Web apps built to scale Since 2015
2MUCHCOFFEE Get a quote
Tag

AI development

AI development Claude Code vs Codex in 2026: Notes From a Team That Runs Both We gave Claude Code and Codex two open GitHub issues, a Python feature and a TypeScript bug fix, three runs each, graded by tests we kept back. All twelve runs passed. Price per change, the shape of each diff and the review step are where they split, with every number and the method shown. PropTech The Real Estate Advisor's AI Copilot: Capture the Call, Draft the Deal Docs When an agent takes the deal terms on a call, those terms can get re-keyed into paperwork more than once. The copilot captures the call and drafts the documents from vetted templates, with a human approving every draft, because a deal document the AI gets wrong is a liability. AI development PDF Parsing for RAG: pdfplumber, PyMuPDF, Docling, Marker and MinerU on Page 300 of a Long Manual We ran pdfplumber, pypdf, PyMuPDF, pymupdf4llm, Docling, Marker and MinerU over a 400-page slice of the PostgreSQL manual, a two-column paper and a 1947 scan, and measured headers in the chunk, footnotes, a page-spanning table, OCR and cost per 1,000 pages, with the licence beside each. AI development Spec-Driven Development Tools: What Each One Is, From a Team That Uses None of Them Spec Kit, Kiro, OpenSpec, BMAD and Tessl, read from their own docs and changelogs rather than their launch posts. What each actually is, what it costs to adopt, and why the tool layer churns while the practice does not. AI development Spec-Driven Development: What Actually Changes When the Doc Comes First Most explanations of spec-driven development define the term. This is the loop we actually run, why the spec became the reviewable artifact once a model could write the code, and what it costs to keep it running. PropTech The AI Layer That Can't Hallucinate: Grounded Listings, Search, and Lead Agents An AI that invents a bedroom, a square-foot count, or the wrong HOA fee isn't a typo in real estate. It's a misrepresentation, and the licensee carries the liability, not the model vendor. The fix is grounding plus a review step that flags invented facts before anything reaches a buyer. PropTech Building a Property Management Platform in 2026: Rent Rails, Maintenance, and the Trust-Accounting Trap A property management platform looks like a CRUD app and behaves like a bank. You're holding other people's money and dispatching emergencies at 2am, and the part that ends careers is the boring one: mixing owner funds, rent, and security deposits in a way the state writes into statute. PropTech The Real Estate CRM: Speed-to-Lead, Multi-Role Data, and Why Generic CRMs Break A real estate lead has a five-minute fuse, and the same contact is a buyer, a seller, and a past client at once. Generic CRMs break in real estate because they model contacts and deals, not the transaction and the roles, and that gap is where deals and client trust leak out. PropTech The Valuation Engine You Don't Own: Building an AVM People Can Trust On-market, an automated valuation looks impressive. Off-market, where the model can't see the new kitchen and can't peek at a list price, accuracy roughly triples in error. You're standing behind a number a family makes a six-figure decision on, and the moat was never the model. PropTech The Compliance Fortress: Fair Housing, Algorithmic Bias, and FCRA for Real Estate Platforms When your platform's AI decides who sees a listing, who gets approved to rent, or what a home is worth, and it discriminates, the question isn't whether the model meant to. It's who is liable: the developer, the deployer, or the brokerage. The architecture you choose is the answer. AI development How to Build a Real Estate Platform in 2026 (and What's Still Standing in 2028) Most real estate platforms don't fail on the engineering. They fail because the team built a listings app, and the problem was never listings. A builder's map to the parts that actually decide it: the MLS data layer, valuation fidelity, the trust layer, and the moat that compounds by 2028. AI agents The YC Batch That Barely Hires: What Spring 2026 Says About the Cost of Software Four days before Demo Day, 34 of 196 companies in YC's Spring 2026 batch were hiring anyone at all. A look at the batch that builds with agents and almost no people, and what it means if you sell engineering work.