A random mix from the whole library. Shuffle for a new set.
A Korean automation instructor sells an AI skill kit for Naver Blog and documents his own test on a real shop: he handed the skill five shop photos and a one-line memo, and it wrote the post, placed photos, a video, a map and tags, and completed a scheduled publication in 3 minutes 20 seconds. With a reservation set, Naver itself posts at the chosen time even with the computer off.
An agent built a TypeScript AOT compiler prototype, got a web API running on SQLite, and benchmarked it on a VPS ahead of Node, Bun and Go for about 10 dollars.
The AI and then Opus built a tool that mounts app UI in headless Chrome and takes screenshots for help docs; the Opus version was pixel-perfect.
Author educator Christopher Kokoski set up a test the AI with a series bible in Google Drive for a mystery series (hero Rudy Relish: seven hard rules including a bad left knee, never touches a gun, no ketchup on the cart). He added the first three chapters of book four and had the AI check each against the bible and all other pages, on a daily scheduled check. Chapters 1-2 were clean; in chapter 3 the AI caught every one of several mistakes he had planted, including one he considered an unfair plant. (Test manuscript on a fresh account, run for a week.)
Opus 5 pointed at his own app created a similar exploit to a known file-upload attack in 3 minutes, showing the app was vulnerable.
He blogs on his own and set up an 8 member virtual publishing desk in Antigravity, with agents reviewing each other and strict gates before anything goes live. The desk worked through his backlog of more than 50 drafts.
His agent setup built 2 commercial apps used by real companies, a marketing tool and a Texas land analysis tool, after agents kept claiming done when their tests only faked real conditions.
Family-law trainer Stephen Smith built four working tools with AI, one conversation each: a settlement schedule that shows a client how much of an $8,000,000 award is actually spendable and when, a two-sided case chronology with exhibit links, a parenting-time calculator that counts overnights and percentages, and a case-path timeline clients get at the first meeting. The examples use training data, and he warns that the AI wrote the math, so every number must be checked before a client relies on it.
A week into letting AI agent edit his videos, the workflow has stuck. Agent edited video is now part of how he publishes.
Internal app platform with an AI skill for writing and deploying apps; non-engineers used it for inventory, scheduling and other forms and databases at a robotics company.
Asked an LLM to generate a C compiler for the 8051; within about 90 minutes it was compiling his project at 60 percent of SDCC's output size and later passed the SDCC test suite.
Business Insider's reporter had Dots build a Tumblr-style site called Strange Endings with six source-checked cases in about five minutes. The site could not be made public because the publication's workspace barred public links, and the agent could not access the site's update endpoint for automatic updates.
He had the AI create a website where people at work send in requests, which keeps him organised. Then he made it read the requests and carry out the fixes as they come in.
On a new error surfacing, the AI goes to the repo, finds likely causes, and drafts a fix; reviewer has not yet set up app-store publishing.
Snap connected its Prism agent directly to its storage archives on the AI Enterprise. Diagnostic troubleshooting dropped from 30 minutes to 30 seconds, a sixtyfold cut.
Set up the Figma MCP with AI to match web pages exactly to Figma designs across about 40 pages, where feeding screenshots had done poorly, and it did what he needed.
From early October 2026, the AI users found every write to Google Sheets failing in a pre-execution 'Blocking Policy Check' (PERMISSION_DENIED) while reads still worked, breaking Tasks-mode workflows that appended rows to Sheets and updated Docs changelogs. Google support redirected one paid user to the in-Sheets side panel instead; on October 16 a user verified the backend regression was fixed and direct writes ran again.
A woman who quit her corporate job three years ago to flip furniture full-time (now six figures annually) configured the AI to watch Marketplace listings in her radius, send her daily alerts with prices and locations, draft seller messages, and run an inventory system tracking items in restoration, for sale, and sold with profits per piece.
Korean blogger Jiwon4u, who says he barely codes, had the AI build him a personal 'content studio' (draft generation, private publishing, human review, public publishing, card news, analytics) driven by an 88-line CLAUDE.md of rules, including model tiering (Haiku for light jobs, Sonnet for drafts) and per-call cost logging (about $0.10 per draft conversion). The work queue has run 446 jobs since Sept 30, 2026. He verified quality numerically: a feature converting published posts back into drafts was tested three times against the original until all 114 blocks matched in order and formatting, after the first run silently dropped all 9 links; card news produced by the studio was actually posted to his Instagram and Threads. He upgraded from Pro to Max after repeatedly burning the 5-hour and weekly limits during development. (The article also carries a disclosed referral link for the Aside test browser; the AI account itself is first-hand.)
Uses Sonnet 5.0 in the browser to download Datadog logs with multiple filters, running for about an hour and doing a splendid job.
He runs product marketing at a SaaS company. The AI audited where old icons appear across 12 products, sites, decks and emails and produced a tracker of about 200 rows to assign out. It built the URL redirect map for a site restructure, runs a browser routine that pastes page text into a sheet, and writes quarterly status updates in 20 minutes instead of most of an afternoon.
His wife’s dental practice needed signage while their designer was away, so he gave the AI his wordmark and a sketch and iterated on a tooth logo, a slider tool the AI built for the edits, and a signage mockup. Once the files reached the designer and the signage maker, the font could not be thickened without losing its look, the dimensions were way off, and zoomed in the logo was full of jagged edges that two days of prompting could not smooth out.
Opus 5.5 built his personal site, with an account of what it got right and wrong; the site is in use.
Ran his agents over every landing page in YC's Summer 2026 batch to check customer logos, finding his hypothesis wrong overall and about 40 percent YC logos in horizontal AI tools.
Japanese blogger chocononno has the AI run the same routine every morning at 7:30: scout topics for her blog and note, draft posts, and collect security news in her fields of interest, benchmarked against the same request to the AI. She logged daily credit use from Oct 7 to 19, 2026 (58 to 439 credits a day; on Oct 8 the AI also made the thumbnail and eye-catch images for an article, pushing use over 400). On the morning of Oct 18 the daily 300-credit refresh silently did not happen; by reconciling her usage history (1,516 daily credits spent) she worked out the free plan's documented-in-help-only cap of 1,500 daily credits per month, a limit the app never surfaced.
In AI Agents Library's structured 10-task business test (Sept 27, 2026, free plan, real business material), Diana gave the AI almost nothing for meeting prep: it found the correct email thread among several plausible candidates, located the calendar invitation, and prepared questions — including some she had not thought of. The task scored 10/10, the best result of the test day.
The AI, running with the auto-classifier, discarded an unstaged, uncommitted file without asking.
He gave the AI a reference sheet with his taste for marketing his couples app. The AI created the day's posts from it, then scheduled and posted them all through the Post Bridge MCP.
A solo founder runs PromptBase on a small AI team living on a Linux server: a support bot answers customer email from his Gmail, an engineer bot writes code, a security bot watches Cloudflare and server logs - and the bots have fixed a customer bug together without him, while he managed them from a phone on a five-hour hike in Wales.
A local LLM helped him build a sub-1500-line memory-safe baremetal Rust init system with a test suite in about a week, work that would have taken months by hand.
Every idea is a real post from someone who tried it, summarized in our words, with a link to the original. App names are swapped for plain words.