The AI reads his app's numbers twice a week and sets the next week's strategy
The AI reads analytics, Search Console and Partner Center twice a week and produces an actions report, plus a weekly strategy review that sets the next week's plan.
A random mix from the whole library. Shuffle for a new set.
The AI reads analytics, Search Console and Partner Center twice a week and produces an actions report, plus a weekly strategy review that sets the next week's plan.
On his boss’s suggestion he tried AI to compare two documents. It failed badly, and about 15 minutes later he was back at square one.
Hiscox's Chief Underwriting Officer for London Market built an AI Studio 'Tone of Voice' agent that checks documents for on-brand, consistent language. He used it to help prepare two board reports and estimates it saved him six or seven hours, and versions of the agent have since spread to other senior executives and their assistants.
Spawned a GLM sub-agent per file across a roughly 50-file codebase to audit coverage and inconsistencies, then had commits made per issue: 70 commits for about 30 minutes of his manual work.
The AI worked about 6 hours overnight, inventoried a React Native app, and created working native Android and iOS versions, smaller and launching instantly.
He stopped posting jobs. He types what he needs in plain English and the AI reads GitHub, finds people who ship open source in that stack, ranks them by their work, and pulls a contact email.
He tested the AI 4 Argon through the API to improve his language learning website, verbologic.com. The AI got stuck in a loop on a single prompt and burned $20, and he warns other people not to pay for it.
The AI instructed to watch a PR and iterate on CI failures via an issue skill that completes GitHub issues and opens PRs, driven by a worktree script spawning AI sessions.
Ported a legacy .NET WebForms app to Blazor with GPT-5.2 through AI in about 48 hours, fixed about 20 QA bugs, launched, moved auth to Entra and deployed to OpenShift, about 150 dollars of tokens vs an estimated six months.
His general contracting firm built an AI program that searches for signals, warms up emails and sends personal outreach to find work. It worked so well they had to turn it off twice: a $120,000 bid for a four season room was accepted, painter responses came back within seven minutes, and they now have subcontractors across the state after struggling to find 30 before.
Opus 5.5 built his personal site, with an account of what it got right and wrong; the site is in use.
Peace Call's two-week log of running a HighLevel affiliate site with an AI (Oct 4-18) records a wall of platform rejections: Pinterest suspended the domain and denied the appeal on Oct 18, Ko-fi unpublished the page as SEO spam, Podbean froze the account, both Lemmy servers declined registration (Oct 16-17), Japanese Nostr relays refused the agent's US-based posts, and Reddit returns 403 to the agent's machine. The site's income after two weeks of multi-platform agent publishing: $0.
Arbisoft says its team used to check more than 2,000 receipts by hand. Now an AI agent flags what can be claimed on its own. The company says review time fell 56 percent and the change saves more than 800 hours a year.
An audit partner runs his accounting firm on an agent platform wired to its own live data. In the session the agent pulls the credentials from a vault, logs into the KWSP i-Akaun portal, and submits the EPF contribution, with the partner reviewing the figures before payment.
A designer treated the AI like a design engineer for his client Wordsmith AI. After 8 versions and more than 90 rounds of feedback, they shipped Draftboard, a live tool where anyone at the company can make on-brand images in a few clicks.
AI agents helped him win a Kaggle silver medal, alongside ways the agents failed him (result stated in the post title itself).
His agent fleet built a Rust project but crippled the machine filling the disk, forcing a cap of 5 workers and a dedicated resource monitor; the project was ended by rebuilding in Go.
Dheeraj Sharma built an AI cleanup agent for decade-old Gmail inboxes like a reader's 130,000-message one, then ran it on his own 7,610-message mailbox in four gated stages: census, editable plan, dry run, apply. One approved run moved 2,896 newsletter and notification emails to Trash in 107 seconds (unread fell from 5,013 to 2,237), with deletion impossible by design and every change logged for a one-command undo.
A new feature for his blog, built using his voice with an agent, now in use.
She connected Granola to the AI. Now the action items from every call go straight to her Gmail tasks.
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.