PropTech Insights
Engineering and AI perspectives from the team behind Germany's #1 real estate technology platform.
Your Coding Agent Passes the Tests. Watch What Happens When It Has to Extend Its Own Code.
SlopCodeBench measured 11 coding agents across 93 checkpoints of iterative extension. No agent solved a single problem end-to-end. Structural erosion rose in 80% of trajectories. Pass-rate benchmarks are hiding this from you.
An AI Agent Canceled Every Stripe Subscription in 7 Seconds. The Model Isn't the Problem.
A cron job written by GPT-5.6-Sol wiped a founder's entire MRR while they slept. The viral post blamed the model. The real lesson is about giving autonomous agents production access without safety rails.
RAG Retrieves. Fine-Tuning Forgets. HyperNetworks Inject — and Now We Have the Scaling Laws.
Nace AI's new paper establishes the first scaling laws for HyperNetwork-based knowledge injection into LLMs — a third path beyond RAG and fine-tuning that doesn't touch base weights, doesn't forget, and generalizes better at scale.
Sakana AI's Fugu-Cyber Tops the Cybersecurity Benchmarks. Their Bigger Argument: The Model Isn't the Hard Part.
Fugu-Cyber scores 86.9% on CyberGym and 72.1% on CTI-REALM, beating GPT-5.5-Cyber and Mythos Preview. But Sakana AI's more important claim is that benchmark-topping is only the beginning — and that frontier models deployed in isolation won't solve enterprise security.
Visa Open-Sourced an 11-Stage AI Security Pipeline. The Key Insight Isn't Discovery — It's Triage Speed.
VVAH — Visa's Vulnerability Agentic Harness, built on Project Glasswing — automates the full journey from code repository to validated fix across 11 stages. The insight worth stealing: in AI-assisted security, the bottleneck isn't finding vulnerabilities. It's triaging fast enough to act on them.
The Hidden Bottleneck in Agent Evolution: What Harness Handbook Reveals About How AI Systems Break
When an AI agent breaks in production, the problem is almost never where you think. A new paper from Tencent and Indiana University identifies the real bottleneck — behavior localization — and proposes a behavior-centric map of the codebase that lets developers and coding agents find where to make changes before they reason about how.
To Every Agent Its Own Database: Why the Shared Data Warehouse Is the Wrong Model for AI Agents
The data warehouse was designed for humans. Joe Reis just prototyped what replaces it: a local DuckDB instance per agent, cryptographically-verified data slices, and a three-layer contract system that makes the shared warehouse obsolete.
One Model, Seven Worlds: What Qwen-AgentWorld Changes About Agentic AI
Alibaba's Qwen team released the first language world model that simulates all seven agentic environments — browser, terminal, OS, mobile, search, IDE, and MCP — in a single unified model. Here is why that matters more than adding another benchmark score.
LLMs for Ambiguity, Deterministic Agents for Policy: Multi-Agent Contract Compliance
A Google Cloud reference sample shows the right way to build agentic contract compliance: an LLM handles messy extraction, a deterministic agent enforces hard policy, and they hand off over the open A2A protocol — with an auditable verdict. Here is why that split matters for real estate.
LLM Inference Optimization: The Line Item That Decides If Your AI Ships
In production, inference — not training — is where the money goes. A practical guide to the techniques that cut LLM serving cost 5-10x: KV-cache and PagedAttention, continuous batching, quantization, and speculative decoding — and why it decides whether AI features are economically viable at scale.
Your Building Already Knows: Turning Property Telemetry into Agentic Action
The most valuable AI asset in real estate isn't a model — it's the operational telemetry your buildings already generate. Here is how to turn high-frequency metrics from commercial and residential properties into agentic action flows that sense, decide, act, and self-correct.
Sakana Fugu: When Orchestrating Models Beats Owning One
Sakana AI's Fugu is a small model that conducts rival frontier LLMs instead of replacing them — and claims to beat Opus 4.8 and GPT-5.5 without training a frontier model of its own. Here is what the orchestration-over-ownership shift means for real estate AI, and the honest caveats.
