No connection? No problem. This is a full session — the exact prompts and the responses you'd get back. Read each prompt, read the response, do the ✎ paper step, move on. When wifi returns, paste any prompt and keep going live. Same flow as the slides: 6 stages, 6 artifacts.
STAGE 01Ideation — generate, then attack
YOU → AI
You are a serial founder who has killed more ideas than you've shipped.
I want to build for engineering teams. My unfair advantage: I've shipped and
on-called for years. Hair-on-fire problem: bugs that escape to production that
the test suite never caught — found by angry users, not by CI. Generate 10
angles, score 1–10 on wedge, buyer urgency, defensibility, distribution.
Kill 8. Defend 2. Write each survivor's obituary.
AI
01 Yet-another test-runner 3/10 — crowded, Jest owns it
02 AI unit-test generator 4/10 — table stakes now
03 Escaped-bug finder + auto-fix prompt 9/10
04 Flaky-test detector 5/10 — feature, not co.
05 Coverage dashboard 2/10 — nobody pays
06 AI code reviewer 4/10 — GitHub owns it
07 Synthetic-user QA bot 7/10
08 Incident-postmortem writer 4/10 — nice-to-have
09 Log-anomaly alerter 5/10 — Datadog land
10 Bug-triage inbox 3/10 — Linear/Jira plugin
SURVIVORS
#03 + #07 are the same engine. The wedge is the
escape: watch prod, find the bug CI missed, and hand back the exact AI prompt that fixes it.
OBITUARY
"Escaped-Bug Finder died at month 9: it just listed bugs like every other tool. It survived only where it shipped the
fix prompt, not the finding — closing the loop, not adding to the queue."
YOU → AI (the attack — in the room you'd attack a pod-mate's idea)
Now destroy it. Attack as a skeptical seed investor, then the buyer who'd churn.
AI
INVESTOR · WEDGE 6/10
"Sentry and Datadog already catch prod errors. What's left?" — They catch
crashes. They don't catch the silent escaped bug that didn't throw, and they never write the fix.
The fix-prompt is the moat, not the detection.
BUYER CHURN · WEEK 2
· "Another tool screaming in my Slack — muted by Friday."
· "The 'fix' it suggested was wrong and I lost an hour."
·
"Finding bugs" is noise — a trusted fix is signal.
THE VERSION THAT SURVIVES
Watch real prod traffic, surface only the
escaped bug with a repro, and paste a fix prompt the dev runs in their own agent. Trust comes from precision, not volume.
✎ ON PAPER: circle the survivor. Write one obituary line for your idea in the workbook.
STAGE 02Product — animate the wow in HTML
YOU → AI
Build a SINGLE HTML page that ANIMATES my wow moment: a live prod feed scrolls,
the agent flags one ESCAPED bug CI missed, expands a repro, then reveals the
ready-to-paste FIX PROMPT. One file, real-ish data, no auth.
BONUS — a fun logo: flat vector "bug caught in a net" mascot, bold simple
shapes, 3 options, transparent-background PNG.
AI
INDEX.HTML — ONE FILE, OPENS IN ANY BROWSER
<div id="feed"> <!-- prod requests stream past -->
<div class="flag pulse">ESCAPED → /checkout 200 but $0 charged</div>
<div id="repro"> <!-- 3-line repro expands -->
<pre class="fix">FIX PROMPT → "In charge.ts, guard..."</pre>
<button>Copy fix prompt</button>
What you see: traffic streams → one row flashes red "ESCAPED — tests were green" → repro unfolds → the fix prompt types itself out → Copy. Five seconds, and the on-call engineer exhales.
LOGO — 3 OPTIONS (transparent PNG)
1) A ladybug inside a catch-net diamond · 2) A bracket
{ } shaped like a net · 3) A bug with a green check forming its shell.
All flat vector, 2 colors, readable at 32px.
ITERATE
Out of ideas? Say "make it better" — I'd add a "tests were GREEN" stamp next to the escaped bug and a one-click "open in your agent" chip.
✎ ON PAPER: sketch your page in 4 boxes: feed → flag → repro → fix-prompt button. Sketch a 10-second logo doodle.
STAGE 03Positioning — 20 lines, keep one
YOU → AI
Generate 20 positioning lines for Escaped-Bug Finder (finds the prod bug your
tests missed and writes the fix prompt). ICP: on-call engineers at AI-first
startups. Alternatives: Sentry, more unit tests, hope. Score specificity,
urgency, memorability. Kill 18. Show top 2. Pick the winner.
AI
"AI-powered observability" 1/10
"Catch bugs faster" 2/10
"Quality, reimagined" 1/10
"Shift-left with AI" 3/10
…14 more, killed.
TOP 2
"Your tests were green. Your users weren't. We find what escaped — and write the fix." — 9/10
"The bug CI missed, in your inbox with a fix, before the user tweets it." — 8/10
WINNER
The first — it names the exact gut-punch ("tests were green, users weren't") instead of the category.
