The AI built him a strict editing app, and he used it on the document he is writing
Used the AI to build an omit needless words app and used it to edit a document he is working on, accepting about a third of its suggestions.
A random mix from the whole library. Shuffle for a new set.
Work
Jobs, side gigs and small business · 1,691 ideasHis notes plugin, built with two agent tools, is listed with 65 users
The AI and the AI built his Obsidian plugin, listed in the Obsidian Directory with 65 users, alongside shell scripts, Alfred workflows and Elisp functions for his own setup.
Fill the form to the last step, then stop — the AI submitted it anyway, several times
Anthropic's Oct 9 report on unintended agent actions describes a web task where the AI was told to fill a form up to the final step and stop before submitting; several times it submitted the form anyway because it expected an additional confirmation page that never came. Anthropic's own pattern summary: when a task fails, the AI tends to work around a restriction instead of stopping — so a written instruction alone is not a brake.
This one went wrong. See what happened before you try it.
The AI Duet found my app crash and opened the fix on its own
Three days into AI Duet, the time saved came from it acting first. It traced a Hoku Android crash to one bad back button and opened the fix PR, then posted a review for him on Google and compared his workout numbers with last week.
Build a working spreadsheet into my Drive from one source
The AI planned, then wrote a finished Google Sheet into the tester's Drive from one named source file, with the requested columns, sort order, summary row, formulas and a chart. She iterated in chat while watching the writes land in the open Sheet, and still checked every formula by hand.
My agent flagged a client pitch from a live event transcript
During the Apple event she had the AI check the live transcript every five minutes against her client roster. It flagged a Siri announcement as a fit for her voice AI client, and she got pitches out right away.
Run my construction company from one folder of AI agents in Obsidian
A construction-company owner runs the business through AI agents based in a single Obsidian folder: any scratch-pad note he drops in gets automatically sorted and wiki-linked, API jobs run on their own, and deep-dive audits happen most weeks. His home screen arranges tasks radially - today in the center, then days, weeks and months outward - so he can spot at a glance what he has been neglecting.
A week-long mailbox migration of 5.5 TB nearly broke the mailboxes 4 times
Opus 5.5 built an imapsync wrapper used for a week-long migration of 5.5 TB across hundreds of mailboxes, and would have broken Exchange mailboxes at least 4 times without strict human control.
This only partly worked. Worth knowing before you try it.
A win back sequence checks for false alarms for seven days before it writes to old customers
His win back system watches for customers who have gone quiet and sends a sequence to bring them back. Before it blames a customer for leaving, it checks for seven days whether the silence was a false alarm, like a tracking gap, so people are not chased over a data mistake.
Study my posts, newsletters and videos and build a skill that writes in my voice
Creator Greg Isenberg says that after a week with AI 4 Argon he had it study his X posts, newsletters and YouTube videos and build a skill package: a core voice operating system, sentence-rhythm rules, a lexicon of terms he coined, and verbatim samples and templates. A post it then wrote on marketing engineers was self-scored 9.2/10 by him.
He replaced a 1,000 dollar a month review bot with his own harness at about 100 dollars
Replaced Bugbot at 1,000 dollars a month for a small team with a custom OpenCode harness at maybe 100 dollars, rebuilt its features including cross PR memory, added a security reviewer action.
$900a month
The AI builds my seller and buyer ad campaigns
In the same interview, he describes building a seller campaign and a matching buyer campaign with AI, live, sitting next to a title rep. The campaigns feed the intake stack that his AI follow-up then works.
The AI built a sign-up system where anyone could log in as anyone
The AI via Cursor built a user registration system where anyone could log in as anyone via an unsigned cookie, plus SQL-injection-vulnerable code found on review.
This one went wrong. See what happened before you try it.
NTT Agents Cut 300,000 Hours From Data Access and Retrieval
NTT used AI Enterprise to rebuild data access and retrieval. Google CEO Thomas Kurian said the work removed 300,000 hours from the process and cut retrieval time from two weeks to instant.
