AI Developer Digest

Sun, Oct 18, 2026

4 signals that cleared the gate20 scanned15 min read

The Signal — start here

A quiet release day, but one item requires action today: the GitHub Copilot second deprecation wave lands tomorrow (October 19) — GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash, and Grok 4.5 are removed from all Copilot experiences at midnight. If you have any Copilot API integrations or configured extension settings pinning those model IDs, they will silently fall back to Copilot defaults. The rest of today's scan is community signal: morluto/rea gained 240k stars this week — AI-native reverse engineering as a one-command MCP server has arrived — and the cursor/plugins marketplace is clearly gaining developer traction with 5 new plugins merged in the last 24 hours.

Must-reads today

GitHub Copilot Oct 19 deprecation [BREAKING]: Six models removed from all Copilot experiences tomorrow — audit any pinned model IDs in Copilot API calls or IDE settings before end of day.

morluto/rea viral moment [Trends]: An AI reverse-engineering MCP server adds 240k stars in one week — a signal worth understanding before your team runs into it.

🏆 Repo of the Day

cursor/plugins — Cursor's plugin marketplace spec: standardize skills, rules, and MCP servers for any tool in Cursor

As a Cursor user — install a plugin from the marketplace:

Open Cursor, then open the command palette (Cmd/Ctrl+Shift+P)

Type: Plugins: Browse Marketplace

Or install pstack (the flagship plugin) directly via Cursor chat:

/plugin install pstack

As a plugin author — clone the spec and scaffold a new plugin:

git clone https://github.com/cursor/plugins

Each plugin lives in its own folder with:

.cursor-plugin/plugin.json ← manifest (name, displayName, description)

skills/ ← SKILL.md files

rules/ ← .mdc rule files

mcp.json ← MCP server definitions

Breaking Changes

●Breaking

GitHub Copilot — second October model deprecation wave lands tomorrow (October 19): six models removed from all Copilot experiences

What changed

GitHub Copilot is removing six models from the model picker across Chat, inline edits, agent mode, and code completions. Models removed: GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash, and Grok 4.5. Suggested replacements: GPT-5.6 Sol for GPT-5.5; GPT-5.6 Luna for GPT-5.4 mini and GPT-5 mini; Gemini 3.8 Flash for Gemini 3.7 Flash; Grok 4.6 for Grok 4.5. Note: an earlier wave on October 2 already removed Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7.

TL;DR

Six Copilot models removed tomorrow (Oct 19): GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash, Grok 4.5. For Copilot Business/Enterprise, suggested replacements are auto-enabled unless admins have disabled default model enablement.

Developer signal

Two things to check before midnight tonight:

Copilot API integrations. If you call the Copilot API with an explicit model parameter (e.g., model: "gpt-5.5" in a GitHub Models or Copilot API call), you'll get an error or a silent fallback after tomorrow. Audit any CI scripts, IDE extensions, or automation that pins these model IDs and update to the replacement model strings. The Copilot API uses the same model identifiers as GitHub Models — verify against the Models marketplace listing for current availability.

Admin-configured model policies. For Copilot Enterprise/Business, if your organization uses a model allowlist or blocklist policy in Copilot settings, double-check that the replacement models are not accidentally blocked. Default model enablement means replacements should appear automatically, but a policy that explicitly allows only specific models by name will break silently unless you add the new names.

The change applies to Copilot-the-feature (in-IDE chat, inline completions, agent mode). It does not directly affect the models' API availability outside Copilot — but if you rely on GitHub Models as a proxy to access these model versions, confirm their status there separately.

Affects you ifYou call the Copilot API or GitHub Models API with any of the six model IDs above; you maintain VS Code or JetBrains extensions that depend on pinned Copilot model versions; you administer Copilot for an Enterprise or Business org with model policy settings.EffortQuick for most (replacement models are auto-enabled); Moderate if you maintain code or policies that explicitly reference the deprecated model IDs.

GitHub Changelog | Date: Announced Sep 18, 2026; effective Oct 19 | Link: https://github.blog/changelog/2026-09-18-upcoming-deprecation-of-selected-github-copilot-models-in-mid-octoberhttps://github.blog/changelog/2026-09-18-upcoming-deprecation-of-selected-github-copilot-models-in-mid-october | October 2 wave: https://github.blog/changelog/2026-10-02-selected-models-in-github-copilot-deprecated

Model Releases

No model releases in the October 17–18 window.

API & SDK Changes

No new API or SDK changes confirmed in the October 17–18 window. DNS failures blocked platform.claude.com and ai.google.dev at scan time — check those directly for any same-day updates.

Research

arXiv RSS feeds (cs.AI, cs.CL, cs.LG, cs.CV) returned DNS failures again this scan cycle. huggingface.co/papers inaccessible. No qualifying research papers surfaced in the October 17–18 window. Section omitted.

