• Solan Sync
  • Posts
  • GOOGLE’S SECRET KITCHEN 🧑‍🍳🤖

GOOGLE’S SECRET KITCHEN 🧑‍🍳🤖

WHAT GOOGLE, MISTRAL, SHOPIFY, AND OPENAI QUIETLY SHIPPED THIS WEEK

Author

Solan Sync
December 13, 2025

If you’re building with AI, this week felt like one of those moments where a lot shipped quietly — but none of it was small.

Google Labs keeps experimenting behind the scenes, Mistral stepped into the terminal, Shopify simulated real customers, and OpenAI raised the bar on security claims. Here’s what actually matters, in plain English.

🧪 GOOGLE LABS IS SHIPPING LIKE A STARTUP AGAIN

Google Labs has been unusually busy lately — not with flashy announcements, but with tools that solve very real problems for builders and creators.

A few highlights:

  • Scheduled Tasks let you automate recurring work without duct-taping scripts together.

  • Jules, Google’s async coding agent, now offers proactive suggestions instead of waiting to be prompted.

  • Pomelli helps marketers generate and animate creative assets faster.

  • Stitch explores design ideas visually, making early-stage design less painful.

None of these feel like moonshots. They feel like tools teams might actually use day to day — which is new, and refreshing, for Google.

💻 MISTRAL ENTERS THE TERMINAL (AND DOES IT RIGHT)

Mistral released Mistral Vibe, an open-source CLI that lets you work with models directly from the terminal.

Why people are paying attention:

  • It’s minimal and hackable.

  • The license allows for real customization.

  • You’re not fighting the tool to extend it.

Alongside the CLI, Mistral also released Devstral 2 and Devstral 2 Small. Performance-wise, they’re competitive with larger models like GLM 4.6 and Kimi K2 — but noticeably smaller.

That matters if you:

  • Want to run models locally

  • Don’t have unlimited GPUs

  • Care about speed and cost over hype

You probably won’t run these comfortably on a MacBook Air, but for small teams with a few GPUs, they’re very real options.

🛒 SHOPIFY’S SIMGYM: FAKE CUSTOMERS, REAL INSIGHTS

Shopify launched SimGym, and it’s one of those ideas that makes you wonder why it didn’t exist earlier.

SimGym creates simulated customers that:

  • Browse your store

  • Complete tasks

  • Reveal friction points

  • Let you run A/B tests without risking real revenue

For ecommerce teams, this means testing ideas before shipping them to real users — no traffic required.

It’s part of Shopify’s broader Winter launch, which leans heavily into AI-native tooling.

🔐 OPENAI SAYS ITS MODELS ARE GETTING “CYBERSECURITY-READY”

OpenAI published a blog claiming its latest models now reach a “high” level of cybersecurity capability.

One data point stood out:

  • GPT-5.1-Codex-Max scored 76%, compared to 27% for GPT-5 back in August.

That’s a massive jump — and signals where model evaluation is heading next. It’s no longer just about writing code or passing benchmarks, but about whether models can reason safely in high-risk domains.

🌐 A few things worth your time this week: 📚

🧰 TOOLS WORTH A QUICK LOOK

A few other tools mentioned this week that are genuinely useful:

  • DeepSky — AI superagent for founders doing strategy and research

  • Mintlify Autopilot — keeps documentation in sync with your code

  • Scouts by Yutori — tracks topics across the web and summarizes them

  • Orchids — an IDE built for “vibe coding”

  • Detail.dev — deep scans your codebase for bugs

  • Figma’s new AI tools — practical image editing for UI work

🧠 WHY THIS WEEK MATTERS

If you’ve been asking:

  • What is Google actually building with AI right now?

  • Are there serious open-source alternatives to closed models?

  • How can AI help me test products before real users see them?

This week quietly answered all three.

No hype cycles — just tools moving closer to real workflows.

Reply

or to participate.