✎ ON PAPER: write your own winner. Read it aloud to your pod — would a buyer repeat it?
STAGE 04Pitching — simple Khosla seed deck
YOU → AI
You are a Khosla Ventures seed partner. Zero patience for buzzwords.
Positioning: "Your tests were green. Your users weren't. We find what escaped
— and write the fix."
Traction: 14 eng teams, 60% of flagged bugs fixed with our prompt as-is.
Build the simple seed deck: what we do · problem · solution + magic · why now ·
market (bottoms-up) · why we win · business model · go-to-market · traction ·
team + ask. For each: a headline and the ONE RISK it must de-risk.
Then a 60-second narrative in MY voice. Then a skeptic pass — 3 weakest slides.
BONUS: which investors fit? Re-tailor to the top firm, draft the intro email.
AI
1 WHAT WE DO "We catch the bug your tests missed + fix it." risk: sounds like Sentry
2 PROBLEM "Green CI, angry users." risk: is it painful enough?
3 SOLUTION Escape detection + paste-ready fix prompt risk: fix accuracy
4 WHY NOW AI-written code escapes more, faster risk: fad?
5 MARKET 30M devs × prod incidents, bottoms-up risk: TAM theater
6 WHY WE WIN The fix-prompt loop, not just detection risk: copyable?
7 MODEL $40/dev/mo, expands per team risk: seat churn
8 GTM Land one on-call eng, spread in the channel risk: slow burn
9 TRACTION 14 teams, 60% fixes accepted as-is risk: tiny
10 TEAM + ASK $1.2M → close the loop to auto-PR risk: why you?
60-SECOND NARRATIVE (your voice)
"Every team I've on-called for ships a bug that CI swore was fine — and a user finds it first. We watch prod, catch the one that escaped, and hand back the exact prompt that fixes it. Fourteen teams run it; most of our fixes land untouched. We want to close the loop all the way to the PR."
SKEPTIC PASS — WEAKEST 3
Why-you (slide 10), market math (5), copyability (6). Fix why-you first — say the years of on-call scar tissue out loud.
BONUS — INVESTOR FIT
Dev-tools seed funds and angels who backed observability / testing (the Sentry, LaunchDarkly lineage). Re-tailored hook: "We're the fix layer their portfolio's monitoring never had." Intro email drafted — 60 words, opens on their portfolio company, one 15-min ask.
✎ ON PAPER: write YOUR hook in your own voice. Say it to your pod in under 10 seconds.
STAGE 05Go-to-Market — first buyers, reach, price
YOU → AI
Who are my 10 most likely FIRST buyers, with the trigger that makes them need
this now? The ONE channel to win first with $0? Pricing — 3 options and the
number they'd actually pay? Then the reach-out that opens on THEIR pain.
AI
FIRST BUYERS
On-call engineers and eng leads at 10–50-person AI-first startups shipping daily. Trigger: a painful escaped-bug incident in the last 30 days (the postmortem is still open). (Also: solo founders who are their own on-call.)
THE ONE CHANNEL
Where burned on-call engineers vent — the right Slack/Discord communities and a brutally honest "here's a bug your tests missed" teardown thread. Win one team, let them tell the next.
PRICING
$0 free for one repo ·
$40/dev/mo team plan (the product) · usage-based for high-traffic. Anchor against one prod incident's cost, not other tools.
REACH-OUT (cold DM)
"Saw your status page hiccup Tuesday. We replayed your public traffic and found one escaped path your tests likely miss — want the repro + the fix prompt, free? No pitch."
LAUNCH + ENGINE
Launch: a public "bugs that escaped CI this week" teardown. Engine: every fix accepted becomes a shareable before/after — the loop markets itself.
✎ ON PAPER: name your first buyer, your one channel, and a price. Three blanks, fill them now.
STAGE 06Validation — synthetic users first
YOU → AI
Simulate 20 on-call engineers hitting the Escaped-Bug Finder page. For each:
first 30 sec, where they get confused, drop or wow, what they'd tell a teammate,
would they pay $40/dev/mo? Aggregate: top 3 bugs + the single biggest fix.
Then re-run.
AI
20 PERSONAS RUN
12/20 hit the wow ("it caught the one my tests swore was fine — and the fix was right").
BUG · 9/20 hesitated: "is this just Sentry with extra steps?" — the escape isn't visually distinct from a normal error.
11/20 would pay $40/dev/mo. 5 said no — "need it to open a PR, not a prompt." 3 wanted self-host.
TOP 3 FIXES
1) A loud "TESTS WERE GREEN" stamp on escaped bugs to separate from Sentry. 2) One-click "open fix in your agent." 3) Trust score on each fix prompt.
RE-RUN AFTER FIX #1
16/20 wow (was 12) · pay-intent 14/20 (was 11). Ship it — then show one real on-call team. ✎ ON PAPER: predict YOUR #1 objection, and the one fix you'd ship. Tell your pod.