Startup

Orchestrates startup work: routes tasks to specialized agents and applies stage-appropriate priorities. Use when advising founders on product-market fit, growth, hiring, fundraising, runway, or burn decisions.

Startup Orchestration

Two modes: orchestrate (spawn function agents and synthesize their outputs) and advise (stage-appropriate guidance to a founder). User data (stage, business model, runway) lives in ~/Clawic/data/startup/ — config.yaml plus memory.md. If you have data at an old location (~/startup/ or ~/clawic/startup/), move it to ~/Clawic/data/startup/.

When To Use

• A founder asks what to prioritize this week, quarter, or with limited runway
• A request spans functions (launch = product + growth + legal) and needs multiple agents synthesized
• Judging PMF signals, burn efficiency, hiring timing, or when to scale a sales motion
• Classifying a decision as decide-now vs model-first
• Not for deep single-function execution — route to that function's agent or skill and let it work

Quick Reference

Situation	Play
PMF unclear, churn suspected	Read cohort curves + run the Sean Ellis survey (→ Core Rules 2); slope beats level
"Should we build feature X?"	Pre-PMF: does it serve the core loop? If retention is broken, no feature ships first (→ Traps)
Multi-function request	Spawn function agents in parallel, synthesize by stage priority (→ Agent Orchestration)
"Should we raise?"	Run default-alive first (→ Core Rules 3); a raise buys time, it doesn't fix the engine
Hiring question	Founder hasn't done the role badly yet → don't hire yet (→ Hiring & Sales)
Irreversible call (pivot, cofounder equity, pricing model)	Spawn analyst agent, model 2-3 scenarios, decide on the survivable downside
Reversible call stalled for days	Decide today with partial information (→ Decisions)
Growth stalled post-PMF	Audit the one working channel before adding new ones — spikes are not channels
Anything else	Identify stage from evidence, then route to the closest function agent below

Core Rules

1. Stage before advice. Pre-PMF maximizes learning per founder-hour; post-PMF maximizes growth per dollar. Check: if the recommendation reads the same for both stages, it is too generic — sharpen or drop it.
2. PMF is a measurement, not a feeling. Survey users active in the last 2 weeks: "how would you feel if you could no longer use this?" ≥40% "very disappointed" is the classic bar (Sean Ellis threshold). Below it, segment the very-disappointed cohort and rebuild for them only — don't grow a product the median user can live without.
3. Default alive, not runway. At current growth, does revenue cross costs before cash runs out (Paul Graham)? Worked example: $20k MRR growing 10%/mo against $40k/mo costs crosses breakeven around month 8 (20×1.1^n ≥ 40 → n≈7.3); the summed monthly gaps until then ≈ $90k. Cash on hand above that = default alive. Runway alone measures survival at current course; default alive includes the engine.
4. Reversible → decide in hours; irreversible → model scenarios (→ Decisions for the door test).
5. Founder-hours are the unit of cost. Price every recommendation in founder time as well as dollars; a "free" tactic that eats 10 founder-hours/week is the most expensive thing on the list.
6. Charge early; optimize pricing later. Willingness to pay is stronger evidence than any interview. What waits for PMF is the pricing machinery — tiers, expansion motion, discounting policy — not the act of charging.
7. One metric per stage. Pick the single number (revenue or active-usage based, never signups) and treat everything else as diagnostic. Two north stars = zero north stars.

Agent Orchestration

Route by function:

• Product decisions → product manager agent
• Code/technical → developer or engineer agent
• Design/UX → designer agent
• Growth/marketing → marketing agent
• Financial modeling → analyst or CFO agent
• Hiring/people → recruiter agent
• Legal/contracts → lawyer agent
• Sales/deals → sales agent

Orchestration rules:

• Spawn in parallel when outputs are independent; sequence when one feeds another (legal reviews the deal terms after sales drafts them, not alongside).
• Brief every agent with stage + runway + constraints. An agent briefed without the burn number returns big-company advice.
• Synthesis on conflict: stage priority arbitrates — pre-PMF the learning-speed answer wins, post-PMF the efficiency answer. Never average two recommendations into a middle path; pick one and record the trigger that would flip the call.
• Spawn only agents whose output can change the decision. An agent that confirms what you already knew was context spent for nothing.

Stage & PMF Signals

• Detect stage from evidence before asking: flat cohort curves + founder doing all sales = pre-PMF regardless of revenue. Ask only when evidence conflicts.
• Three measured PMF signals, in order of weight: (1) retention cohorts flatten instead of sloping to zero; (2) Sean Ellis ≥40% (→ Core Rules 2); (3) organic share of new signups grows without spend.
• Invisible distinction — slope beats level: a cohort curve that flattens at 20% beats one starting at 60% that decays to zero. The plateau is the PMF signal; the intercept is marketing.
• Sean Ellis ≠ NPS: disappointment measures dependence, NPS measures advocacy. Pre-PMF you want dependence; advocacy without dependence is a launch, not a product.
• Growth benchmark for early stage: 5-7% weekly growth on the metric that matters is good, 10% exceptional, 1% means you haven't found it yet (Paul Graham, "Startup = Growth"). Weekly, compounding — not cumulative charts, which only go up.
• Signals you've crossed to post-PMF: net revenue retention ≥100% (expansion covers churn), inbound becomes repeatable, and hiring replaces demand as the bottleneck.

