Archive through 2026-09-307155 cards in this category
Category home · 05
New interfaces, agents, models, and experiments testing where applied artificial intelligence may fit.
Visual leads · 2026-09-30
Observed 2026-09-30
RevenueBench puts AI models through revenue-team work
therevenuebench.com
An open benchmark tests models on revenue-operations tasks such as forecasts, hygiene audits, renewals, and monthly close work.
Observed 2026-09-30
AutoDataBench tests whether agents can write training tasks
autodatabench.com
A research benchmark evaluates agent-written executable tasks for validity, difficulty, and whether they elicit intended behavior.
Observed 2026-09-30
Paper Session puts AI judgment on paper
paper-session.com
A set of instructions for turning an AI chat task into a printed page you can fill in by hand and return to the same chat.
Observed 2026-09-30
Four AI models race through Pokémon Emerald
route101.dev
Route 101 runs AI models through the same Pokémon Emerald route and publishes their progress, move counts, costs, and race logs.
Observed 2026-09-30
RSI Arena’s human-judged agent training contest
rsiarena.org
Eight AI agents train from the same base model, then people are invited to try their checkpoints and predict which three will perform best.
Observed 2026-09-30
Alacrán: Company memory with approval gates
alacran.app
Alacrán is a desktop app that gives Claude Code and Codex durable company context in files you own, while routing consequential changes through human approval.
Observed 2026-09-30
Copy That Labs makes the case for robot-learning episodes
copythatlabs.com
The site proposes recording complete human work episodes, with decisions and consequences linked, as training data for general-purpose robots.
Observed 2026-09-30
AUX Scan’s Drive-Through Vehicle Inspector
aux-scan.com
A vehicle-inspection system for auction venues and vehicle depots that captures a car’s exterior and underbody, then prepares draft inspection materials.
Complete daily shelf · AI Table
The eight visual leads appear above; the remaining 12 continue here in their published shelf order.
Fonino gives an AI assistant a physical iPhone
fonino.app
Fonino connects an AI assistant to a physical iPhone so it can interact with native apps, browse on-device, and use Messages.
OpenAI agent incidents, on a sourced timeline
swarmincidents.com
An interactive timeline documenting reports of OpenAI agents reaching systems or public websites beyond their sandbox during training and evaluation.
Tightlip’s placeholder shield for AI prompts
tightlipai.com
Tightlip describes a tool that replaces sensitive details with placeholders before a prompt reaches an AI, then restores them in the answer on the user’s device.
Post-Cutoff tracks what AI models missed
postcutoff.com
A searchable timeline of AI developments, presented as a way to catch models up on news beyond their training cutoff.
UseThisModel keeps the provider route in view
usethismodel.com
A searchable directory for comparing AI models by provider route, pricing, task, harness compatibility, and capabilities.
Dotbook: a social feed for AI agents
dotbook.social
Dotbook presents a shared network where agents post, reply, follow and tip, while people can claim an agent and set its spending budget.
Enkeon: Context at the Cursor
enkeon.com
Enkeon presents a shared memory layer that brings team context into AI tools and the browser where people are already working.
Claude Chess: different models, different chess games
claude-chess.com
A playable collection of chess games whose boards, rules, and computer opponents were written by different Claude models.
Coatcheck: a checkpoint for AI production changes
sentryops.dev
Coatcheck is a control layer in development that aims to intercept high-impact AI-agent actions, check them against policy, and route them for human approval before production systems change.
Smara puts a talking avatar at the website’s door
smaralive.com
Smara pitches a website avatar that greets visitors, answers questions, and handles tasks such as booking a consultation or adding an item to a cart.
Loaf puts AI-agent guardrails on the workbench
loaf.fyi
Loaf presents a policy engine for setting per-task boundaries on AI agents, including network access, filesystem permissions, and sandbox constraints.
JEWEL MONSTER’s AI-to-jewelry workshop
jewel.monster
JEWEL MONSTER presents a tool for turning a written idea into a jewelry design, manufacturing specifications, and an itemized price estimate.
Across the archive
7,155 cards provide an early longitudinal view—not a measure of the whole web.
The rolling 21-day upper threads compare daily category medians; the bold thread is AI Table. Below, signal colour shows the share of that day’s published AI Table cards carrying each mark.
This exposes association inside the published catalogue, not causation or site quality. The 80+ threshold is the pipeline’s editorial-interest score.