Avi Hacker, J.D.

I help Real Estate & Legal professionals leverage AI to multiply their revenue and scale - WITHOUT adding overhead | Reclaim 18+ hours a week

[Avi Hacker, J.D. @Avi_hacker1]

@TopherNOW I'm looking forward to this one

[Avi Hacker, J.D. @Avi_hacker1]

@DavidOndrej1 I feel the exact same way.

[Avi Hacker, J.D. @Avi_hacker1]

@girdley A lot of AI products would get better if they deleted half the demos.

[Avi Hacker, J.D. @Avi_hacker1]

@realEstateTrent @Agora_RE The fund jump is where the business stops being one-off deals and starts needing real systems.

[Avi Hacker, J.D. @Avi_hacker1]

@AnthropicAI The domain expertise piece is huge: AI gets dangerous when nobody knows what good looks like.

[Avi Hacker, J.D. @Avi_hacker1]

@commobserver The data layer is where CRE AI stops being impressive and starts being useful.

[Avi Hacker, J.D. @Avi_hacker1]

@BraceSproul @jakebroekhuizen Yep, once agents touch real tools, testing in the live business gets expensive fast.

[Avi Hacker, J.D. @Avi_hacker1]

If you tried AI last month and thought "this is not useful," that conclusion may already be stale.

Codex Record & Replay is the signal:

Show a workflow once.
Let Codex inspect it.
Turn it into a reusable skill.

AI adoption is moving from better prompts to repeatable workflows.

[Avi Hacker, J.D. @Avi_hacker1]

This is actually incredible!

[OpenAI Developers @OpenAIDevs]

Show Codex a workflow once. Reuse it as a skill.

Record & Replay lets you show Codex a recurring task, like filing an expense report or submitting a time-off request.

Codex turns that demo into an inspectable, editable skill.

You control when recording starts and stops.

[Avi Hacker, J.D. @Avi_hacker1]

I've been saying this for so long already. This is the absolute best way to go.

[Aravind Srinivas @AravSrinivas]

Context graphs will be the best way for businesses to enable and deploy agentic harnesses. There's a lot of context fragmentation across so many different tools in every company. A god-mode view that self-improves and self-organizes is the way tacit knowledge can be captured.

[Avi Hacker, J.D. @Avi_hacker1]

Surreal moment this week:

Sitting in my Tesla while it handled the drive.

At the same time, talking to my OpenClaw agents about work.

One autonomous system moving me through the world.

Another moving my business through the day.

Less manual operation. More supervision.

[Avi Hacker, J.D. @Avi_hacker1]

CRE Agent Skills is now 66 standalone prompts you can use without the full orchestrator.

Yesterday's additions:

Office v1: leases, stacking plans, rollover, TI/LC, financing fit.

Capital Markets v1: maturities, refi gaps, workouts, rescue capital, CMBS, recap IC memos.

[Avi Hacker, J.D. @Avi_hacker1]

Yesterday's work on the open-source CRE agent workspace was unglamorous but important:

CI, Dependabot, tighter PR/issue templates, clearer security boundaries, and a release proof matrix.

For agentic CRE tools, demos aren't enough. The proof path has to be inspectable.

[Avi Hacker, J.D. @Avi_hacker1]

A mistake I keep seeing:

Building the AI system for the client, but not teaching the method.

That creates dependency.

Better:

- build v1
- explain the design
- turn it into a repeatable pattern
- teach the client to extend it

The goal is not magic.

It's capability transfer.

[Avi Hacker, J.D. @Avi_hacker1]

Yesterday I wanted to understand how local AI models actually work.

A year ago that meant a weekend of half-finished blog posts, three abandoned YouTube tabs, and a Reddit thread that contradicted itself twice.

So I tried something different.

I went to ChatGPT and asked Deep Research for a full report on local models.

How they work, where they run, how I'd actually use one.

It went out, read the internet, and came back with a real report. Sourced. Structured. The kind of thing that used to take an analyst a day.

Then I took that report and dropped the whole thing into NotebookLM.

NotebookLM turned it into a podcast.

Two AI voices, talking through my research like a couple of experts who'd actually read it.

Back and forth. One would raise a point, the other would push on it, then they'd land somewhere I hadn't thought of.

I put my headphones in and listened to two AIs explain the exact thing I wanted to learn. In a conversation. Made for me.

It was one of the coolest things I've done in a while.

Here's what I keep coming back to.

We have never had better tools, or a better moment, to learn anything we want. Curiosity used to be capped by access, by time, by who'd bother to explain it to you.

That cap is gone.

You can take any subject, point three tools at it, and have a custom report and a private podcast about it before lunch.

The bottleneck isn't information anymore. It's whether you decide to be curious.

I do this for a living, so I live in these tools daily.

But none of this needs a consultant.

The three steps are free or close to it, and you can run them today.

What's the one thing you've been meaning to understand but never had time to dig into?

Try the chain on that. I'd genuinely like to hear what you pick.

[Avi Hacker, J.D. @Avi_hacker1]

@rowancheung AI education is becoming a distribution business, not just a content business.

[Avi Hacker, J.D. @Avi_hacker1]

@OfficialLoganK @GoogleAIStudio A million apps a week changes the question from who can build to who has taste.

[Avi Hacker, J.D. @Avi_hacker1]

@DarioAmodei I think policy will keep trailing capability until governments build faster feedback loops.

[Avi Hacker, J.D. @Avi_hacker1]

@claudeai The useful signal is not the launch, it is what people build in week one.

[Avi Hacker, J.D. @Avi_hacker1]

@OpenAIDevs The best docs are starting to feel less like pages and more like support staff.