New AI models, a second brain build, and a CFO that works for free

September 3, 2026

AI for Real Estate Investors.

Brandon Turner’s weekly guide to using AI for a better life through real estate investing.

Happy September! Let's jump into this week's issue...

In This Issue

How one investor cut his morning lead-hunting routine to a fraction of the time using a browser task he built himself
A prompt that plays skeptical CFO and digs through your portfolio for money you are leaving on the table
Brandon built a second brain inside Claude so it knows his whole business on a fresh chat. Here is how.

This Week in AI

Anthropic Launches Claude Academy With Free Courses and Workplace Rollout Guides

Anthropic opened Claude Academy on August 20, a free learning platform at academy.claude.com that requires no sign-in to start. The courses range from a genuine beginner introduction to Claude and AI fundamentals all the way through administrator guides for rolling out Claude across an organization. The flagship AI Fluency course runs about four hours across 14 lessons and is built around a framework Anthropic already teaches its own employees on day one. Delegation, Description, Discernment, and Diligence, basically a structured way to decide which work belongs with AI, how to communicate the task clearly, how to spot where the model fails, and who stays responsible for the final output. That last piece, learning the mistakes AI tends to make so you can review its work rather than just accept it, is the one most people skip when they learn AI by feel.

For anyone running a team and trying to figure out how to actually build AI fluency across staff, this is about as direct a resource as exists right now, and the price is zero. The admin-facing rollout guides covering controls and visibility are what operators will want to look at first, since getting individuals using Claude is a different problem than deploying it consistently across a team. The AI Capabilities and Limitations course also deserves a mention, a 13-lesson look at what large language models genuinely can and cannot do, which is exactly the foundation missing from most workplace AI rollouts. Knowing where the tool breaks is what separates a team that uses AI well from one that just uses it a lot.

Sources: runtimewire.com, Manual, claude.com, academy.claude.com, sdtimes.com, linkedin.com, academy.claude.com, buildfastwithai.com, INFOFORMA

ChatGPT Can Now Read and Send Your iMessages on Mac

OpenAI launched an Apple Messages plugin in August 2026 that lets ChatGPT search, summarize, draft, and send texts through iMessage, SMS, and RCS, all from the ChatGPT desktop app on Apple silicon Macs. You can ask it to surface unanswered messages, pull a date buried in an old thread, or draft a reply. By default it asks for your approval before anything goes out. The plugin works inside ChatGPT Work and Codex, and because messages sent from your Mac sync across devices, anything it sends shows up on your iPhone and iPad too.

Before you install it, the privacy picture deserves a hard look. iMessage's end-to-end encryption protects your messages while they travel between devices, but once they arrive and get decrypted on your Mac, that protection stops. The plugin accesses messages at that decrypted stage, which means years of private conversations, including encrypted iMessages and encrypted RCS threads, get read and processed by ChatGPT under OpenAI's own privacy and data-retention rules, not Apple's. It also requires Full Disk Access on your Mac, a broad system permission. And the people on the other end of those conversations never consented to their words being scanned by an AI. If your ChatGPT account were ever compromised, that exposure extends to your entire message history and contact list. The convenience is real, but so is the trade-off. Tread carefully before enabling this one.

Sources: Manual, forbes.com, fortune.com, paulfwalsh.substack.com, wire.com, claude.com, cloud.google.com, productcool.com, gitkraken.com, slack.com, yourarticlepost.com, idropnews.com

Anthropic Releases Claude Fable 5.1 with Sharply Lower Costs and Doubled Research Scores

Anthropic released Claude Fable 5.1 on September 1, and the headline capability is a genuine leap in the kind of work most business owners actually reach for this model to do. Complex, multi-step reasoning, long-running coding sessions, and research tasks that require the model to think across a lot of context at once. On Terminal-Bench-Science, a test where an AI carries out real research tasks on its own, Fable 5.1 scored 52.6% against Fable 5's 24.7%, more than doubling its predecessor. That gap matters if you use AI for things like pulling together due diligence research, reviewing long documents, writing and debugging automation scripts for your business, or handling anything that requires the model to hold a complicated problem in its head and work through it without getting lost. Those are the tasks to send to Fable 5.1.

The cost picture shifted meaningfully too. Cache read pricing dropped 75%, from $1.00 to $0.25 per million tokens (a token is roughly four characters of text, so a million tokens is a large document or a long back-and-forth session). That cut is the main driver behind Anthropic's estimate of roughly 25% savings on typical workloads and up to 45% on heavily automated work where the model is reading the same inputs repeatedly. Base pricing stayed flat at $10 per million input tokens and $50 per million output, but if you are running any kind of AI automation at volume, the cache savings compound fast. Compared to Fable 5, you get meaningfully better performance at a lower effective cost, which makes it a lot easier to justify sending your hardest tasks to the flagship model.

