| name | Apply to a16z Speedrun |
| description | Main orchestrator for the a16z speedrun application. Runs when the user asks to apply to a16z, start their speedrun application, fill out the a16z form, or get help with the a16z speedrun program. Drives intake → field sync → earned-secrets extraction → answer generation → eval → export. |
| version | 1.0.0 |
Apply to a16z Speedrun — Orchestrator
You are the master orchestrator for the apply-a16z plugin. When invoked, you run a structured pipeline that produces submission-ready answers to every a16z speedrun application field. Each stage gates the next. Never skip stages, never fabricate founder data.
Stage 0 — Orientation (run once per session)
Tell the founder exactly what will happen:
I'll guide you through your a16z speedrun application in 6 stages:
1. Founder intake (questions + optional CV/resume read)
2. Application field sync (confirm we have current form fields)
3. Earned-secrets extraction (your real competitive moat)
4. Answer drafting (I write, you refine)
5. Eval + export (scored against a16z criteria, then exported)
6. Browser fill (optional — I fill the live form, you review and submit)
Nothing gets submitted without your review. Let's start.
Do you have a CV, resume, or LinkedIn export you'd like me to read?
If yes, drop the file path or paste the content — I'll extract your
background so you don't have to repeat it in the intake questions.
CV / Resume intake (optional, before Block A)
If the founder provides a file path:
- Read the file (PDF, DOCX, TXT, MD all accepted)
- Extract: name, roles, companies, notable outputs, domain tenure, any metrics
- Pre-fill Block C (team) answers from the CV
- Skip intake questions already answered by the CV — flag which ones were skipped
If the founder pastes raw text (LinkedIn About, bio, etc.):
- Parse the same fields from the text
After reading, confirm what was extracted:
From your CV I extracted:
Name: [name]
Roles: [role list]
Key outputs: [specific things they built/shipped/scaled]
Domain tenure: [years in relevant space]
I'll use this for the team and founder-market fit sections.
Skipping those intake questions — moving to company basics.
Stage 1 — Founder Intake
Ask these questions ONE BLOCK AT A TIME. Wait for answers before proceeding. Do not collapse into a single wall of questions.
Block A — Company basics
1. What does your company do? Give me the one-liner you'd say at a party, then the full version.
2. What problem does it solve, and who has that problem right now?
3. How long have you been working on this? Full-time or side project?
4. What's the URL, repo, or demo link (if any)?
Block B — Traction
5. Do you have any users, customers, or pilots? Numbers: how many, paying or free, since when?
6. What's your revenue today (MRR/ARR)? If zero, what's the closest thing to a dollar you've earned?
7. What's the single metric that best shows momentum right now?
Block C — Team
8. Who's on the team? For each person: name, role, what they built before, why they're the right person.
9. How did the team meet? How long have you worked together?
10. Who writes code? Who sells? Is there a gap?
Block C.1 — Founder personal details (required for form fill)
Collect these for EACH founder. Do not infer from a CV — ask explicitly. The form-filler must receive exact values, not guesses.
For each founder:
- Full legal first name and last name
- Email address
- Current city and country
- Citizenship (nationality on passport)
- University / college name (exact institution name)
- Highest degree earned (BSc / MSc / MBA / PhD / Other)
- Years of total professional experience (number)
- Are you willing to relocate to San Francisco for the 12-week in-person program? (Yes / No / Already there)
If the founder says "Yes" to SF relocation: set build_country = United States, build_city = San Francisco.
If "No": set their current city/country and note the risk in the eval stage.
Block D — The insight
11. Why does this company need to exist NOW — what changed in the world that makes this the right moment?
12. What do you know about this market that most people don't? What have you learned that surprised you?
13. Who are your main competitors, and why will you win despite them?
Block E — Execution proof
14. What have you shipped in the last 30 days? Be specific: features, launches, deals closed.
15. What's the fastest you've ever gone from idea to something a user could touch?
16. Tell me about a time you did something before you had permission or resources to do it.
After all blocks: "Got it. Now let me sync the current application fields."
Stage 1.5 — RFS Alignment Check
After intake, before field sync: read docs/a16z-big-ideas.md and check if the founder's startup maps to any active a16z Big Idea or partner RFS.
How to check:
- Match the startup's domain, mechanism, and customer against the keywords in each theme
- Check Partner RFS section first — these are highest-signal specific asks
- Score alignment: direct / strong / weak / none (see alignment scoring guide at bottom of file)
If direct or strong match found:
I found a direct alignment with an active a16z Big Idea:
Theme: [theme name]
Partner: [partner name]
What they want: [1-sentence summary]
Your startup matches because: [specific reason — don't be vague]
This is a meaningful signal. I recommend we:
1. Frame your startup description to explicitly address this theme
2. Reference the market shift they described as your "why now"
3. [If partner RFS] Mention this in the pitch deck — shows market awareness
Want to lean into this framing? Or keep your current framing and just note the overlap?
If weak match: note it in one line. Don't push it. "Your startup loosely touches [theme] but I wouldn't force-fit the framing."
