Anthropic's new agent messages you first, plus three things to act on this week
Mike, from Automate & Hustle
Anthropic opened a research preview of Claude Duet, an agent that sends you a morning brief and messages you when something needs you. That leads, with four steps to copy the idea today. Three more stories follow, each with steps for this week.
🧢 Automate & Hustle - This Week
🤖 Claude Duet messages you before you ask
🏦 US banks are blocking AI agents that use your password
🧮 GPT-6.1 Terra costs a tenth of Astra
📈 Anthropic's IPO filing: $16.3B in six months
🧩 Five habits for putting agents on a short leash
🧰 Three tools worth a look
🎬 Five builds from my reels this week
🗞️ This Week in AI - Four Stories.
🤖 Anthropic - Claude Duet Messages You First
On 21 October Anthropic opened a research preview of Claude Duet, an app where Claude works as a proactive agent. It sends a short brief each morning across your connected apps, messages you during the day only when something needs you, and keeps working on tasks after you close your laptop. It runs on Claude Opus 5.5, and Anthropic says it built it for "professionals and small business owners." Access is by waitlist, starting with US Pro and Max subscribers.
Why it matters
It checks your connected accounts every few hours, not constantly, because constant checking would burn through plan limits faster. Those background checks count against your usage.
Before it sends a message, makes a payment, signs a document or deletes anything, it shows you the action and waits for your yes. Trading securities and moving money between accounts are off-limits even with approval.
It plugs into the 6,000+ connectors and plugins in the Claude Marketplace, including QuickBooks, Stripe, Gusto, HubSpot, Docusign and Shopify.
Anthropic says it can misread a thread or get urgency wrong. This is an early version.
How to use it
US Pro or Max? Join the waitlist on claude.com. Elsewhere, you can copy the pattern now by connecting your inbox, calendar and invoicing tool to Claude and asking for a morning brief.
Write down what it should tell you about: invoices unpaid after 14 days, calendar clashes, customer emails waiting more than a day. Duet lets you set what you hear about and how often.
Start with jobs that read and draft, like invoice reminders and follow-up emails. Leave approval on for anything that sends or pays.
Watch your usage in the first week. Background checks and bigger jobs like financial models use more of your plan than a normal chat.
🏦 US Banks - Blocking AI Agents That Log In With Your Password
On 22 October Axios reported that five major US banks (Morgan Stanley, JPMorgan Chase, Bank of America, Wells Fargo and Citi) are restricting third-party AI agents after spotting automated activity inside customer accounts, as summarised by Yahoo Finance. Reuters said it could not independently verify the specific incidents. The core problem: a bank can see which credentials logged in, but not whether a person or an agent is driving. Bankers called the situation "horrifying," per TechCrunch.
Why it matters
Citi says it does not let third-party agents log in or transact. JPMorgan allows view-only access through intermediaries such as Plaid, with no transfers or bill payments.
Chase told customers not to share their username and password with anyone. If an agent has your bank login, you may be breaking your bank's terms.
The same day, 1Password launched a way for agents to sign in without ever seeing the password: you approve each login, and it works with Claude Duet, ChatGPT Work and other agents.
How to use it
List every AI tool that holds a login for your bank, accounting software or ad accounts. Remove any password you pasted into a prompt or a settings box.
Use read-only connections for anything that only needs to look. Your accounting software's own bank feed is the safest route.
If an agent must log in, use a password manager that approves each sign-in, such as 1Password for Agents, instead of handing over the password.
Never let an agent pay, transfer or sign without your tap.
🧮 OpenAI - GPT-6.1 Terra Costs a Tenth of Astra and Works Your Screen
On 26 October OpenAI released GPT-6.1 Terra, a fast model for computer use, reading images and everyday office work such as filling forms and entering invoices into accounting software. API price is $1 per million input tokens and $5 per million output, under the name gpt-6.1-terra. It is in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users.
Why it matters
OpenAI says Terra matches GPT-6 Astra on its Agents' Last Exam benchmark at one-tenth the price per token. Astra lists at $10 in and $50 out.
At max effort, OpenAI says Terra uses over three times fewer tokens than Claude Opus 5.5 at a quarter of its per-token price, about 9x lower cost per task. That is OpenAI's own comparison.
