Hi {{first_name}}!
Last week I mentioned GPT-5.2 launched, but didn't get into the details. This week I'm sharing the results from several people testing it. I’ll dive into what works for real business use versus what just looks impressive in demos. Plus Google had another massive week with multiple exciting launches and new capabilities, including the release of Super Gems!
This week we're covering:
Google's massive week - Real-time translation, upgraded Deep Research agent, NotebookLM integration, Disco browser, and more
New Google Super Gems - Google’s starting to roll out Super Gems which are like custom GPTs with built-in automation capabilities
GPT-5.2 breakdown - The pros and cons of ChatGPT’s newest update
Other AI developments - Manus 1.6 mobile agents, Claude Skills and Memory worth revisiting
OK, let's get into it!
New and Noteworthy
ChatGPT Images Gets a Major Upgrade: OpenAI just released GPT Image 1.5 with some promising improvements. They're claiming the big upgrade is precise edits that preserve what you want to keep (IE lighting, composition, facial likeness staying consistent across edits instead of changing randomly). It's also supposedly 4x faster and the instruction following is more reliable, so when you ask it to "only change the background" it should actually do just that. This is available now for all ChatGPT users and in the API. Nobody seems to be saying it’s better than Google’s Nano Banana Pro model but for those of you that work within ChatGPT throughout the day it’s a nice upgrade to have!
OpenAI Ends Stock Compensation Waiting Period: OpenAI eliminated its 6-month cliff before new employees can access stock options. This is the latest move in the heated battle to attract and retain top AI talent.
Google's Real-Time Translation with Headphones: Google is rolling out real-time translation that works through any headphones. Powered by Gemini AI, it handles language, context and slang across 70+ languages, making international business meetings and travel conversations significantly smoother.
OpenAI GPT-5.2 Release: OpenAI dropped GPT-5.2 with expert-level capabilities across most knowledge work. The long-context handling is substantially improved and it can now process and remember massive amounts of information without losing the thread. I'm covering this in detail in the spotlight section below.
Google Labs Launches Disco Browser: Google's new experimental browser uses Gemini 3 to look at your open tabs and chat history, then builds interactive web apps to help you finish whatever you're working on. No coding required. macOS only for now, waitlist required.
Google Denies Gemini Ads: Google pushed back on reports claiming Gemini would get ads in 2026, calling the claims from anonymous sources inaccurate and unfounded. I’m guessing this reaction is due to the backlash ChatGPT saw with their advertising experiments.
A more powerful Gemini Deep Research agent: Google released a significantly upgraded version of their Deep Research agent, now powered by Gemini 3 Pro with vastly improved web search capabilities. It's now available via API so developers can embed Google's autonomous research capabilities directly into their own applications. Previously, Deep Research was only accessible inside the Gemini app. This agent can now navigate deeper into websites for specific data, achieves state-of-the-art benchmark results, and will soon be integrated into Google Search, NotebookLM, and Google Finance. For businesses, this means the research agent that was limited to consumer use is now something you can build into your own workflows and tools.
NotebookLM Integration in Gemini: Google added NotebookLM directly inside Gemini so you can work with your research notes without switching apps. The AI gets more context from your notebooks and gives better results. Limited access for now, but this makes it easier to use powerful AI tools in one place. NotebookLM is one of my absolute favorite AI tools and they’ve rolled out lots of cool updates recently, so if you haven’t experimented with it lately you should definitely take a look!
Manus 1.6 Release: Manus launched version 1.6 with major upgrades including Max reasoning capabilities, mobile builds, and Design View creation. This update brings more powerful AI agent functionality to mobile devices and improves how developers can build and test applications. Are you using Manus? If so I’d love to hear how!
Claude Skills and Memory Worth Revisiting: Claude’s new “Skills” capability lets you package workflows and company knowledge that Claude can automatically use when needed, eliminating repetitive context explanations. Skills take about 20 minutes to set up and work well for recurring tasks. Another great upgrade in Claude has been the addition of “memory” (available on Pro and Max plans) which is similar to ChatGPT in that it keeps context and gets to know you better so you don't have to rebuild the same background in every conversation. Both features have been available since October but are worth exploring if you're doing the same tasks repeatedly in Claude or working on ongoing projects.
Google Labs Launches Disco Browser: Google dropped an experimental browser called Disco on December 11th. It features "GenTabs" powered by Gemini 3 that turn your open tabs or prompts into custom interactive web apps. Think: converting a cluster of research tabs into an actual trip planner or study tool. This isn't meant to replace Chrome—it's Google testing future browsing ideas. Currently macOS only with waitlist access at labs.google/disco. Worth watching as successful features may eventually land in Chrome.
