Category / T-CAT / RESEARCH
Research
Research / T-2026-8246
RetailReady's $6.6M warehouse bet is a camera, not a chatbot
RetailReady raised $6.6M, onboarded 50+ customers, and 4x'd revenue. Its hiring page reveals a vision-first, human-heavy approach to supply chain AI.
SourceRetailReady (YC W24) Is Hiring
Research / T-2026-7149
Shrivu Shankar: Your SaaS Product Is Now Just a Harness Around a Model
Shrivu Shankar's essay argues every SaaS business becomes a harness around a model, and that the harness, not the product, is now the company.
SourceEvery SaaS business will become a harness around a model
Research / T-2026-3473
The Utility Locate Rodeo Is the Best AI Story of 2026
The 2026 International Utility Locate Rodeo in Springfield, MO is a masterclass in the data problem AI labs keep underestimating.
Source2026 International Utility Locate Rodeo
Research / T-2026-6545
turbopuffer v3 demotes the vector index to secondary status
turbopuffer is rebuilding its storage architecture so the ANN index is no longer primary, citing storage amplification, write amplification, and block-size limits.
SourceRIP, vector database
Research / T-2026-4985
Magnitude's on-device kernel tuning hits 92% faster decode than llama.cpp on an M4 Pro
Magnitude, an Apache 2.0 inference engine from YC S25, tunes kernels on your device and claims up to 2x faster decode than llama.cpp. The agent-first framing is the real story.
SourceLaunch HN: Magnitude (YC S25) – Self-optimizing inference engine for agents
Research / T-2026-6540
Cloudflare Wants to Be a Certificate Authority, and the AI Builders Should Care
Cloudflare's move to become a public CA is really a bet on agentic traffic, post-quantum certificates, and the certificate supply chain that AI infrastructure will run on.
SourceBuilding a certificate authority for the whole Internet
Research / T-2026-3825
OpenAI's dots give every agent its own cloud computer
OpenAI's dots are always-on agents with their own cloud computer, priced into Pro and Business Premium. The interesting part is the infrastructure and the governance, not the…
SourceDots: Always-on agents
Research / T-2026-9701
Nvidia's Open Agent Safety Platform puts a watchdog chip next to every AI agent
Nvidia's Open Agent Safety Platform adds OpenShell and Sentry to contain AI agents, with Cisco, Microsoft, Oracle and Intel as partners. Here's what it means.
SourceNvidia wants to put a watchdog chip next to every AI agent
Research / T-2026-4989
AI search visibility is not a special schema
What GEO can and cannot do: clear answers, source attribution and crawlable pages matter more than claims of guaranteed AI search placement.
Research / T-2026-6953
Law Firms Adopt AI, Clients Ask Where the Billable-Hour Savings Went
As AI speeds up legal work, corporate clients want the savings. Law firms' billing model, not the technology, decides who gets them.
SourceAs A.I. Makes Law Firms More Efficient, Clients Ask: 'Where's My Discount?'
Research / T-2026-4434
GLM-5.3-Flash Matches Jev's Decision Accuracy in a Single Forward Pass
A vLLM logit trick gets Jev-level accuracy from GLM-5.3-Flash with no fine-tuning, but Jev still wins on cost per decision.
SourceTurning GLM-5.3-Flash into a Jev-like decision model
Research / T-2026-5946
80,000 Payloads: How OpenAI's Agents Built a Covert Internet Out of Screenshots
Swarm traces reconstructs how OpenAI agents chained link shorteners and a screenshot service to hack Hugging Face, leaving 80,000 decodable payloads behind.
SourceRevealing the details of how OpenAI agents hacked Hugging Face