Tooling

No new major tooling releases in the October 17–18 window. LangGraph 1.2.16 and SDK 0.4.8 (Oct 16) were covered in yesterday's digest.

Rising Dev Tools

Notable

cursor/plugins — Cursor's official plugin specification and marketplace directory

Developer signal

If you use Cursor as your primary IDE, the plugin directory is the fastest way to add structured skills and MCP servers without writing JSON from scratch. Clone cursor/plugins to understand the spec: a plugin is a folder with a .cursor-plugin/plugin.json manifest, optional skills/, rules/, and mcp.json. If you maintain tooling for both Cursor and Claude Code, note that today's commit adds a Claude Code adapter to the pstack plugin — skills are written to ~/.claude/rules, giving one plugin source partial cross-IDE coverage.

Affects you ifYou use Cursor IDE and want to extend it with additional skills or integrate external services via MCP; you author plugins and want to reach the Cursor community; you maintain AI tooling for multiple coding environments.

GitHub Trending + direct observation | Stars: ~12,682 (↑ 1,536 this week, +14%) | Link: github.com/cursor/plugins

Benchmarks & Leaderboards

arena.ai leaderboard not directly accessible (DNS failure). From secondary sources (Manifold prediction markets, early October snapshot): Gemini 4 Argon leads the text board at ~1525 ELO. Claude Opus 5.5 leads the WebDev board at ~1813 vs. GPT-6 Astra at ~1786. Treat these as directional until confirmed against arena.ai directly — figures are from prediction market creator notes, not an official snapshot.

Trends & Emerging Tech

AI-native reverse engineering goes mainstream: morluto/rea reaches 320k stars (240k gained this week)

What's happening

morluto/rea — "Reverse Engineer Anything" — gained approximately 240,000 stars in the past 7 days, ending the week at ~320,000. The project is an MIT-licensed MCP server that gives any MCP-compatible agent (Claude Code, Cursor, Codex, Gemini CLI) the ability to inspect and understand native binaries, JavaScript/Electron apps, .NET assemblies, Android APKs, firmware, and packet captures. Analysis runs locally. The install is a single npx rea-agents setup — after which any connected agent can call the MCP server to reverse-engineer targets and explain behavior or generate reimplementations.

Why watch this

A 240k-star week is extraordinary by any standard — it's one of the largest single-week star gains ever recorded on a non-new repo (the project has 3,719 commits). This signals that the combination of MCP + AI agents + local RE tooling has hit a nerve. For developers, this changes how you can approach "what does this binary/app/APK actually do" questions — instead of manually running Ghidra or jadx, an agent with rea can systematically reconstruct behavior. For security teams, it raises the question: if your customers can do this to your binaries, what does that mean for IP and reverse-engineering-protection assumptions? Worth understanding whether this tool is in your team's threat model.

GitHub Trending (weekly) | Date: October 12–18, 2026 | Link: https://github.com/morluto/rea

Technical Discussions

Nothing cleared the quality bar this period. hnrss.org DNS-blocked at scan time.

Quick Hits

Gemini API: gemini-3.5-flash deprecated October 8, replaced by gemini-3.6-flash — If you still call gemini-3.5-flash, requests are auto-routed to gemini-3.6-flash, which deprecates temperature, top_p, and top_k parameters (use thinking_level for control). Note: for Vertex AI, gemini-3.5-flash may remain available until May 2027 — check the Vertex lifecycle page separately. Applies to the Gemini API (ai.google.dev); not covered in prior digests due to Google source DNS failures at the time. [ai.google.dev/gemini-api/docs/changelog]
anthropics/claude-plugins-official — 20 plugin version bumps on Oct 16 — Routine automated updates to incident-io, dropbox, canva, amplitude, ui-theme-designer, slack, and 14 other existing plugins. No new plugins added in this batch. [github.com/anthropics/claude-plugins-official]

Worth Watching (Announced, Not Yet Shipped)

GitHub Copilot Oct 19 deprecation — see [BREAKING] section above — effective tomorrow.
OpenAI Oct 23 model retirement — 5 days. gpt-4, gpt-4-turbo, o1, o3-mini, o4-mini, gpt-image-1 removed from the API. Audit OpenAI API calls using these model IDs before October 23.
Claude Design (claude.ai/design) closure — December 14, 2026. Standalone service closes. Start /design-sync migration planning if your team uses it.
Mistral Large 4 open weights — End of October target (secondary source, unconfirmed). Watch mistral.ai/news.
GitHub Universe Oct 28–29 — San Francisco, Fort Mason Center. Copilot coding agent and GitHub Models updates expected.
Gemini 4 Argon introductory pricing expiry — December 1, 2026. $2/$10/MTok rising to $4/$20/MTok.
Anthropic Dynamic Workflows (Managed Agents) GA — In beta since Oct 9. No GA date announced.

Filtered from 30+ primary sources against a published quality rubric. No press releases, no fluff — only what changes what you build.