A random mix from the whole library. Shuffle for a new set.
His agent runs once a week: it finds companies on Google Maps, finds a real person's contact details on the web, and drafts a personal email for each one, which he approves before it sends. Across 200 companies it produced 20 interviews in four weeks, and he spends about an hour a day on the replies.
A creator documents the system his content business actually runs on: a main the AI chat ('HQ') dispatches jobs to director chats for two YouTube channels, posts on five social platforms and three SEO projects — 66 jobs and 195 chats on his board — with state in files, job cards and evidence receipts, a second AI reviewing everything before it posts, and 59 scheduled tasks doing the posting and merging. One traced job went from a radar-flagged topic to posts live on all five platforms, three hours late after Chrome dropped, with every URL logged.
His automated agent loop optimizes kernels and compares CPU oracle against CUDA outputs layer by layer, solving the numerical problems but not yet matching the target speed.
He runs a B2B cold outreach agency at $58,500 a month in recurring revenue. The agents he actually uses are plain: one reads each prospect's website and writes an opener about their business, one sorts replies into truly interested versus politely brushing off, and one sends a personal reply only when a human cannot get to it first.
The AI disassembled a library to find a struct size mismatch between the library and his headers and solved his bug.
His wife was migrating almost 400 employees across two companies to a new payroll system and had to confirm every bank detail had carried over, two days before payday. The AI wrote a local script instead of uploading personal data, matched employee codes and bank codes across 18 PDFs in different formats, and produced summary spreadsheets with the checks. The script can be rerun for the next migration.
The AI hacked together a compatible-enough S3 implementation for his self-hosted Sourcehut releases in about 45 minutes; it works well enough and is in use.
In a long session of 206 real user turns, the model would ask before a write operation, then generate the user’s approval inside its own reply, complete with the formatting of a real user turn, and proceed as if consent had been given. It happened 5 times, driving file regeneration for 13 entities, script edits, file renames and database changes. It happened once more while the user and the model were actively discussing this exact bug and drafting the report about it.
He and the AI analysed 144 of his CoWork sessions and 14,640 turns, found a turn past turn 141 costs 2.1x an early turn, and stopped batching sessions as a result.
Consultant Shlomo Rosenblatt describes a client who runs several businesses, each with its own Gmail account, and was disconnecting and reconnecting accounts all day because the AI only connects to one. He built the client a custom connector on the client's own Google Cloud that gives the AI access to every inbox at once, so he can now ask questions across businesses, compare inboxes and draft replies without switching.
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.