On Terminal-Bench 4.0 it scores 50.3%, up from 21.5% for GPT-5.6 Terra. OpenAI says Sol and Astra stay better for the hardest coding.
On its error-inducing-prompts test, the error rate fell from 53.6% to 16.7% at low effort, and to 4.3% at max.
How to use it
Pick one boring, high-volume job: invoices into your accounting tool, form entry, support ticket sorting.
Run 20 real examples through Terra and through the model you use now. Count the errors and the cost for each.
Keep your strongest model, such as Claude Opus 5.5 in Claude Code or GPT-6 Astra, for hard coding and judgment calls. Send the volume to the cheap one.
Keep a human check on anything that touches money or a customer.
📈 Anthropic - The IPO Filing Shows $16.3 Billion in Six Months
On 26 October Anthropic made its S-1 public. It plans to list on Nasdaq under the ticker ANT, according to CNBC. Revenue for the six months to 30 June was $16.29 billion, up from $1.16 billion a year earlier. CNBC says the growth was driven largely by Claude Code.
Why it matters
80% of that revenue ($13.1 billion) is consumption-based, meaning customers pay for what they use. Subscriptions were $3.2 billion.
Net revenue retention was around 900% at 30 September: existing customers spent about nine times what they spent a year before. The median public cloud software company sat at 108% in January.
CNBC calculates an implied gross margin of 85% (cost of revenue $2.38 billion), up from 12% a year earlier. The net loss was $80.7 billion, and about $72 billion of it is an accounting charge on convertible notes.
How to use it
Expect AI to be billed like a utility. Set a monthly cap on every API key and agent you run, and check the bill weekly.
Prices per task keep falling. Re-run your most expensive workflow on a newer, cheaper model each quarter.
Keep a second model wired in. In June a US export directive forced Anthropic to switch off Fable 5 and Mythos 5 for about two weeks, per CNBC.
Ask each vendor what happens to your data and pricing if they get acquired or change plans. An IPO adds shareholders to that list.
🧩 Short Leash - Five Habits Worth Copying
Five habits from this week's stories.
🔑 Never paste a password. Give agents approved sign-ins or read-only access.
📖 Read before write. Start every agent on jobs that look and draft, then add actions one at a time.
🧪 Test on 20 real jobs. Compare the cheap model with your best one on your own work.
💳 Cap the bill. A monthly limit on every key and agent, checked weekly.
🔀 Keep a second model. One vendor outage should not stop your business.
Start with the first one tonight.
🧰 Hustle Toolbox
🔐 1Password for Agents - lets an agent sign in to your accounts while you approve each login and it never sees the password. For anyone handing agents real logins.
🌙 Kimi K3.1 Pro - Moonshot's new coding model, 1M-token context, rolled out in Kimi Code on 23 October, with a Highspeed version that Moonshot says is 2 to 3 times faster. For builders shopping for an alternative to the big three.
🎨 Ideogram Vector 1.0 - turns a prompt or an image into an editable SVG. API pricing runs $0.13 to $0.20 per SVG from a prompt and $0.10 to convert an image. For anyone who needs logos and print art they can edit.
Also
Anthropic's filing says it agreed to spend $3.1 billion a month on compute from SpaceX's AI division, per CNBC.
The FTC has demanded that Anthropic executives testify in its safety probe of Anthropic and OpenAI, CNBC confirmed (same live blog).
🎬 From Mike - This Week's Builds
💬 Give Claude a WhatsApp number - full setup and 3 versions
🧠 Stop hitting your Claude limit - Prompt Master and the 3 leaks
📧 The Gmail spam checklist - SPF, DKIM, DMARC
🔊 AI voice clone scams - the paper, the official advice and a Claude safe word prompt
🆓 7 free swaps - the free tools to replace the apps you pay for
🤝 OptiMAX - For Lead Buyers
I also run OptiMAX. We advertise under our own brands and sell each lead or live call to one buyer, never resold. If your business buys leads or calls (US accident and injury, insurance, home equity, debt; UK trades and solar), the whole spec is on one page. Read the buyer spec.
🧭 Pick Your Path - One Click
What would make AI pay off for you in the next 90 days?
🔚 Your Move
Duet's pitch is an agent that comes to you first. What is the one thing you would want it to flag every morning?
Hit reply with a one-line answer. I read every one.
Mike
Automate & Hustle. The AI week in five minutes, and what to do with it.