Happy Wednesday! Welcome back to the AI Edge (by Miles Deutscher) weekly newsletter.
Google's most capable AI model yet is live now - this is your full guide.
For the first time ever, Google has released a truly great AI model. Argon 4 can handle video and audio inputs, build custom Google tools & much more.
This week, we're breaking down how to use the new Argon 4 model, optimize it for greater AI productivity, and test a few practical workflows.
After our main guide, be sure to stick around for our Looking Ahead section - we have some very interesting things this week.
Including: How to get Hermes agent to call you, setting Grok Bot as your action button on iPhone & more.
IN TODAY'S READ
Table of Contents
WEEKLY BUILD & WORKFLOW
Gemini 4 Argon Explained in 30 Seconds
Gemini 4 Argon is Google's latest flagship AI reasoning model - think Google’s response to Fable.
We’ve found the model is built for these tasks
→ Deep reasoning: Coding, planning, research, SPEC design
→ Long context: One of Argon’s superpowers.
→ Multimodal analysis: Working with text, images, audio, video, and PDF files.
→ Agentic work: Actually handling agentic tasks in Google Workspace.
TL;DR: Argon is designed for work that needs deeper analysis, more context, or multiple steps. It isn't necessary for every everyday AI task.
Argon vs Other Models
Model | Best For | Choose It If... |
|---|---|---|
Flash-Lite | Simple, quick requests | Fast answer without much reasoning |
Flash | Everyday work | Balance of speed and capability |
Pro | More demanding work | General deep research |
Argon | Difficult, multi-step tasks | Deep coding, SPEC design |
How to Access Gemini 4 Argon
You can use Argon in a few different ways
Route 1: Directly in Gemini
Visit https://gemini.google.com and sign in with your Google account.
Argon is available to eligible Google AI Pro and Ultra subscribers.
Step 2: Select Argon
At the bottom of the chat box, open the model selector.
Choose Argon from the available models.
Step 3: Choose Your Thinking Level
Gemini provides three thinking levels:
Low: Faster responses for straightforward tasks.
Medium: A balance between reasoning depth and speed.
High: More reasoning effort for complex, multi-step work.
Sample prompt to test:
You are reviewing a business decision.
A small company sells a product for $50.
Its variable cost is $30 per unit,
and monthly fixed costs are $2,000.
The company is considering lowering the selling price to $45.
Calculate:
1. Current break-even sales volume
2. New break-even sales volume
3. Monthly profit at 150 units sold,
under both prices
4. What changes financially
5. What information is missing before recommending the price reduction
Show the calculations.
Do not assume that lowering the price will automatically increase demand.
Optimizing Argon
Four tips we’ve found helpful for this new model:
1. Attach Full Context
Argon can handle large inputs, so make use of that when the task needs it.
For example, if you're reviewing a business proposal, you can literally attach dozens of contextual documents.
2. Reverse-Engineer Final Outputs
This is an underrated hack.
With Argon, you should always focus on the end state, then use the model to reverse-engineer the exact steps to get [final output].
For example:
"Review these three proposals and return a comparison table, the main risks, your recommendation, and anything that still needs verification."3. On Thinking Levels
Argon supports Low, Medium, and High thinking levels.
For regular use, Medium is perfectly fine.
You can also use High when model-stacking with models like Fable for the best possible outputs.
4. Iteration Loops
Use a separate review instruction for this (copy the prompt):
Review your previous answer.
Check the calculations and conclusions against the original information.
Identify:
1. Any factual errors
2. Unsupported assumptions
3. Missing information
4. Contradictions between sources
5. Conclusions that are stronger than the evidence supports
Correct any errors you find.
Clearly state what still needs independent verification.
Do not introduce new information unless you identify its source.Argon Workflows Testing
Three valuable workflows for research, data analysis, and building.
Workflow 1: Research → Finished Work (end-to-end)
Imagine you're reviewing whether your business should launch a new product.
You have a sales report, customer feedback, and competitor research.
Instead of asking Gemini to summarize each file separately, we'll use Argon to review them together and prepare a decision brief.
Step 1: Upload Your Files
Open your Argon chat in Gemini.
Click Add files → Upload Files and attach the relevant documents.
For this example, we'll use:
→ A sales report
→ Customer feedback
→ Competitor research
Gemini supports multiple file uploads in one prompt, subject to your plan's limits.
Step 2: Analyze all Evidence
Paste this prompt into your Argon chat after attaching the files.
You are helping evaluate a proposed product launch.
Review the three attached documents:
1. Sales report
2. Customer feedback
3. Competitor research
Your goal is to determine whether the available evidence supports launching the product.
First, extract the most important findings from each document.
