Welcome back to Agentic Coding Weekly. Here are the updates on agentic coding tools, models, and workflows worth your attention for the week of Nov 23 - 29, 2025.
Tooling and Model Updates
Claude Opus 4.5
Anthropic shipped Claude Opus 4.5, their new SOTA model that’s particularly strong on coding and reasoning benchmarks.
A few things that stand out:
Hits 80.9% on SWE-bench Verified, the first model to cross 80% there.
Also leads on Terminal-bench 2.0 and ARC-AGI-2 (Verified) against Gemini 3 Pro and GPT 5.1 Codex Max.
Still has a 200k token context window, similar to earlier Opus models.
Knowledge cutoff: Aug 2025 (vs Jul 2025 for Sonnet 4.5 and Jan 2025 for Gemini 3 Pro).
3x cheaper than Opus 4.1: now $5 / $25 per million input / output tokens.
There's a new effort parameter to control token usage when Claude responds. Takes high, medium, or low values, and defaults to high. It affects reasoning tokens as well as the final answer. For non-trivial coding and design work, you’ll likely want high plus a generous thinking-token budget; for quick edits and refactors, using low could save cost and latency.
Check the announcement and system card for details.
MCP turns one
Model Context Protocol hit its one-year mark on Nov 26th. The retrospective is worth reading. Latest release adds support for task-based workflows to manage long-running, multi-step operations.
Workflow of the Week
Beads is an issue tracker designed for AI coding agents. If you already do spec-driven dev where you plan in a markdown file, implement in fresh sessions, keep updating the plan as you go, Beads formalizes that flow. It acts as persistent memory, allowing an agent to plan, create epics/issues, and then implement them in isolated sessions based on the those issues. No need for markdown files.
Install it using npm install -g @beads/bd and then follow this one-time setup as suggested in readme:
# In your project root:
bd init
# For OSS contributors (fork workflow):
bd init --contributor
# For team members (branch workflow):
bd init --team
# For protected branches (GitHub/GitLab):
bd init --branch beads-metadata
# bd will:
# - Create .beads/ directory with database
# - Import existing issues from git (if any)
# - Prompt to install git hooks (recommended: say yes)
# - Prompt to configure git merge driver (recommended: say yes)
# - Auto-start daemon for sync
# Then tell your agent about bd:
echo -e "\nBEFORE ANYTHING ELSE: run 'bd onboard' and follow the instructions" >> AGENTS.mdThe author's best practices doc is worth reading.
Community Picks
Gemini CLI Tips & Tricks
About 30 pro-tips for Gemini CLI. Covers basics like persistent context via GEMINI.md, custom slash commands, MCP server extensibility, and multimodal input with @ syntax. Advanced section includes checkpointing for undo, headless scripting for automation, multi-directory workspace support, and using your entire system $PATH as the AI's toolset. Read more.
Codex, Opus, Gemini try to build Counter Strike
Gemini 3 Pro, Codex Max 5.1, and Claude Opus 4.5 tackle building a basic Counter Strike. The game had to be a 3D UI and it had to be multiplayer. Opus 4.5 won on frontend, Gemini 3 Pro did best on backend, Codex Max 5.1 took second place on most criteria. Read more.
How I Use Claude Code on My Phone with Termux and Tailscale
20-minute setup for running Claude Code on mobile devices. Uses SSH, Termux, Tailscale, and tmux to connect to a persistent desktop environment. Read more.
That’s it for this week. I’ll be back next Monday with the latest agentic coding updates.