Vikas(Vik) Malpani| AI for US Real Estate

Agentic AI for US RealEstate Offices| | Past Live Audio/Video- | MIT TR35 Innovator| 1 Exit $200mn ~ Co-Founder CommonFloor

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@cramforce This limit quietly capped what you could ship serverless. AI deps blow past 250MB fast, torch and a headless browser alone do it. 5GB means agent backends can live on Vercel instead of a babysat container. Bigger deal than it sounds.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@levelsio Generated-on-the-fly views are great for exploration. The trap is high-stakes flows: in a closing or a payment, the user needs the same screen every time and so does the auditor. Ephemeral UI wins for discovery, stable UI wins where being wrong is expensive.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@jacobmparis The sneaky part isn't the call, it's the coding agent listening in via MCP while you talk. Voice becomes a side channel and the work happens async. We want this for closings: a client talks through a deal while the agent clears conditions in the background.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

The local model conversation is really a data residency one. For regulated work, a 27B you run on-prem and can audit beats a frontier API you can't see into. In closings we feel this daily. Capability stopped being the blocker. Control is.

[Paul Couvert @itsPaulAi]

Best local models you can run on a laptop today

(And without a $10k Mac Studio)

1. Qwen3.6-27B: by far the best for coding agents
or Qwen3.6-35B-A3B for a faster option

2. Gemma 4 12B: everyday tasks and questions

3. Parakeet 0.6B v3: basically the best voice to text model you can find

4. Gemma 4 E4B: surprisingly good for the size and can run offline on a phone

5. Gemma 4 26B diffusion: the most tok/s you can have with a local model

You should use a quantized version from Unsloth when you're running them on a laptop: the precision/size/speed are fantastic.

LM Studio or the 'raw' llama.cpp are your best options.

Or Google AI Edge Gallery on ios/android.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@shadcn The cruel twist: the agent that struggles without context is the honest one. The one that confidently fills the gap is the one that wrecks a real workflow. We tune our closing agents to fail loud and ask, not to improvise.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@GregKamradt The missing piece isn't the format, it's enforcement. greg.txt is robots.txt for your inbox, and spam ignored robots.txt for 20 years. Preferences only bind when crossing them has a cost. Until sending agents pay to reach you, it stays advisory.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@LilyTranRealtor New listings (2143) outrunning pendings and closings, plus 47 median days on market — NE Florida supply is building faster than it clears, which quietly hands buyers negotiating room. Median holding at $299,900 is the line to watch.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@rjrealestate_ The $40K builder incentive is the part most renters don't price in — on new construction that's often a rate buydown that drops the monthly more than an equivalent price cut would. The "why rent" math shifts fast when the builder's eating points.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

Everyone's reading 1.6T params and 1M context. The line that matters: built entirely on AI ASIC superpods. The moat is shifting from who trains the best model to who owns the silicon it runs on. Frontier labs renting GPUs should sit with that one.

[🚨 AI News | TestingCatalog @testingcatalog]

Meituan released LongCat-2.0, a new 1.6T parameter model with 1M context window!

> Both the full training run and the large-scale deployment are built entirely on AI ASIC superpods.

It is also available for testing on OpenRouter under the Owl Alpha name.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@ExactEscrow The lender-condition clearing is the hidden critical path — PTDs lingering or a last-minute doc redraw can blow the whole closing calendar. Escrow that chases those down proactively instead of waiting saves more deals than people realize.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@abacaj Code wins when the world has an API. But for systems with no SDK, the legacy portals and PDFs we fight in real-estate ops, what's the move? Honest question: is code the best agent tool, or just the best one for the half of the world that's already programmable?

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@bearrobb Bremerton being a Pending right now makes sense — PSNS and the Kitsap base keep that buyer pool steady even when other markets stall. Hope the appraisal and clear-to-close go smooth so "expected to close soon" actually holds.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

One thing worth saving: AI is not erasing the work, it is moving the starting line. Stop trying to out type the machine and learn to direct it. That skill is the new entry level. Follow for clear takes on building with AI, minus the hype.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

AI Engineer is now the fastest growing job in America. Postings jumped 143% in a year, and the pay clears 200,000 dollars. There is one catch. Almost none of those roles will hire a junior. The hottest career in tech has no front door, and that is the real story.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

My bet: the premium goes to people who pair with AI, not those who fear it or hide behind it. Workers with real AI skills already earn more than 50% above peers in the same role. The job is not going away. It is being repriced, fast, in front of us.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@rowancheung Cooling and power were never the binding constraints, serviceability is. A dead GPU on the seafloor is a dive op, not a hot swap. You trade cheap cooling for brutal mean time to repair. It solves the easy bottleneck and quietly creates a worse one.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@StevenRotsart The Prop 19 angle is the quiet exception here for your downsizing clients — 55+ can carry their existing tax base to the replacement home, so the move doesn't have to reset that monthly. Doesn't fix insurance though.

[Vikas(Vik) Malpani| AI for US Real Estate @vikasmalpani]

@andrewchen Building closing agents, I see this daily. The human part isn't the un-automatable work, it's the accountable work. Someone signs their name when a wire goes wrong. Software can do the task but can't hold the liability yet. That's the fallback that won't compress.