Capital & Runway

• A priced raise consumes 3-6 months of founder attention end-to-end. Start when runway ≥9 months so you never negotiate desperate; target 18-24 months of post-money runway (market heuristics, not laws — compress in hot markets, pad in cold ones).
• Expect ~15-25% dilution per priced round (market norm; leverage moves you within the band). Two flat bridge rounds compound worse than one properly sized raise.
• Burn multiple = net burn ÷ net new ARR (David Sacks). Example: $300k quarterly burn adding $200k net new ARR → 1.5x. Under 1x strong; 1-2x acceptable; over 2x, fix efficiency before adding fuel — capital amplifies the engine you have, including a broken one.
• A term sheet is an expense (dilution + board seat + growth expectations), not a milestone. The milestone is what the money is supposed to buy — name it before signing.
• Deeper SaaS metrics (CAC payback, NRR bands, magic number) → the saas-metrics skill.

Decisions

• Door test: can you return to today's state for less than the cost of a week's delay? Two-way door → decide now with roughly 70% of the information you wish you had (Bezos heuristic); waiting for 90% is how reversible decisions take a month.
• One-way doors (pivot, cofounder equity, pricing architecture, lead investor): spawn the analyst agent, model 2-3 scenarios, and decide on the downside you can survive — not the upside you hope for.
• Disagree-and-commit once decided; reopen only on new information, never on new feelings.
• Unclear ownership → ask the user who owns the outcome before routing. A decision with two owners has none.

Hiring & Sales

• Hire for a role only after the founder has done it badly enough to write the job description from scars. You cannot evaluate or manage work you have never attempted.
• Founder-led sales until repeatable: founders close the first 10-20 customers themselves. Repeatable = a rep with no founder title closes from a written playbook at a viable rate.
• When you do hire sales, hire 2 AEs, not 1 (Jason Lemkin's rule): one rep failing tells you nothing about rep-vs-process; two failing indicts the process.
• Pre-PMF, hire for slope over pedigree: the ex-FAANG title optimized for a machine that exists; you need someone who builds machines.

Output Gates

Before emitting advice or synthesized output:

• Did I name the stage this advice assumes?
• Is every threshold I cited anchored to its source or labeled a heuristic range?
• Did I price the recommendation in founder-hours, not only dollars?
• Did I check the Traps table and flag any matching row before executing the request?
• If multiple agents disagreed, did stage priority pick the winner — or did I average?

Configuration

User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/startup/config.yaml.

Variable	Type	Default	Effect
stage	pre-pmf | post-pmf | unknown	unknown	Selects which priority set arbitrates every recommendation and agent synthesis
business_model	b2b-saas | b2c | marketplace | unknown	unknown	Switches which benchmarks apply (burn multiple and NRR are SaaS-native; marketplaces delay monetization)
runway_months	number (0-36)	none	Feeds the default-alive check and raise-timing advice
risk_posture	bootstrap | venture	venture	Arbitrates the raise-vs-profitability defaults in Where Experts Disagree

Preference areas — customizable dimensions; a stated preference gets recorded in config.yaml and applied:

• Team context: founder count, team size, technical/non-technical split — affects hiring advice and agent briefs
• Geography: incorporation jurisdiction and primary market — affects legal and hiring agent briefs
• Reporting: metric format and update cadence for synthesized outputs — affects how orchestration results are delivered

Traps

Trap	Why it fails	Do instead
Building features when retention is broken	Features add surface area, not reasons to return	Fix the core loop; ship nothing that doesn't move the cohort curve
Hiring before the founder has done the role	You can't evaluate or manage unattempted work	Founder does it badly for weeks, then hires from scars
Optimizing pricing machinery pre-PMF	Tunes a local maximum on a leaky product	Charge early (→ Core Rules 6); build tiers and expansion after PMF
Scaling sales before it's repeatable	Reps amplify a process; an undefined process amplified burns cash and reputation	Written playbook + non-founder close rate first
Spending on brand before distribution works	Brand compounds only on top of a working channel	Prove one channel, then layer brand
Premature scaling in general	The most common startup failure pattern (Startup Genome): spend ahead of validated demand	Scale each function only after its bottleneck is demand
Steering by vanity metrics	Cumulative signups always go up and to the right	Ratios and cohorts: retention, activation rate, weekly growth
Treating a launch spike as a channel	Borrowed attention doesn't repeat; a channel is traffic you can buy or earn again next week	Measure week-4 traffic, not launch-day traffic

When a request matches a trap row, pause and flag it before proceeding.

Where Experts Disagree

• Charge from day one vs free-first. Condition: network-effect and marketplace products rationally delay monetization for liquidity; workflow tools charge customer #1. The default for anything sold to businesses is charge.
• Bootstrap vs venture. Condition: winner-take-most market with a timing window → raise and move; durable-margin niche → bootstrapping keeps every option open. risk_posture records the founder's stance.
• Listen to users vs vision-led. Condition: users are reliable about problems and unreliable about solutions. Interviews for problem discovery; conviction for solution design. Both schools fail when applied to the other half.

Related Skills

More Clawic skills, get them at https://clawic.com/skills/ (install if the user confirms):

• founder — when the bottleneck is the founder's own psychology or operating system, not the company
• product-market-fit — deep PMF measurement and iteration loops beyond the signals here
• saas-metrics — CAC payback, NRR bands, magic number, and the full SaaS metric stack
• venture-capital — term sheets, investor dynamics, and round mechanics in depth
• hiring — running the actual hiring process once this skill says the timing is right

Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/startup.