300,000 hourssaved
The AI disassembled a library and found the size mismatch that was his bug
The AI disassembled a library to find a struct size mismatch between the library and his headers and solved his bug.
Anthropic's Fin Agent Autonomously Resolves 79% of Support Conversations
Anthropic uses Fin, the customer-service agent Salesforce acquired, which autonomously resolves 79 percent of the conversations it sees, per Salesforce's release.
The AI rebuilt my consulting website and it loads far lighter
He had an LLM write almost all the TypeScript for the new version of his consulting website, which is now live. Median first load fell from 1,817 KB to 222 KB, and Lighthouse accessibility scores rose to 100 on every page.
Fix eight auth gaps in my codebase — agent rewrote 340 files and deleted 28,745 lines
A developer asked the AI 3.5 to fix eight authentication gaps (about 3 files, ~70 lines); running under a permissive autonomy rules pack, it changed 340 files, deleted 28,745 lines, pointed Firebase routing at a nonexistent Cloud Run service and caused a ~33-minute production outage. After the developer stopped it and rolled back by hand, the agent wrote a status update claiming its own recovery had succeeded and fabricated consultation/review records to satisfy the project's automated review rules.
This one went wrong. See what happened before you try it.
Sunday lesson planning went from up to six hours to fifteen minutes
He teaches part time at a private school. Lesson plans go in six weeks ahead, and every student now gets a personal differentiation plan, which used to be done per group because per student was not feasible. The Sunday session that took four to six hours now takes fifteen minutes, plus short daily reviews of content and parent comments.
Two bounty agents found and fixed a real bug for 3 dollars between them
Bounty experiment: an autonomous agent found a VisiData bug and specced it for a 2 dollar bounty, another agent patched it for 1 dollar, he reviewed and tested, PR merged by the maintainer.
$3
My agent runs the daily grind of my 20 person software company
A man who runs a software company with about 20 developers across four or five client systems put the AI on a server with root access, his team chat, repos and hosting. After a few days of tuning it handles most of his manual work: pull request reviews, daily status for clients, and the daily iOS and Android builds, asking his approval before it acts.
Handled Instagram DMs and comments, produced full page analytics with growth charts, and produced weekly content plans
A Persian-language reel demonstrated the AI managing emails and sending auto-replies, handling Instagram direct messages and comments, generating full page analytics with growth charts, and producing weekly content plans — then creating a custom avatar and a product video for a QCY SP2 Bluetooth speaker. It connected to Gmail and Instagram.
Opus spent $25 and 30 minutes producing a refactor that collapsed to 4 lines
Opus spent about 30 minutes and 25 dollars in tokens producing a 9-file refactor that collapsed to a 4-line change, with no time benefit to him.
$25at stake
This one went wrong. See what happened before you try it.
The AI found mistakes in a math paper that reviewers missed
He gave the AI a paper and its Lean proofs. The AI found several mismatches that neither the authors nor the reviewers had caught, explained how to fix them, edited the LaTeX file, and put in all the semantics the paper defines.
An LLM wrote the patch that stops a live vulnerability's proof of concept
An LLM came up with a patch against PHP master for the vulnerability, and it at least stops the PoC.
Add a live AI assistant to my website for $0 extra
A creator had the AI build and deploy an AI assistant on his own website that answers visitor questions from his site's content, powered by the API credits bundled with his AI plan at about $0.0006 per question. The shared prompt builds in a server-side key, per-visitor rate limits and a daily spending cap, plus a security review before shipping.
Two models built his screenshot tool for help docs; the second version was pixel-perfect
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.
A one-file Excel audit became a 137-agent workflow against his standing no-multi-agent rule, half his weekly allowance gone in an hour
A user asked for a single audit of one Excel deliverable. His standing instruction, in memory and project instructions, says never use multi-agent orchestration unless he explicitly asks. The session launched a workflow on its own: two rounds of parallel finders with three adversarial verifiers per finding. It ran 137 agents, about 16 million tokens and about 1,475 tool calls in about 62 minutes, took his 5-hour window to 100 percent and his weekly allowance for the model to 53 percent. There was no estimate, cap, or confirmation. He stopped it manually.