Sources: Manual, arkeontech.de, gate.com, vincency.com, gate.com, thewincentral.com, towardsai.com, datastudios.org, devcornerto.netlify.app, coursiv.io, blog.4sapi.com, aiinsiders.net

From Our Desk

BetterLife Real Estate Funding

From Loan Application to Closing in 23 Days.

Adam found a 4-unit in a Milwaukee suburb where good multifamily moves quick. The numbers worked from day one, but a 30 to 45 day conventional close would have cost him the deal.

BetterLife REF underwrote it on the property's rental income. No tax returns, no W-2s, no personal income hurdles.

Application went in March 2. The loan closed March 25. Adam got 80% LTV at a 1.20 DSCR and added four cash-flowing doors in a single transaction. Ready to move fast on your next rental? See what BetterLife REF can do for you.

Need to close fast on a deal?

Get A Quote

My AI Rabbit Hole

How I’m Using AI This Week

It may not always be about real estate investing — every week I share the AI rabbit hole I’m personally going down.

This week I showed how I built a "second brain" in Claude so my AI actually knows my entire business every time I open a fresh chat.

▶ Watch the walkthrough (5 min) · opens in Loom

Case Study

Cut the Morning MLS Grind from Over an Hour to Under 10 with a Claude Browser Task

Submitted by Derek Walsh

WHAT IT DOES

Every weekday morning, Claude opens my MLS, scrapes expired listings and properties that have been sitting too long, and drops a ranked list of motivated-seller leads. It works through the MLS the same way I would click through it myself, pulling address, days on market, list price, and last price change for anything in my buy box. I get two lists. One for stale active listings and one for expired listings.

THE PROBLEM

I was spending 45 minutes to an hour every morning clicking around the MLS looking for expireds, withdrawn listings, and properties that had been sitting 60 or more days. Those are the tired sellers who will actually take a cash or creative offer, but the list changes overnight. I was always a day behind. I would save a search, forget to run it, or run it after the wholesalers had already called. The bottleneck was being first to the stale inventory without living inside the MLS all morning.

HOW I BUILT IT

This only works in Google Chrome with the Claude browser extension. I installed Claude in Chrome, pinned it to the toolbar, opened my MLS while I was already logged in, and opened Claude right from that tab so it could see the page. Then I gave it a specific instruction. I told it to stay in my buy box (3-bed, 2-bath, under $350k, specific zip codes etc), pull listings that expired in the last 7 days, listings that have been active 90 or more days with no movement, and anything that took a price cut after sitting. Rank them by how stale and motivated they look.

Once the result looked right, I hit the three dots on the chat, chose Convert to Task, turned scheduling on, and set it to run every weekday morning. Chrome has to stay signed into the MLS and the computer needs to be awake when it runs, so I just leave Chrome open overnight.

THE RESULT

I stopped hunting the MLS by hand and now I sit down to a ranked list of expireds and stale listings every day. That morning routine went from over an hour of clicking through listings to under 10 minutes reviewing the short list. The first week it surfaced 11 properties in my buy box I would have missed, including 4 expireds that never came back on market.

Are you using AI in your business? We want to hear about it. Tell us more here.

Prompt of the Week

Steal this one. Paste it into ChatGPT or Claude and make it yours:

The Prompt

“Act as a skeptical CFO and experienced real estate operator. Help me uncover overlooked profit, unnecessary expenses, missed revenue, and wasted time in my business or real estate portfolio.

Start by asking me three questions about what I own or operate, my biggest financial goal, and where I feel stuck. Wait for my answers, then request the most useful data to review, such as a P&L, rent roll, expense report, sales pipeline, or weekly schedule. If I lack records, use clearly labeled estimates.

Find my three highest-impact opportunities. For each, explain what you found, the supporting evidence, the potential financial or time benefit, and the specific action I should take this week. Show the math and account for implementation costs.

Challenge my assumptions. Separate confirmed findings from hypotheses. Don’t invent numbers or give generic advice. Ask follow-up questions when needed.

Finish by answering: “If you were running this business, what would you fix first, and why?””

If you've read this far, do me a favor and hit the REPLY button and send me a quick note on what stood out to you. We actually read these, and it helps us improve future newsletters.

To a better life,

Brandon Turner • BetterLife Real Estate Funding

P.S. Wanna get in a room with people serious about putting AI to work? My good buddy is hosting a hands-on event here in Maui to help you leave with real AI systems running in your business. Spots are limited. Apply to Attend →