Welcome back. It’s been a massive Friday for devs. Between Z.ai’s new model outperforming Fable 5 and DeepSeek dropping a new open-source coding harness, we have plenty to dig into. But beyond the hype of these new launches, we’re breaking down the essential advice for engineers in 2026 that most people are overlooking. Read below to dive deep. 

Also: The creator of Claude Code shares his app maintenance workflow, how Meta’s layoff strategy backfired, and Cursor teases its new GitHub rival.

Google's new ‘workhorse’ model gets a serious coding upgrade: The search giant just unveiled Gemini 3.7 Flash, calling it their most capable workhorse model yet for coding and agents. It delivers more accurate first-pass code with better performance on multi-step tasks and tool calls. The model now powers Gemini Spark, the company's personal AI agent, and API prices have been temporarily cut in half. See it in action.

Z.ai drops a new model that rivals OpenAI and Anthropic in coding: The Chinese AI lab just unveiled GLM 5.3, which is built on the same 743B base as GLM 5.2. This time, the improvements come entirely from scaled up post training. The model surprisingly beats Fable 5 and DeepSeek V4 on Terminal Bench 2.1. Weights are expected to ship in about two weeks, but you can try it now via the Z.ai API, GLM Coding Plan, or ZCode.

DeepSeek drops an open-source rival to Claude Code: The Chinese AI startup just released DeepSeek Harness v0.1, a new open source coding harness that gives developers an alternative to paid ones. The lab also unveiled DeepSeek-V4-Pro, an updated flagship model heavily optimized for agentic workloads. But this comes at a cost: if you’ve been using V4 through DeepSeek's API, you might soon pay significantly more for it as the company plans on raising the prices. Find more details on the cost and try the harness out here.

AI agents that answer phone calls work great in demos, then fall apart on real calls. The reason is usually the audio.

Speech-to-text handles noise just fine. What breaks it is a second person talking near the caller. A noise filter can't tell voices apart. The transcript comes out wrong, and the agent acts on it: wrong answer, wrong booking, "could you repeat that?"

Krisp Voice Isolation 2.5 sits in front of your speech-to-text and isolates the caller's voice. When other voices are talking, word errors drop from 36% to 11%.

Try it on your own calls →

An outdated test. Phil Chen, a founder who’s worked at OpenAI, Google DeepMind, and Scale AI, says that the standard technical interview is basically dead. At his AI-native company, devs barely write code by hand anymore. LeetCode and old school system design interviews don’t really show how someone performs on the job anymore. If the skill you're testing is already automated, you're mostly just checking how well they can use the tool. 

The useful metric. Chen’s team focuses on two things agents still can’t do alone: pick the right problem and solve it precisely. That also shapes his advice for engineers trying to stay valuable as agentic coding changes the game:

Chen's full essay is definitely worth a read, and if you're curious about how an AI-assisted interview actually feels, this coding track lets you try it out with some timed practice repos. 

Soniox Text-to-Speech lets you create extraordinary voice experiences in 60+ languages with expressive control, exceptional precision, instant voice cloning, and ultra-low-latency streaming.

Direct whispers, laughter, hesitation, excitement, and emotion, or precisely speak phone numbers, email addresses, technical terminology, and multilingual text.

Built for voice agents, global products, customer support, media, and any application where speech needs to sound human and get every detail right.

$0.70 per generated hour.

Try it now

Building AI coding workflows from cloud to local (DeepLearning.ai): You’ll learn how to build tailored AI coding workflows from cloud to local. This tutorial covers how to switch coding agents, route tasks across specialized models, and deploy local LLMs to achieve optimized speed, reduced costs, maximum privacy, and better code quality.

CodeBurn: You can now track exactly where your AI coding budget is going. This tool flags waste, handles the fix, and shows you exactly how much you saved. 

How to optimize your AGENTS.md file for AI coding agents: Agents usually start out great but quickly get bogged down by context bloat and vague instructions that lead to hallucinations or broken loops. This guide teaches you how to dial in your AGENTS.md file to keep things on track. 

You ask your agent a question and get five paragraphs of jargon back. A senior developer got tired of it and built “/show-me”, a skill that makes the agent answer with small visuals instead of component trees, call stacks, file layouts, diagrams.

You can also use these same techniques to walk through large diffs after the fact so you know exactly what to focus on during your review.

P.S. Get 50+ AI coding hacks for Claude Code, Cursor, and Codex here.

Check out the full timeline of what's changed since a group of OpenAI researchers dropped "AI 2027," a roadmap predicting what AI would look like over the next two years, where 19 out of 53 tracked predictions are now either confirmed or ahead of schedule.

Grow customers & revenue: Join companies like Google, IBM, and Datadog. Showcase your product to our 300K+ engineers and 150K+ followers on socials. Get in touch.

You can also reply directly to this email if you have suggestions, feedback, or questions.

Until next time — The Code team