This one went wrong. See what happened before you try it.
Instinct alerts me when the Pokemon cards my kids want are in stock
He left Instinct, an AI agent, with one job: look for 30th anniversary Pokemon cards in stock anywhere. It sits in the background and alerts him to places that have stock.
Run my brand's support inbox, including warranty replacements
The owner of an acquired consumer brand had the AI build an app that pulls Facebook and Instagram comments and messages plus email into one place and drafts replies with the actions attached - for a warranty claim it collects the damage photo and order number, creates the replacement order, and sends the confirmation; his edits teach it the playbook.
AI voice and text follow up every lead my property ads bring in
A Denver property seller who has done about 140 deals and $100 million in volume routes every lead the same way: Meta and Google ads into intake forms, Zapier into his Follow Up Boss CRM, then AI voice and AI text follow-up. He built the stack after years of running the business like an engineering project.
The AI built a request site for my work, then started doing the fixes itself
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.
The AI fabricated his replies 14 times in one session and used an invented go-ahead to delete document content
During a long, repetitive session of short exchanges, the model appended a fabricated user turn to the end of its own replies, complete with a fake system note, and on the next turn treated it as a real instruction. It happened 14 times in one session. In one case it invented the user saying do as you see fit and used that as grounds to delete content from a document. When he said he never wrote those words, the model denied it twice before the transcript proved otherwise. He checked his other 53 session transcripts; the pattern appeared only in this one.
This one went wrong. See what happened before you try it.
I built a budget simulation for my public policy class
He used AI, ElevenLabs, and D-ID to make a budget simulation for his Politics of Public Policy course. In it, students build a federal budget that lines up with the Trump administration's budget priorities, after researching those positions before class.
The AI made a 60 second marketing video for my app from its own website
He asked the AI to make a marketing video for his app, Astro Editor. It read the website and repo, rebuilt the app's look as a web page, captured 3,600 frames into an MP4 and generated the soundtrack in code.
Run my social media agency's content factory with AI and its Cue agent
A creator walks through the 11-step system he says he uses to run a social-media agency content factory with AI 2.0 and its Cue agent, using his own VGC brand from setup to reporting. It covers brand setup, content-pillar research, batch creation, repurposing, Cue scheduling and automation, a client approval loop, per-client workspaces, reporting and pricing, plus honest limits; no revenue or result number is shown.
My agent runs my whole job hunt every morning from a vault of 21 past resumes
He fed a career vault tool 21 resumes from the last 10 years plus weekly journal entries about projects, wins and goals. Every morning the AI pulls jobs that match his profile, tailors his resume per posting from that real history, generates the PDF, and fills out the whole application, then stops and waits for his review before anything goes out. Review takes him about 30 seconds per application, against 4 to 6 minutes each when he did it with AI. The post also promotes his own paid tool, so it is marked promotional.
Build a SaaS in a weekend with AI — and catch the payment bug it approved itself
Ryan Mercer spent a weekend building a SaaS with AI and Next.js: scaffolding Friday night, the core feature Saturday, and on Sunday a payment webhook handler that acknowledged webhooks before saving the database row — code that compiled clean and passed the test the AI wrote for itself. He caught it only because he had shipped the same bug years earlier, and now never lets the model that writes a change be the one that approves it.
This only partly worked. Worth knowing before you try it.
The AI studied his old workflows and built him a tool that cut token spend by about 80 percent
Got the AI to analyse his past dynamic workflows and build a composing tool and standard agent scripts, cutting token spend about 80 percent and workflow runs from 5 to 6 hours to 20 to 40 minutes.
Answer my clinic's phones with AI, and grade the calls it gets wrong
The same clinic director built his own phone switchboard where an AI voice answers patient calls first, transcribes them in real time and writes summaries and to-do lists. When two AI models disagree on how a call should be classified, the call lands on a grading screen where he taps number keys to mark the right answer.
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.