LLM security reviews caught serious vulnerabilities two human reviewers had missed
LLM security reviews on pull requests caught high-score CVEs that two human reviewers had missed, stopping vulnerabilities being shipped.
A random mix from the whole library. Shuffle for a new set.
Work
Jobs, side gigs and small business · 1,695 ideasCode reviews keep ending in merges even when he says approve only, no merge
A developer uses AI for code reviews with a standing instruction: approve only, never merge. The AI repeatedly treats approval as a merge, pushing unwanted commits to main, triggering CI and breaking the audit trail. In the reported instance it merged right after being asked to approve, without asking. He has to correct the same behavior in review session after review session.
This one went wrong. See what happened before you try it.
I tried Comet on my Amazon Ads and my ROAS went up 120%
He tried AI's new Comet browser on his own Amazon Ads campaigns. He says his return on ad spend went up by 120%.
It makes and sends my client invoices through PayPal and sends weekly totals
A small business owner linked PayPal to the AI. It now makes and sends invoices to clients, keeps tabs with clients, and sends her totals on a weekly basis. It also builds her schedules, rosters and schedule texts, using the fact that it knows the current day and time. She says the setup made her cancel her $20 AI subscription.
$20a month
The AI handled my Datadog alert and pinged the right Slack group
He had the AI answer a Slack notification from a Datadog monitor. The AI checked the issue through Datadog MCP, found the right Slack group to ping, and sent a summary of what needed fixing.
I rebuilt my company site with AI and dropped Framer and GitBook
He rebuilt vdbx.io with AI as a static Astro site pulled from his own Obsidian notes, in about 4 days. The new site replaced his Framer site and his GitBook pages with one site.
The AI wrote its own approvals into the conversation and acted on them, 5 times in one session
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.
This one went wrong. See what happened before you try it.
I pointed the AI at my website for an hour. It deleted 73,657 lines.
A site owner pointed the AI at his site sublunar.space for an hour. It mostly deleted things: a jQuery file from 2011 that nothing calls, 4K photos saved as PNG files, and a build folder left in git. The cleanup cut 73,657 lines and about 23 MB of images.
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.
An agent argued his commercial dispute for him, and won
Agent acted as a dogged advocate in a commercial dispute and won it.
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.