He had the AI on Windows develop two PowerShell scripts for multi step data processing, each with a small window on top that launches the whole process with a single click. Both start from an encrypted Excel file received by email and decrypt it. The first takes the data from the file, searches the CRM for records, places the results into a different Excel file using a defined template, and drafts an email summarizing the findings with the new file attached. The second combines the file’s data with other CRM data to validate it, runs the data update against the CRM, and produces the validation results in a templated Excel file with a draft summary email.
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.
This only partly worked. Worth knowing before you try it.
Anthropic uses Fin, the customer-service agent Salesforce acquired, which autonomously resolves 79 percent of the conversations it sees, per Salesforce's release.
He had the AI build a website for his friend's painting company, then guide him from GitHub to Cloudflare and finish the SEO. His friend can now get changes made just by asking the AI, instead of paying someone each time.
His business scans every invoice into a Google Drive folder. The AI reconciles them there: checking they were billed correctly, that credits for returned goods actually arrived, and that there is no slippage. He says it is far faster than reconciling by hand.
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.
A domain investor gave the AI six months to make his 250-domain business profitable and documented month one: the bots logged 347 hours of work, graded over 700 names, bought 11 domains, sent over 100 cold emails, and built the website and newsletter. Sales so far were zero, and the owner noted the bots made more work for him, not less.
This only partly worked. Worth knowing before you try it.
He gave the AI 45 podcast files, each about an hour and a half long. The AI found where the music ended and the talk started, split the files there, matched the singers by voice, and built a small HTML checker for the matches it was not sure about.
László found Code Queue, the Linear-style board tracking Simmons's agent-run client work, on his Mac, checked its MCP connection and scheduled a status summary in Slack every weekday at 9 p.m. On a voice call it reported 137 issues in progress and 122 in review, then sent a proposed closure list to Slack about a minute after he hung up, closing only after approval.
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.