You'd think this needs a senior engineer's gut feel — Tycoon Agent reads the repo and tells you exactly which 5 fixes pay back the fastest.
The short answer
Tycoon Agent answers "how much tech debt do we actually have" by scanning your GitHub repositories for TODO/FIXME/HACK comments, Sentry for recurring errors, deploy logs for failed-then-rolled-back releases, and Linear for tickets tagged debt or refactor — then estimates engineering hours owed. She quantifies it: "~127 hours of debt across 4 repos, costing ~3.2 hrs/week in slowdown." Then she ranks the top 5 items by ROI — picking ones where the fix is small but the ongoing pain is large (a 2-hour fix saving 30 min/week beats a week-long refactor saving 1 hr/month). The output is a Lark message: total debt estimate, weekly drag in hours, top 5 fixes with effort + payback, and the items safe to ignore. You stop debating debt in the abstract and start fixing the ones that pay.
How Tycoon Agent actually does it
- Scan repos for debt markers
GitHub search for TODO/FIXME/HACK/XXX comments across all repos, file age, last-touched dates.
GitHub
- Pull recurring errors
Sentry top 20 errors by frequency, filtered to the same root cause occurring >5x/week — those are debt symptoms.
Sentry
- Cross-reference deploy stability
Vercel and Cloud Run deploy history for last 90 days — failed deploys, rollbacks, hotfix-after-deploy patterns indicate fragile areas.
Vercel
- Estimate hours owed
Each marker gets a rough hours estimate based on file size, complexity (LOC + cyclomatic), and whether tests exist.
- Rank top 5 by ROI
Effort to fix vs ongoing weekly drag. Lark message with total estimate, weekly cost, top 5 prioritized list, plus safe-to-ignore items.
Lark
What it looks like in chat
YouHow much tech debt do we have?
Tycoon AgentAcross your 4 repos: **~127 hours of debt**, currently costing **~3.2 hrs/week** in slowdown (failed deploys, rollback time, error firefighting). **Top 5 fixes by ROI:** 1. **Inventory write race condition** (api repo) — 3 hr fix, currently causing ~45 min/week of Sentry firefighting + 2 customer complaints/month. Net payback: 3 weeks. 2. **Hardcoded retry logic in payment webhook** — 2 hr fix, eliminates 1 weekly on-call ping. Payback: 4 weeks. 3. **Untyped checkout state** (web repo) — 6 hr to add types, prevents the bug class behind 3 of last quarter's incidents. 4. **Stale dependencies in admin repo** — 4 hr to bump 14 packages, unlocks the next Next.js minor. 5. **`UserService.findOne` 8 callers using the slow path** — 1 hr refactor, ~200ms saved per request. **Safe to defer:** 38 TODOs in the marketing site repo (low-traffic, no errors). 12 "someday" comments in the discord bot. Want me to file these as tickets and assign?
What you get back
Lark digest with total debt in hours, weekly slowdown cost, top 5 fixes ranked by payback period, and a safe-to-ignore list so you stop second-guessing.
Cadence: Monthly + on-demand before each planning cycle.