GLM-5.2 and the Open-Weight Tipping Point for Real Estate Data
Z.ai's GLM-5.2 is an MIT-licensed open-weight model performing near the closed-source frontier on coding and agentic tasks, at a fraction of the cost. Here is what self-hostable frontier-class AI changes for real estate data sovereignty — and the honest caveats.
From Chatbot to Teammate: What Claude Tag Signals for Real Estate Ops
Anthropic's Claude Tag lets teams tag an AI agent into Slack as a governed, async teammate. Here is what the shift from prompting a chatbot to delegating to an agent means for real estate operations — and why governance, not the model, is the hard part.
When AI Agents Rewrite Their Own Rules: Self-Improving Harnesses for Real Estate
A new framework called Self-Harness lets AI agents fix their own operating rules by analyzing their failures — no model retraining, up to 60% better on hard tasks. Here is how harness engineering works and why it is the next reliability lever for agentic AI in real estate and PropTech.
Self-Improving AI Agents: Why Evolving Agentic Systems Win in Real Estate
Self-improving AI agents get better with every run — not by retraining the model, but by evolving their orchestration. Here is how the rollout-and-reflection loop works, and why it is the key to reliable agentic AI in real estate and PropTech.
Building an Agentic Orchestration Layer for PropTech Platforms
How to move beyond isolated LLM features to a governed control plane that coordinates specialized AI agents across real estate workflows — with deterministic orchestration, typed tool contracts, and full observability.
Agentic AI Best Practices: Shipping Reliable Agents in Production
Guardrails, human-in-the-loop approvals, evaluation, and cost control — the engineering discipline that separates a production-grade real estate agent layer from an impressive demo.
How AI Is Transforming Real Estate Asset Management in Europe
Discover how AI-powered platforms are revolutionizing portfolio management, reducing operational costs, and enabling data-driven decisions across European real estate markets.
Building a PropTech SaaS Platform: From POC to $1M ARR in 9 Months
A deep dive into the engineering decisions, architecture choices, and team structure that took a real estate AI platform from proof-of-concept to $1M monthly recurring revenue.
Top 10 PropTech Companies in Europe and USA to Watch in 2025
An in-depth look at the leading PropTech innovators in Europe and the United States — their technology, business models, and the engineering challenges they're solving.
LLM Automation in Property Management: A $6.5M Cost Reduction Case Study
How a major European real estate provider used large language model automation to reduce total operating costs by $6.5 million — and the engineering approach that made it possible.
Microservices Architecture for High-Scale Real Estate Data Platforms
A technical guide to designing microservices architecture for real estate SaaS platforms — covering data isolation, service boundaries, Kubernetes deployment, and Terraform IAC.
ML-Powered Payment Gateway Optimization for PropTech Platforms
How machine learning cascade routing reduced payment gateway fluctuations by 40%, cut human error by 95%, and accelerated time-to-market by 50% for a real estate payment platform.
The Managed Capacity Model: How PropTech Companies Scale Engineering Teams
Why leading PropTech companies are replacing traditional T&M and fixed-price contracts with managed capacity models — and what it means for engineering delivery speed and cost predictability.
AI Predictive Maintenance for Real Estate: Reducing Downtime and Costs
How AI-powered outage prediction and anomaly detection systems are transforming property maintenance — with a case study from a natural resource extraction company managing industrial facilities.
KPI-Driven Engineering Culture: How PropTech Leaders Measure What Matters
A framework for engineering KPIs in PropTech organizations — covering delivery velocity, quality metrics, DevOps performance, and how to align engineering metrics with business outcomes.
From Startup to Award Winner: Engineering Lessons from Germany's #1 PropTech
The engineering decisions, team structures, and delivery practices that took a real estate AI platform from startup to Germany's #1 Asset & Portfolio Management Tool — lessons for every PropTech builder.