Then:
1. Identify evidence supporting the product launch.
2. Identify evidence against it.
3. Compare the findings across all three documents.
4. Flag contradictory figures, claims, or assumptions.
5. Identify information that is missing or unreliable.
6. Recommend one of three options: launch, test further, or delay.
Explain the recommendation using the evidence provided.
Reference the source document for every important claim.
Do not invent figures or treat assumptions as confirmed facts.
Finish with a concise decision brief.Step 3: Turning Findings Into a Presentation
Once you've reviewed the decision brief and confirmed its important claims, you can turn it into a presentation.
Gemini Canvas supports creating and editing slides.
To use it, open Canvas through Gemini's Add files menu and provide the verified brief as context:
Create a six-slide presentation from the verified decision brief below.
[PASTE VERIFIED DECISION BRIEF]
Audience: Business owners or project decision-makers.
Structure:
1. The proposed decision
2. Evidence supporting it
3. Evidence against it
4. Key financial or operational risks
5. Recommended next step
6. Open questions and required actions
Keep each slide focused on one main point.
Use simple, professional language. Preserve the original figure and source references.
Do not invent data or add claims that were not established in the decision brief.Example output
Workflow 2: Multimodal Prompting
Argon can also work with audio and video inputs.
This is useful when you've recorded a meeting, interview, etc.
Instead of directly transcribing audio, you can attach it directly into Argon.
How to Use It
Open Gemini with Argon selected.
Click Add files and upload a supported recording.
Make sure the file is within your plan's upload limits.
Paste the following prompt below:
You are reviewing [recorded meeting]
Analyze the attached recording.
Extract:
1. The main subjects discussed
2. Decisions that were confirmed
3. Proposals that were discussed but not approved
4. Action items and assigned owners
5. Deadlines mentioned
6. Questions that remain unresolved
Separate confirmed decisions from suggestions and opinions.
Where possible, provide timestamps for important moments.
If you cannot reliably identify a speaker, owner, or deadline, mark it as unclear.
Do not invent commitments.
Return:
- A short executive summary
- A decision log
- An action table
- Three suggested follow-up questionsExample Output
Workflow 3: Build a Custom Tool With Gemini Canvas
Building custom tooling with Argon
Let's use a simple example:
You want a calculator that shows how changing your prices or costs affects profit.
Use Gemini Canvas to build a small interactive tool.
How to Use It
→ Open Gemini and select Argon.
→ Open Canvas through the Add files menu.
MIDWEEK EDGE: Manually curated list of the most important weekly AI news updates
Weekly AI News Recap
→ Grok Voice 3.0 Takes the #1 Spot in Voice AI. SpaceXAI released its latest voice model, ranking #1 on Artificial Analysis's Speech-to-Speech Index.
It responds in under a second.
Try it in Grok Bot's voice mode or through the Voice Agent API.
→ Meta Launches Muse Wonder and Muse Wonder Full. New models that introduce Active Memory.
→ OpenAI Brings GPT-6 and Intelligent UI to ChatGPT. GPT-6 is rolling out across free and paid ChatGPT plans, alongside interactive responses that can generate calculators, diagrams, comparison tools, and more.
→ Anthropic Releases Claude Haiku 5.5. Anthropic just released its fastest and cheapest small Claude model yet.
→ OpenAI Introduces GPT-6.1 Sol Ultrafast. OpenAI's new Ultrafast mode can generate tokens up to 8× faster than Sol.
LOOKING AHEAD: Manually curated list of useful AI tools, skills, workflows & more
Tools and Research Worth Your Time
→ How to Build a Company in the AI Era
An eight-stage (realistic) roadmap for building a business in 2026.
→ Hermes Can Now Call You on WhatsApp.
A new community plugin lets your Hermes agent call you when a scheduled task needs attention, hold live voice conversations, and send a summary afterwards.
→ Talk to Grok Bot With One Button on Your iPhone.
Grok Bot now supports the iPhone's Action Button, letting you start a voice conversation without opening the app.
Settings → Action Button → Shortcut → Grok Bot → Start Voice Chat.
→ 10 Ways to Use Grok Bot Inside 𝕏
Grok Bot can now plug into your 𝕏 account.
→ AI-Related Employment Is Surging
LinkedIn estimates that +750,000 AI-linked jobs have been added in the U.S. since 2023. The biggest categories include:
CONTENT PIPELINE: Latest content from AI Edge
New From AI Edge
YouTube: How 5 AI Agents Run My Computer While I Sleep. Watch now →
𝕏: Opus 5.5 for Motion Design Read now →
Skool: Free prompts, cheat sheets, PDF guides & more. Join here (100% free) →
See you next Wednesday,
The AI Edge Team
Want to sponsor or partner with us? Reach out at [email protected].