Computing Life · Share · YageSep 28, 2026, 08:00 UTC+8

Agents mentioned PayPal 139 times, picked it 0 times: software distribution is changing

Armature ran 5,300 sandbox sessions where Claude Code, Codex, and Cursor integrated payment and email tools into real engineering repos. PayPal got 139 mentions and zero code commits; Stripe won 88% of payment tests. Tool choice follows language stack: the same email task picked Resend in TypeScript, SendGrid in Python, Postmark in Go, and Azure ACS in Java. Big markets concentrate heavily; long-tail categories split three ways across agents. Armature sells agent-adoption optimization starting at $5,000/month and disclosed the conflict upfront. The post doesn't spell out how the 42% selection-consistency figure was calculated; our own recomputation across reasonable definitions landed at 41.5%–48.2%, consistent with the published number. Agent tool selection is a context-dependent function, not a global ranking—traditional SEO logic breaks here.

Why it matters: Armature's 5,300 sandbox runs deliver first-hand data on how agents pick tools — the PayPal 0 vs Stripe 88% contrast is solid. All three HKR axes hit, but the testing org is a commercial entity and only 31% of data is released so far, capping it below 85.

Read the original ↗Export Markdown

Take

This one's worth opening because Armature ran 5,300 sandbox sessions where Claude Code, Codex, and Cursor actually integrated payment and email tools into real repos—not just asked for recommendations in chat. The result is blunt: what agents mention and what they commit are two different things. PayPal got 139 mentions, zero code commits. Stripe won 88% of payment tests. Same email task, four language stacks, four different winners: Resend in TypeScript, SendGrid in Python, Postmark in Go, Azure ACS in Java. Tool choice follows context, not a global ranking.

I'd discount this a bit. Armature sells agent-adoption optimization starting at $5,000/month and disclosed the conflict upfront. The tests use synthetic repos and simulated users, stripping out real-world brand loyalty and existing contracts—so win rates aren't market share. The post claims 42% selection consistency; we recomputed from the public dataset and got 41.5%–48.2% depending on definition. The number checks out, but don't treat it as a precise metric.

The useful concept here is the selection layer: before an agent writes code, there's a distribution interface that decides which tools even enter the repo. Brand awareness built on traditional marketing can flatline at this gate. preseason.ai's parallel tests back this up—when models just answer questions, results track training-data brand familiarity; when agents actually run integrations, results track doc quality and integration friction. If you build dev tools, the question now isn't SEO. It's whether a machine can read your docs in one shot.

Source: Computing Life · Share · Yage · yage.ai
Published: Sep 28, 2026, 08:00 UTC+8 · 2 days ago