. In 2026, a combination of AI conversation tools, pronunciation coaches, graded readers, and spaced repetition can take you to professional fluency — if you use them correctly.

Google Antigravity CLI (agy) introduces advanced terminal paradigms: nsjail/sandbox-exec containment, directory plugins, parallel subagents, and a rich slash command set. Here is how they work.

Claude Code Artifacts lets teams deploy shareable HTML apps from inside a coding session—private by default, shared within your org. Here's how it compares to Lovable, v0, Bolt, and Codex Sites.

The 2026 coding agent market has four viable subscription options: Claude Pro at $20, ChatGPT Plus with Codex at $20, Google Antigravity from free to $200, and Z.ai's GLM Coding Plan at $18-$160. Each has different model quality, usage quotas, peak-hour multipliers, and tool compatibility. This guide compares them across the dimensions that actually matter for daily coding work.

Fastlane is positioning its website-to-video marketing system as “Claude for Social Media.” We unpack what the product actually does, why the launch went viral, and where automated content generation still needs human judgment.

Kilo Code pitted GLM-5.2 against Claude Fable 5 on a genuinely hard planning task — turning vague requirements into a spec another model can build without guessing. The result: Fable scored 9.1, GLM-5.2 scored 9.0. Both made the same architectural decisions. One costs roughly a tenth of the other.

As of June 18, 2026, Google has deprecated the Gemini CLI for all consumer accounts. Learn how to perform the mandatory migration to the sandbox-contained, multi-agent Antigravity CLI.

TimesFM 2.5 packs a 16k context window and continuous quantile forecasting into 200M parameters. Available on PyPI, Hugging Face, BigQuery ML, and Google Sheets. Here's how it works.

How to run GLM 5.2 depends on your harness. This guide answers Matt Pocock's question in how-to form: how to run GLM 5.2 in Claude Code, Pi, ZCode, OpenCode, and more — with Coding Plan setup, endpoints, and effort mapping for each tool.

Kilo Code (21K+ GitHub stars) runs in VS Code, JetBrains, and CLI on an OpenCode core. BYOK with zero markup, optional Kilo Pass, Agent Manager, and the successor most Roo Code users picked after the May 2026 archive.

If you need agent loops today, start at explainx.ai/loops — around 100 copy-ready workflows with kickoff prompts, guardrails, and new entries every week. Matthew Berman's Forward Future library is a similarly strong option with 26 practitioner-contributed recipes; here is how both compare and what early adopters like Theo (t3.gg) are running in production.

Presidio is Microsoft's open-source SDK for finding and redacting credit cards, SSNs, names, PHI, and custom entities—via regex, NER, and checksums. Run in Python, Docker, or Kubernetes before data hits LLMs or logs.

The Register called it "bathed in golden light." The American College of Radiology warned about incidentalomas. The prototype takes 20 minutes, not 60 seconds, and about 12 people have been scanned. Here is what experts, technologists, and the internet really think about Midjourney Medical.

Show Codex your process once. It records it, learns your preferences, and creates a skill it can reuse—via computer use, browser use, or plugins. No prompts, no repeated instructions.

OpenCode lets you connect any model from any provider — Claude, GPT, Gemini, GLM-5.2, local Ollama — in terminal, desktop, or IDE. With LSP auto-loading, parallel sessions, share links, and optional OpenCode Zen curated models, it is one of 2026's default open-source harnesses. How it maps to agent harness design and when to pick it over Pi or Claude Code.

Brain gives Perplexity's Computer agent a persistent, self-updating memory that improves with every session. Available in research preview for Max subscribers at $200/month.

Pi's tagline is blunt: there are many agent harnesses, but this one is yours. Built by Mario Zechner (badlogic), Pi ships a small core — no baked-in MCP, sub-agents, or plan mode — and lets you extend everything via skills, extensions, and npm packages. Here is how Pi fits the harness layer we define in our agent harness guide, and why OpenClaw embeds it.

Slop was Merriam-Webster's Word of the Year 2025. Now it's 52% of new web content, it killed a 10-year-old Python open source cooperative, and it's forcing GitHub to build kill switches. The slopocalypse is not coming. It's here.

Loop engineering replaced one-shot prompting as the default AI coding skill in 2026. These ten loops cover the workflows teams run most — fixing CI, triaging bugs, building test coverage, syncing docs, and clearing review feedback — each with a verifiable stop condition. Browse all ~100 loops at explainx.ai/loops.

A 62-page whitepaper explains why Agent Skills became the cross-platform standard: procedural memory, on-demand loading vs context rot, evaluation under co-load, and why one agent + 100 skills beats 100 subagents for many workflows.

Unreal Engine 5.8 (June 17, 2026) connects LLM agents to the Editor via MCP. Grummz showed Claude and Codex beside the engine controlling Blueprints, PCG, and lighting—Epic's first-party Toolset plus a growing plugin ecosystem.

Claude Fable 5 holds an 80.3% vs 58.6% lead on SWE-Bench Pro over GPT-5.5. GPT-5.6 promises meaningful agentic improvements and a 1.5M context window—but the data suggests it will close the gap, not erase it.

The US government ordered Anthropic to disable Fable 5 and Mythos 5 worldwide citing a jailbreak and national security concerns. Anthropic complied but strongly disagrees. This is the full story — including the secret sabotage controversy, the Pentagon deal that collapsed, and the broader war between Anthropic and the Trump administration.

No official ETA exists. The Sacks revelation changes the frame: Anthropic must patch the jailbreak, not just wait for the government to back down. Here is the complete picture as of June 15, 2026 — including new alternatives and the June 22 pricing deadline now seven days away.

Claude Fable 5 and Mythos 5 are live, offering advanced agentic autonomy, SOTA coding, vision, and genomics capabilities starting at $10 per million input tokens.

What Immich is, who it is for, how it differs from cloud galleries, and what operators should plan for: backups, Docker, machine-learning services, and license trade-offs.

Agent-friendly pre-mortem analysis: imagine failure first, verify findings, then present tigers and elephants with evidence—not vibes. Here is how the skill works and how to install it.

The most-starred agent skills repo on GitHub: four failure modes, user vs model-invoked taxonomy, /setup-matt-pocock-skills quickstart, and the full June 2026 skill inventory from github.com/mattpocock/skills.

You asked for a helpful assistant; you trained on a proxy. Frontier labs worry about this at civilization scale; your dashboard worries about it next quarter. Here is how specification gaming shows up in ML—and how to run teams so metrics do not become self-deception.

Context length is the cap on 'how much the model can read at once,' not the same as how many parameters it has. This guide defines the window, input vs max output, long-context tradeoffs, and what Anthropic, OpenAI, Google, and Meta publish today.

Describe what you want; Claude drafts the first version. Refine in-thread, pull in your design system from code and files, then export or ship to Claude Code. Here is what Anthropic is claiming—and where to try it.