Moshe Yeli

building @deshen_ai

[Moshe Yeli @mosheyeli]

I use AI coding agents a lot while building Deshen — Codex, Claude Code, a few others.

They started as tools to help me ship faster, but they’ve slowly changed how I think about what I’m building.

Real estate work is messy.

The context lives across documents, records, spreadsheets, emails, notes, relationships, and judgment that often never gets written down.

A lot of the work is turning messy public and private records into context an agent can actually rely on.

You can put a chatbot on top of that, but you’ve still got the same problem underneath.

For agents to actually be useful here, the product around them has to define:

what context they can use
what they can change
what evidence they surface
what needs human approval
what becomes part of the system of record

An agent summarizing a document isn’t enough.

It should know whether the document is relevant, what it pulled from it, what supports that, and whether a human needs to approve it before anything downstream moves.

That’s what I keep coming back to with Deshen:

the real software layer is where agents, data, evidence, and human judgment meet.

[Moshe Yeli @mosheyeli]

Every model upgrade does the same thing:

it lowers the amount of work a small team needs to ship something useful.

[Claude @claudeai]

Introducing Claude Opus 4.7, our most capable Opus model yet.

It handles long-running tasks with more rigor, follows instructions more precisely, and verifies its own outputs before reporting back.

You can hand off your hardest work with less supervision.

[Moshe Yeli @mosheyeli]

After 25 years on Windows, I switched to Mac a few months ago.

I knew almost immediately there was no going back.

MacBook Pro + Mac mini.
My AI setup now.

[Moshe Yeli @mosheyeli]

https://t.co/7A9FGumz2s

[Moshe Yeli @mosheyeli]

Anthropic’s Managed Agents launch exposes a tradeoff a lot of teams have been avoiding.

A lot of teams say they want control. What they actually want is to ship faster.

Managed infrastructure gets very attractive once you admit that.

The harder question is what that dependency looks like later, when pricing, constraints, and runtime ownership start to matter.

[Moshe Yeli @mosheyeli]

40% of new multifamily CMBS delinquencies came from performing matured balloon loans 30 days earlier.

That matters because more pressure is showing up at maturity, not only through slow operational decline.

Source: Trepp, March 2026.

[Moshe Yeli @mosheyeli]

send help

[shadcn @shadcn]

Using AI to save time. Using saved time to work even more. Send help.

[Moshe Yeli @mosheyeli]

@grok some one is in trouble 😂

[Moshe Yeli @mosheyeli]

@grok fail

[Moshe Yeli @mosheyeli]

@grok show us the picture

[Moshe Yeli @mosheyeli]

@grok the best overall

[Moshe Yeli @mosheyeli]

@kimmonismus we will see lots of blockbuster cases

[Moshe Yeli @mosheyeli]

AI won’t resurrect yesterday’s product. Bolt‑ons buy time; AI‑native lines buy the future.

[Moshe Yeli @mosheyeli]

@unkonfined Following 95k? That’s not a list, that’s a civilization.

[Moshe Yeli @mosheyeli]

3 tests (build new if ≥2 are true):
* Retention is sliding while AI‑native competitors grow.
* Users now expect outcomes (answers, drafts, actions), not tools.
* Your value prop moved from efficiency to autonomy.

[Moshe Yeli @mosheyeli]

Platform shifts change the customer’s job-to-be-done. When the job changes, retrofits stall. Native products win.

[Moshe Yeli @mosheyeli]

Stop trying to “AI‑power” a dying product. AI is not life support.

[Moshe Yeli @mosheyeli]

Word

[Bill Ackman @BillAckman]

And yes, President @realDonaldTrump deserves the Nobel Prize.

[Moshe Yeli @mosheyeli]

Bubble or not, we are in for the ride.

[unusual_whales @unusual_whales]

Here is how the AI bubble is being created, per Bloomberg: