Seldon: Behavioral QA agents that use your product like customers do
Seldon launched Wren, its behavioral QA agent, on Product Hunt on November 10, 2026. The original Product Hunt listing reported 238 upvotes and 94 comments. Teams describe what should work in plain English, and Wren runs those descriptions in parallel. The agents pick each next step based on what they see on screen, so a button that gives no feedback can prompt an agent to retry it or look elsewhere. The stated aim is to surface test cases teams never wrote, such as a customer who goes back mid-checkout to change an address or retries a failed payment.
When a run exposes a problem, Wren produces a replay and the steps needed to reproduce the bug. Seldon's Wren page gives an example in which a shopper clicked 'Add to bag,' saw no confirmation, clicked again, and ended with two items in the cart. Engineers can review the recording to see which other paths the agent explored. Rather than scripting each browser action in a framework such as Playwright, users describe the task and the expected result.
Pricing starts with a free allowance of 10 runs. Seldon says runs that cannot finish because of an agent or platform failure are not charged and do not count against that allowance. A public GitHub pull request shows a seldon.yml file for the Seldon GitHub App, which runs the tests on each pull request against a Vercel preview and again on each deployment.
Why it matters
For companies selling AI-built products, this launch shows testing moving from fixed scripts toward agents that act like users. An explorer that tries unscripted paths can catch failures in flows like payment retries or edits mid-checkout, the points where customers drop off and support tickets accumulate. For founders, the practical question is whether a plain-English description can produce reproducible evidence without maintaining brittle test code.
For investors, the product is a bet that QA spending can shift from test maintenance toward usage-based runs. Seldon's free tier and its charge only for completed runs make the unit of value easy to see. Buyers will ask about false positives, whether a finding reproduces on rerun, and what a run costs to execute. Those answers will say more about whether behavioral testing becomes a standard budget line than launch-day vote counts do.
Key points
- 1Seldon launched Wren, its QA agent, on Product Hunt on November 10, 2026, with 238 upvotes and 94 comments.
- 2Users describe the expected result in a sentence or two, and agents choose each next step from what they observe.
- 3Each failure comes with a recorded run and steps to reproduce it, according to Seldon's Wren product page.
- 4The first 10 runs are free, and runs that fail for agent or platform reasons are not charged.
- 5