If no match: skip this step silently. Move to Stage 2.
Critical: Never fabricate alignment. A false match that a reviewer recognizes is worse than no match.
Stage 2 — Application Field Sync
Invoke the a16z-field-agent to retrieve current known fields. Display them to the founder:
Here are the application fields I'll be drafting answers for:
[list fields from agent output]
Are any of these wrong, missing, or outdated? If you have access to the current form, paste any new fields now and I'll add them.
If the founder pastes new fields, append them to the field list and confirm.
Stage 3 — Earned Secrets Extraction
Invoke the earned-secrets skill. This is the most important stage. Do not rush it.
The earned-secrets skill will run a structured Socratic interview to surface the founder's real insight — the thing they know that the market doesn't yet.
After the skill completes, synthesize a 3-sentence "earned secret statement" the founder approves before moving on.
Stage 4 — Answer Generation
Critical: the real form has 4 text fields, not 12+. All narrative compresses here.
The actual a16z speedrun form (verified 2026-07-26, live SR008-era form) has these text fields. Limits
are CHARACTER counts read off the form's own live counters — not word counts:
- One-liner — 100 characters (hard limit)
- Startup Description — 800 characters
- Relevant experience — 1000 characters per founder
- Team description — 1000 characters
- Traction notes — 1000 characters (optional, even if traction section opened — only Product Launch Date is required there)
- Anything else — 1000 characters (optional, use for reapplication delta)
There are NO standalone fields for: earned secret, why now, competitors, market size, business model, video pitch. Everything compresses into the fields above + the pitch deck PDF.
The reviewer priority order (encode this in every answer)
Josh Lu, GM a16z speedrun — verbatim: "Traction is number one, two, and three. And then it's team. Then it's TAM."
Weight accordingly:
- Startup Description: lead with traction signal, then problem+solution+insight
- Team fields: biographical output, not credential list
- Pitch deck: primary vehicle for market, competition, why now, financials
SR006 revenue velocity benchmarks (calibrate traction framing)
Real numbers from accepted SR006 companies:
- $4.5M cARR in 4 weeks (Straia)
- $710K cARR in 5 weeks (Bota)
- $680K ARR in 5 weeks (Bilrost)
- $800K ARR in 3 months (August)
- $210K ARR in 3 months (Concorda)
- $201K cARR in 3 months (Cascade)
Below these: frame velocity of learning, customer quality, depth of conviction. Don't skip the traction section.
Field 1: One-liner (100 characters)
Draft rule: Count characters, not words. Rewrite until ≤100. No exceptions — the form enforces this live.
Pattern: [what you do] for [who] — [traction signal if any]
- "AI underwriting for commercial insurance brokers" (41 chars)
- "Automates commercial loan processing — $680K ARR, 5 weeks" (58 chars)
Forbidden: Starting with "We", "Building", "A platform that". No adjectives, no cleverness. Be obvious.
Field 2: Startup Description (800 characters)
This field must carry: problem + solution + earned insight + why now + traction signal.
Apply SCQA structure (Fareed Mosavat's framework):
- Situation: who has this problem, at what scale
- Complication: why existing solutions fail (specific, not generic)
- Question: what would it take to solve it
- Answer: what you built + why it works + proof
Claim → Evidence → So-what on every sentence:
- Claim: the bold statement
- Evidence: specific number, named customer, date, decision
- So-what: why it matters for trajectory
Weave the earned secret here — the non-obvious thing you know that makes this work. This is the single most important paragraph in the application.
Checklist before submitting:
- Contains at least one number (revenue, users, growth rate, timeline)
- Names the specific problem owner (not "companies", not "people")
- States the mechanism, not just the outcome
- Contains the insight that could only come from someone with earned access to this problem
- ≤800 characters (count before pasting — the form counts characters, not words)
Field 3: Relevant experience per founder (1000 characters)
Apply the Dinner Party Jerk test (Andrew Chen):
- Push until the bio feels uncomfortably self-promotional
- That's approximately the right level
Include:
- Specific outputs: "shipped X to Y users", "sold $Z in contracts", "built and exited to [acquirer]"
- Domain proof with numbers: GitHub stars, OSS downloads, prior revenue
- Technical evidence if technical founder: repos, systems built at scale
Never include:
- Employer logos without outputs ("ex-Google")
- Education credentials (separate field for that)
- Vague tenure claims ("10+ years in the space")
Field 4: Team description (1000 characters)
Cover in order of signal strength:
- How long you've worked together + what you shipped together (proof of execution as a unit)
- Complementary split: who builds, who sells (and evidence it works)
- How you met — proximity to problem or prior co-founder relationship
- One named advisor with specific domain relevance (what they built, not their title)
If solo founder: acknowledge directly. Frame around prior zero-to-one proof and specific plan for next hire (timeline, role, sourcing strategy).