Google Super Gems: Turn Ideas Into Working Business Tools
Google just shared something new called Super Gems and they really are SUPER!!!! ;)
Super Gems are an advanced evolution of Gemini Gems integrated with Opal workflows, enabling users to build custom AI agents, mini-apps, and automated tools for complex tasks.
If you have used regular Gems before, you know they are like custom assistants or custom GPTs for specific tasks. Super Gems take that significantly further.
To keep this simple: you can now build your own functional tools without writing a single line of code.
Real world examples
One idea could be to use these for:
Research assistants: These can gather info from different places and format it exactly how you want.
Data tools: You can connect your Google Sheets to create visual reports automatically.
Content systems: These help you move from a rough outline to a finished draft in one spot.
Client onboarding: You can build a system that handles new client info and updates your CRM.
How to get started
Getting access is pretty straightforward:
Head over to gemini.google.com.
Open the Gems manager in the left sidebar.
Look for "Build AI apps" or "Gems from Labs."
If you do not see it yet, check back in a few days. It is rolling out to everyone now.
I’ll be sharing more insights and recommendations in the Ampra Circle community as I experiment with Super Gems over the next couple weeks.
The GPT-5.2 Hype Is Real... But So Are The Mistakes
OpenAI released GPT-5.2 last week calling it "the smartest generally-available model in the world," and within hours people were posting wild demos across X. Some look like genuine breakthroughs. Others look impressive until you check the actual outputs.
Here are the most business-relevant capabilities I've seen tested so far:
It helps experts think through complex problems. An immunologist asked GPT-5.2 Pro for the most innovative questions needed to understand the immune system, using it as a thinking partner to surface research directions he wouldn't have considered. This matters for strategy sessions where you need someone to challenge your assumptions and ask the hard questions.
It builds complete systems in single prompts. Developers are generating full 3D graphics engines with interactive controls in one shot, no back-and-forth refinement needed. For businesses, this means building internal dashboards, tools, or prototypes significantly faster than the usual iterative process.
The reasoning capability is substantially upgraded. In detailed testing, GPT-5.2 independently derived advanced mathematical conditions and used code to search for counterexamples. It scored 92.4% on complex science questions. This translates to better analysis, deeper problem-solving, and more reliable outputs for technical work.
It handles long-context work without degrading. Multiple testers showed GPT-5.2 maintaining quality over huge inputs where previous models would start repeating themselves or losing coherence. Your multi-page reports, lengthy client communications, and extended strategy documents get processed with consistent quality throughout.
Its multimodal capabilities still lag behind Gemini. While GPT-5.2 excels at text-based work, video analysis remains a weakness. A developer tested both models on the same video clip: Gemini delivered a detailed transcript, pulled the core concept, and contextualized the discussion, while GPT-5.2 failed to even load the video source. For businesses relying on video content analysis, meeting recordings, or training materials, Gemini's multimodal capabilities are still far superior.
But it still makes expensive mistakes. Someone asked GPT-5.2 to build a complete financial model: 5,000+ cells, 18 interconnected sheets, dynamic scenarios, sensitivity tables, and beautiful charts. It generated everything. But as the creator said, "none of the numbers added up", the DCF valued his lemonade stand at $2.7 billion. The model can build impressive-looking outputs that are fundamentally wrong.
GPT-5.2 represents a real step forward for business applications, especially for document analysis, strategic thinking, and building tools. But the lemonade stand example is your reminder that you still need to verify outputs, particularly for financial models, data analysis, or anywhere accuracy matters more than speed. When you are working on something were accuracy is paramount i’ll remind you to use the “deep thinking” mode and ask it to take longer. The more compute and the more time AI takes the more likely it will be accurate.
Worth exploring if you're working with large document sets, doing strategic planning, building internal tools, or need a thinking partner for complex problems. The standard Plus version ($20/month) handles most business needs, though heavy users might benefit from Pro ($200/month).
If you've tested GPT-5.2 yourself, hit reply and let me know what you're seeing. I'm curious which use cases are actually working for real businesses versus just looking impressive in demos.
If you spotted a quick win here, or if you’re just wondering where to start with AI and automation, please feel free to reach out. I’m always happy to share insights and help you figure out the best first step.
Next week, keep an eye out for a special Christmas Edition. Based on the positive feedback from my Thanksgiving edition, I’m putting together another fun holiday edition to celebrate the season!
Cheers,
Julien
PS: If this was valuable, feel free to forward it to a friend or colleague. They can sign up at www.ampra.ai/join-our-newsletter.