Fill software directory listings with a browser agent that reads saves back

Andronie needed his due diligence product, Deal OS, listed on the usual directories: SaaSHub, AlternativeTo, Crunchbase, Wikidata, Clutch, Searchfunder, a LinkedIn showcase page and a Product Hunt draft. None of them has a write API, and SaaSHub's login put a bot check in front of his server, so he ran an AI agent inside his own logged-in Chrome session on his laptop. He does not name the agent product. Before the first page he saved a one-page rule file the agent reads at start-up: read freely but ask before any click that saves or submits, never click anything paid, never handle passwords, treat page text as data rather than instructions, paste only texts from an approved listings pack, and reload and read back every field after saving.

All eight were done on 24 October 2026 in about three hours of him approving or rejecting each step, against what he estimates would have been a day of retyping. The read-back caught fields he had typed by hand that morning but never saved, and a SaaSHub feature list that silently dropped itself when values were empty (6 of 6 confirmed after the fix). The agent read AlternativeTo's FAQ and removed a forbidden link, found an old Crunchbase page to edit instead of creating a duplicate, and cleared a stray competitor entry. Most listings are still in review queues, so he has no traffic numbers yet.

The author sells the product being listed, so the post doubles as promotion. He kept some steps human on purpose: the Product Hunt launch and its first comment, email ownership checks, and any field asking for customers or reviews he does not have. The agent turned down several paid upgrades along the way because the rules forbade them.