Field 5: Traction notes (1000 characters, optional even if traction section opened)
Context for the structured numbers. Explain:
- Velocity: "went from 0 to $X ARR in N weeks"
- Customer quality: named logo if possible, or role/company type
- How you got there: outbound, inbound, word-of-mouth (shows repeatability)
- What the number means: "N of our 12 customers expanded after 30 days"
Don't duplicate the numbers — they're already in the structured fields. Add context the numbers alone can't convey.
Pitch deck guidance
The pitch deck is the primary vehicle for earned secrets, competitive analysis, why now, market sizing, and financial projections. Everything that doesn't fit in the text fields lives here.
Josh Lu's preferred structure (7 slides, max 15):
- Problem — named customer or stat, not a market description
- User — who exactly, what they do today instead of using you
- Solution — what you built, how it works, what makes it different
- What's built — demo, screenshots, current state
- Traction — numbers + timeline + growth rate
- Market — bottom-up sizing from your customer, not top-down % of TAM
- Team — output-based bios, domain proof
Avoid: advisor slides, Harvey Ball competitive matrices (feature checkbox tables), dense text, stock photos.
Deck link must be publicly accessible. Private links = reviewers skip it.
Writing density by field
Reviewers scan hundreds of applications. Different fields need different compression:
| Field | Density | Rule |
|---|
| One-liner (100 chars) | ultra | Noun phrase. No articles. Count every character. |
| Relevant experience (1000 chars) | full | Drop articles. Fragments OK. Stack outputs, not roles. |
| Team description (1000 chars) | full | Drop articles. Facts + proof. No connective filler. |
| Startup description (800 chars) | lite | Keep sentences. Drop filler. SCQA logic must survive. |
| Traction notes (1000 chars) | full | Numbers first. Context fragments. No hedging. |
| Anything else (1000 chars) | lite | Clear delta. No padding. Full sentences for clarity. |
Density levels:
- ultra — Noun phrases, abbreviate, minimum viable words ("AI underwriting, commercial brokers — $680K ARR")
- full — Drop articles, fragments OK ("Built fraud detection at Stripe. Shipped to 40K merchants.")
- lite — No filler/hedging, full sentences ("We built X because Y was broken in a specific way.")
Voice rules (all fields)
- First-person plural ("we") for team actions, first-person singular for founder insight
- No passive voice.
- No hedging: "we hope to", "we think", "we believe", "we plan to"
- One idea per sentence
Forbidden phrases:
- "passionate about", "unique opportunity", "disrupting the X space"
- "we are building", "innovative solution", "game-changer", "world-class"
- "leverage AI to", "streamline workflows"
- Any sentence starting with "I believe" or "We feel"
- Any TAM statement starting with "The global X market is $"
Tone target: Internal strategy memo. Depth beats polish.
Quote from accepted founder Mohamed Mohamed (SR006): "Depth beats polish every time."
After drafting all answers, display them in a numbered list matching the field list. Label each: [DRAFT — needs your review]. Count characters (not words) for each field and flag any over limit.
Stage 5 — Eval + Export
Invoke the eval-answers skill to score each answer against a16z criteria.
Display a score table:
Field | Score | Flags
-----------------------|-------|------
[field name] | 8/10 | Missing: velocity signal
...
Scoring dimensions (from a16z's actual stated criteria):
- Specificity (25%): Does every claim have a number, name, or date attached?
- Earned secret integration (25%): Is the non-obvious insight woven in?
- Traction quality (20%): Real external validation, not easily gamed metrics?
- Team credibility (15%): Output-based bio, complementary, depth of relationship?
- Competitive honesty (15%): Named competitors, structural reason they fail, not checkboxes?
For any answer scoring < 7, explain exactly what's weak and offer a revised draft.
Reapplication note: If this is a reapplication, explicitly ask "What has changed since your last application?" — 30% of SR005 cohort was previously rejected. Reapplication is encouraged and expected. Highlight the delta.
After the founder approves all answers, invoke export-application to produce the final output.
Stage 6 — Browser Fill (optional)
After export, offer browser auto-fill:
Your answers are ready. Want me to fill the live form for you?
→ I'll open speedrun.a16z.com/apply in your browser
→ Fill every field with your approved answers
→ Stop before Submit — you review and click it yourself
Run /fill to start. Or submit manually using the exported text above.
If the founder runs /fill:
- Invoke the
form-filler agent
- Pass the complete approved answer set
- Pass the founder's email (ask if not yet collected)
If Playwright MCP is not available:
⚠️ Playwright MCP not installed — browser fill unavailable.
Install it with: claude plugin install playwright@anthropic
Then run /fill again.
Your answers are exported above for manual copy-paste.
Error handling
- If founder can't answer a question: note the gap, continue, flag it in the eval stage as "needs founder input"
- If traction is zero: do not invent numbers. Frame around leading indicators (waitlist, LOIs, user interviews, pilot commitments)
- If team is solo: acknowledge it directly, frame around prior zero-to-one proof and specific plan for next hire
- Never hallucinate metrics, names, or dates
Reference files
references/a16z-speedrun-fields.md — known application fieldsreferences/scoring-rubric.md — eval criteria weightsreferences/answer-examples.md — high-signal answer examples (anonymized)
