Issue #: 013
Send date: Oct 19, 2026
Subject line: Read the contract, write down the process
Preheader: Gemini 4 Argon is finally open. Here’s the work it’s actually good for.

The most expensive documents in your business are the ones nobody reads, and the ones nobody wrote.

World Commerce & Contracting puts the average loss from poor contract management at almost 9% of annual revenue. That money disappears through cost overruns, invoicing errors, missed entitlements, and disputes nobody saw coming. Most of it traces back to terms someone skimmed at signing: the auto-renewal with a 60-day notice window, the net-90 payment clause, the liability cap that only protects one side.

The second leak is quieter. It’s every process that lives only in your head. If you can’t hand a task to someone else, you can’t sell the hours it eats.

Last week we left Gemini 4 Argon out because you couldn’t use it yet. Now, you can. It isn’t the best model at everything; more on that in the Stack Update. But it ranks first on the Vals Index, an independent test of finance, legal, tax, and coding work. Google also reports leading scores on legal drafting and on understanding long video.

Put plainly: reading a 30-page contract and watching a 20-minute screen recording just got cheap. Those are the two jobs operators skip most.

So this week we’re putting it on both. Beginners get every contract read before it’s signed. Operators get their recurring work out of their heads and into documents someone else can run.

● Beginner: read before you sign · ● Operator: write it down so someone else can do it

MOVE 01 · BEGINNER
Text. A red-flag memo is a one-page list of the clauses in a contract that could cost you money.

Objective: you finish this move with a memo on the next contract you’re about to sign, so you know which three clauses to question before your signature makes them permanent.

Upload the PDF to the Gemini app with Argon selected. Paste this:

Tools: Gemini (Argon)

MOVE 02 · BEGINNER
Workflow. Check which Google account you’re using before you upload anything.

Objective: you finish this move knowing whether the contract you upload can be read by a human reviewer at Google.

This matters more than the prompt. On personal Google accounts, Google’s Gemini privacy notice says human reviewers read some chats, and it tells you not to enter confidential information. On eligible Google Workspace accounts, chats and files aren’t human-reviewed or used to train models.

If you only have a personal account, replace party names and dollar amounts with placeholders before you upload. Also check the contract itself: if it has a confidentiality clause, read that clause first. This is Issue #004’s Move 09 rule applied to a specific tool.

Tools: Gemini · Google Workspace

MOVE 03 · BEGINNER
Text. A pushback email asks the other side to change two or three specific clauses, with a reason for each.

Objective: you finish this move having sent one request to change a term, instead of signing as-is.

First, open the PDF and check every clause the memo rated “pushback” against the actual wording. Models misquote clause numbers, and that’s the Issue #004 Move 03 failure. Then ask Argon: “Draft a short, friendly email asking to change only the pushback items, with one business reason for each.”

Know where this stops. On Harvey’s legal-agent test, Argon led the field at 19.6%, which Google reports is far ahead of its rivals. That’s still nowhere near a lawyer. For anything high-stakes, take the memo to your lawyer. You’ll pay for a shorter conversation.

Tools: Gemini (Argon)

MOVE 04 · OPERATOR
Workflow. A process recording is a screen capture of you doing one recurring task while you explain what you’re doing out loud.

Objective: you finish this move with one recording of a task you do every week, narrated well enough that the reasoning is on tape, not just the clicks.

Pick the task that eats the most hours and that you’d most like to hand off. Record it with Loom or your computer’s built-in screen recorder. Do one process per recording. Say the why out loud: “I check this column first because refunds show up here.” The why is the part a new person can’t guess.

Close client tabs, or run the task on test data. Anything visible on screen ends up in the upload.

Tools: Loom · Gemini

MOVE 05 · OPERATOR
Text. An SOP (standard operating procedure) is the written version of that recording: numbered steps someone else can follow without you.

Objective: you finish this move with a draft SOP where every step links back to a moment in the recording you can check.

Upload the recording to Gemini with Argon selected and paste this:

The flags are the important part. They’re Issue #004’s Move 05 rule: give the model permission to say “I don’t know” instead of quietly inventing a step.

