Welcome to the first issue of Agentic Coding Weekly. Here are the updates on agentic coding tools, models, and workflows worth your attention for the week of Nov 16 - 22, 2025.
Tooling and Model Updates
Gemini 3 Pro
Google released Gemini 3 Pro, their new flagship model. It now sits at or near the top of most public coding benchmarks: LMArena WebDev, LiveCodeBench Pro, and Terminal Bench 2.0, and is close to SOTA on SWE-bench Verified.
Notable details:
Ships with a higher default “thinking” level enabled
Knowledge cutoff: January 2025
Pricing is $2 / $12 per million for input / output tokens. Between Gemini 2.5 Pro ($1.25 / $10) and Claude 4.5 Sonnet ($3 / $15).
Gemini CLI has already added support for Gemini 3.
Read the model card for more details.
Antigravity (Google’s AI IDE)
Google also launched Antigravity, their AI-powered IDE. It's a VS Code fork, though they don't advertise that.
The launch demo showed a pretty interesting workflow loop:
Generate UI mockups with their Nano Banana Pro model
Let agents implement the code and verify via browser use
Iterate on the design and the implementation by leaving comments directly on screenshots to drive changes
If they execute this well, might be worth trying over Claude Code for prototyping. Early access is rough though. Users report burning through credits in a few prompts with no way to add payment yet. Read the announcement.
GPT‑5.1‑Codex‑Max
The day after Gemini 3, OpenAI released GPT‑5.1‑Codex‑Max, their new SOTA coding model.
Details:
Beats Gemini 3 Pro and Claude 4.5 Sonnet on SWE-bench Verified and Terminal Bench 2.0
Uses ~30% fewer “thinking” tokens than GPT‑5.1‑Codex at medium reasoning effort
Built for long-running and detailed tasks using “compaction” across multiple context windows
Can operate over millions of tokens for a single task
Currently, it’s available on ChatGPT paid plans but not yet via API. Read the announcement.
Workflow of the Week
Try smartcommit for git commit messages. It's a middle ground between writing detailed messages yourself and letting AI generate them blindly from diffs.
How it works:
It inspects your staged changes
Asks you a few clarifying questions (why, not just what)
Generates a structured commit message using both your intent and the actual diff
Works with both remote LLMs (using OpenAI API) and local ones (using Ollama).
Community Picks
What if you don't need MCP at all?
An argument against monolithic MCP servers. The alternative: minimal CLI tools executed via Bash.
Author replaced MCP with small Node.js scripts wrapped in Bash commands since agents already understand Bash. Token consumption dropped from 18k to 225 in his tests.
Recommendation: only add essential MCP servers. Better yet, implement needed tools yourself to keep context clean. Read more.
How AI will change software engineering
Martin Fowler on The Pragmatic Engineer podcast discussed the shift from deterministic systems to non-deterministic LLM workflows. Read more.
The key insight: Vibe Coding is detrimental to long-term software quality because it eliminates the crucial learning loop for developers.
If you’re not looking at the output, you’re not learning. The thing is that so much of what we do is we come up with ideas, we try them out on the computer with this constant back and forth between what the computer does with what we’re thinking. We’re constantly going through that learning loop. And Unmesh’s point, which I think is absolutely true, is you cannot shortcut that process. What LLMs do is they just kind of skim over all of that, and you’re not learning. And when you’re not learning, that means that when you produce something, you don’t know how to tweak it, modify it, evolve it, and grow it.
Command Lines
The AI coding market is splitting. "Hands-off" tools like Lovable target non-engineers doing prototyping. "Hands-on" tools like Claude Code serve professional engineers using AI as pair programmers.
The analogy: AI tools are to boilerplate what compilers were to assembly. They handle grunt work so you can focus on higher-order problems. Read more.
That’s it for this week. I’ll be back next Monday with the latest agentic coding updates.
