FAQ
Frequently asked questions
Clear answers about wallet credit, usage, subscriptions, and how Tycoon charges for work.
How do you avoid Sentry noise spam — we get 400 alerts a day and most are garbage?
Noise suppression happens at classification time, not at alerting time. The AI CTO triages every Sentry group but only opens Linear issues for groups that meet a minimum bar: affecting ≥3 distinct users, OR impacting a paying customer, OR on a critical path (auth, payments, core flow), OR net-new this week. A 404 from a dead crawler hitting /old-api a hundred times a day won't trigger anything. The triage log is still kept so you can audit 'why did we not act on this' but the noise doesn't make it to your Linear. Most founders see their triaged issue count drop from ~400/day signals to 2-8/day actually-actionable issues, which is the volume a human could have handled if they had 4 free hours.
What about bugs that only the customer can reproduce — like 'this doesn't work on my iPhone XR with iOS 16'?
The AI CTO can't reproduce those without the customer's help, and it's honest about it. For customer-specific bugs it generates a structured info-request to the customer: 'We need a screen recording, your browser version, and the URL you were on — reply with these and we'll have a fix candidate in 24h.' The AI Customer Support sends the message in a non-annoying way (not a form, a natural reply). About 60% of customers respond within a day with the info needed. The 40% who don't get logged as 'info needed, low priority' and don't block real work. What the AI does that a human ops person wouldn't: automatically correlates every customer-specific bug with PostHog session data if the customer's user ID can be linked, which often reproduces the issue without the customer doing anything.
Can the AI actually fix bugs itself, or just triage them?
It triages and files; fixing is the AI CTO's developer subagent, which is a separate thing. For low-risk categories — typos in copy, missing null checks in clearly bounded code, documentation fixes — the AI CTO will draft a PR and queue it for your approval. For anything touching auth, payments, schema, or production data flow, it stops at the triage step and hands you a ready-to-fix issue with reproducer. The dividing line is explicit in the AI CTO's configuration — founders can expand or tighten it. Most founders start with 'triage-only' and expand auto-fix scope once they've seen the quality of the triage for a few weeks.
How does this interact with Sentry's own AI-grouping features?
Sentry groups errors by stack trace similarity — same function, same file, same exception type. That's the first layer and Tycoon doesn't re-do it. What Tycoon adds is the next three layers: (1) grouping across data sources (Sentry + customer report + Twitter mention about the same thing become one issue), (2) business context (revenue exposure, user segment, workaround availability) that Sentry has no access to, and (3) memory across time (this group of errors is a regression of issue #712 from February, even though Sentry sees it as new). Sentry is the telemetry layer; Tycoon is the decision layer on top. They're complementary, not competing.
What's the escalation if a bug is S0 at 3am and I'm asleep?
You configure escalation per severity. Default S0 behavior: (1) text you via Twilio with the evidence bundle summary, (2) auto-post an acknowledgment on your status page so customers know you know, (3) ping you on Slack mobile with push notification, (4) if no ack from you in 10 minutes, escalate to a fallback contact (co-founder, on-call dev). Some founders configure tighter: auto-revert the last deploy if the error rate on the new release exceeds X%. Tycoon supports that but doesn't default to it because auto-reverts can cause worse outcomes than letting the incident ride with the existing mitigations. You decide the trade-off.