Tools: Gemini (Argon) · Notion

MOVE 06 · OPERATOR
Workflow. A cold test means handing the SOP to someone who’s never done the task and giving them no other help.

Objective: you finish this move with an SOP a stranger has actually completed, which is the only proof it works.

Give it to a VA, a contractor, or a new teammate. Every question they ask is a missing step. Paste their questions back into the same chat and have Argon revise the document. Store the final version in Notion next to your context file from Issue #001.

Once you’ve done this three times, you have a sellable service. Recording a client’s processes and returning tested SOPs is Model 01 from Issue #002, priced by the project.

Tools: Gemini (Argon) · Notion

Your first red-flag memo, on a contract you’ve already signed. About 20 minutes.

Start with a signed contract, not a new one. You already know where its problems are, so you can grade the model. That’s the whole point of a first run.

Check the account. Personal account: replace names and amounts with placeholders. Workspace account: go ahead.

Pick Argon. Open the Gemini app and choose Argon from the model menu. It’s available on Google AI Pro and Ultra.

Upload and paste the Move 01 prompt. Read the whole memo before you judge it.

Grade it. Check three flagged clauses against the PDF. Then write down one problem you already knew about that it missed. That miss tells you more about where to trust it than the hits do.

Keep the prompt. Save it as a text-expander snippet, like the pre-mortem from Issue #004. Next time, it takes two keystrokes.

The slot’s open this week. We won’t invent a question to fill it.

If you run a red-flag memo or record your first SOP this week, tell us where it went wrong: a clause it misread, a step it invented, a recording it couldn’t follow. Include what you tried.

Send it to [email protected]. One question gets answered every Monday, and founding member questions go first.

BUY · Gemini 4 Argon, for reading long documents and recordings. WATCH · Argon, as your coding agent.

Where it’s strong: it ranks first on the Vals Index, an independent test of professional work, ahead of both Claude 5.5 models. Google’s own reported results lead on legal drafting, financial research, and long-video understanding. Google also says it’s trained to pause before irreversible actions. Inside Google, more than 100,000 employees use it every month.

Where it breaks: on Artificial Analysis’s broader index, it ties GPT-6 Astra and sits below Opus 5.5 and Sonnet 5.5. On command-line coding, Google’s own comparison has Opus ahead. One early tester called it really good but not the best, and said Antigravity isn’t close to Claude Code or Codex. The headline output limit from launch was later reframed as a combined limit that only works through a special API setup. You won’t see it in the app. And Google’s internal examples are its own reports, not independent tests.

Use it to read what you’d skip. Keep building in the tools you already trust.

The rest of the shelf: three things from the last week worth your attention.

BUY, if you sell on Instagram · Muse for Small Business. WATCH · Muse Wonder, the model. Issue #008 covered Muse as a shopping agent. The business side is newer. Meta says Muse can connect your Instagram professional analytics, Facebook Pages, and Meta ad accounts in a few clicks, plus Shopify, Stripe, QuickBooks, Klaviyo, and Canva. It works from your own account data, which ChatGPT and Claude don’t have. Meta also says nothing publishes, sends, or spends without your approval. Keep it that way. Muse Wonder, the new model, is a different product. The API version scores 52 on Artificial Analysis. The Full version that wins Meta’s charts is only in Muse Code, and Meta says Wonder will power parts of the Muse app later.

WATCH · ChatGPT in Chrome, Work mode. Since October 15, the ChatGPT extension can read your open tabs. In Work mode, it can control them to send messages, fill forms, and download files, on every ChatGPT plan. OpenAI’s own docs say to treat page content as untrusted. Run it only in the clean Chrome profile from Issue #005.

FIX THIS WEEK · OpenAI API retirements. Older GPT-3.5, GPT-4, and o-series models, GPT-4.1 nano, and the first GPT-4o snapshot retire October 23. GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini follow on January 14, 2027. Open every Zapier, Make, or n8n step that calls OpenAI and check the model field. An automation you set up in 2024 is the one most likely to break without warning.

A weekly dispatch for founders, freelancers, and operators who treat AI as a business tool, not a hobby. Every Monday, one question: what changed in AI this week that you should actually do something about?

Anthony | Founder of Uncover AI