Pattern Library
Purpose: Recurring startup outcomes Magi cites in Founder Mode. Each pattern is durable across stages and decades. Read when: You are in DIAGNOSE / ADVISE phase and need pattern grounding.
Founder Mode advice must cite ≥1 pattern by ID. Patterns are priors, not absolute truths — when the founder's evidence overrides the pattern, follow the evidence, but acknowledge the pattern explicitly.
Foundational Patterns
P-01: Make Something People Want
Demand precedes everything. If users won't pull the product, no amount of build, sell, or fundraise fixes it. Test: paid users, organic referrals, daily active use, or payment intent within 4 weeks of contact.
P-02: Talk to Users (continuously)
Founders who talk to ≥5 users/week converge on PMF faster than those who don't. Most founders don't. Default action when in doubt: schedule user conversations this week.
P-03: Default Alive vs Default Dead
At current burn and growth rate, will you be cash-flow positive before runway ends? Default-alive companies have permission to focus on quality. Default-dead must change something within 30 days.
P-04: Do Things That Don't Scale
Manual onboarding, white-glove support, hand-built integrations are correct in early stages. The premature instinct to automate is a signal of avoiding customer contact.
P-05: Launch
Shipping is itself a learning event. The information from a real launch exceeds 100 hypothetical user calls. "If you're not embarrassed by your first launch, you waited too long."
P-06: Focus on One Metric
Pick one metric that captures real value (revenue, retention, engaged users). Track weekly. Everything that doesn't move that metric is a candidate for cut.
P-07: 1% Weekly Growth Floor
Sustainable startup growth is roughly 5-7% week-over-week at seed stage. <1% is stagnation. Below the floor, the bottleneck is product-market fit, not execution.
P-08: Founders Do the Work
Until traction, founders are in code, sales, and support. Delegation comes after PMF, not before. Hiring is not a substitute for founder time on the most important problem.
Customer Patterns
P-10: One Specific Customer
"Everyone" is no one. Pick one job title, company size, problem profile. Sell only there for 4 weeks. Generalization is earned by traction in a niche, not promised in a deck.
P-11: Pull > Push
If you're pushing the product on customers, you don't have demand. If they're pulling it from you, you do. Signals of pull: organic referrals, unprompted re-engagement, paid retention.
P-12: Talk to Churned Users
Churned users teach more than active ones. Schedule exit interviews. Don't optimize what works; investigate what broke.
P-13: Customer Pain > Customer Want
Founders confuse "users say it'd be cool" with "users will pay to remove pain". Only the second pulls revenue. Probe for what the user is currently doing painfully without the product.
Product Patterns
P-20: Cut Scope, Don't Add Features
The default response to "users want X" is to cut Y, not add X. Total complexity is the constraint. New features rarely fix old retention problems.
P-21: Activation > Acquisition
Acquiring 1000 users with 5% activation is worse than acquiring 100 with 50% activation. Fix the funnel from middle out before pouring more in.
P-22: One Killer Feature
Successful early products usually have one 10x feature, not five 1.5x features. If you can't name your one killer feature in one sentence, that's the work.
P-23: Manual First, Automate Later
The first version of any operational process should be a Google Doc, a spreadsheet, or a human in the loop. Code comes after the manual version proves the workflow.
Founder Patterns
P-30: Cofounder Drama is Fatal
A majority of failed startups cite cofounder conflict as a primary cause. Treat cofounder dynamics as P0. Equity, decisions, communication cadence — written down.
P-31: Founder-Mode (literal contact)
Founders who stay in customer/code/sales contact through scaling outperform delegators. Strategy without contact is fiction. "Working on strategy" while not talking to users is a red flag.
P-32: Persistence > Ideas
Most successful startups iterated through wrong ideas before finding the right one. Persistence in problem space matters more than first idea quality. Pivots that abandon the problem space are usually a different startup.
P-33: Founder Energy is the Hidden Constraint
Founders burn out from context switching, not from total hours. Two priorities is one too many. One bottleneck per week is a discipline.
Fundraising Patterns
P-40: Customer Signal Precedes Investor Signal
Investors fund traction. Founders chasing investors before customers are usually substituting investor validation for product-market fit. Default order: 5 paying customers → investor meetings.
P-41: Raise Enough, Not More
Take what you need to reach the next milestone with ~12-18 months buffer. Larger rounds dilute and create false confidence. Excess capital often funds anti-patterns (premature hiring, premature scaling).
P-42: Fundraising is a Sprint, Not a Default Activity
Fundraising should be a 4-6 week dedicated sprint, not a continuous background task. Continuous fundraising correlates with under-execution on product.
Operating Patterns
P-50: Weekly Cadence
Weekly review of the one metric, weekly commitment to one priority, weekly check on user conversations. Monthly is too slow at startup speed.
P-51: Cut Three Things This Week
The default action in a stalled startup is subtraction, not addition. Cancel a partnership, kill a feature, drop a customer segment. Reclaim attention for the core.
P-52: Reversible Decisions Deserve Speed, Not Deliberation
Decisions reversible within 30 days should be made in <1 day. Hours of deliberation on reversible decisions is a sign of paralysis, not rigor.
P-53: Idea Maze Mapping
Before claiming a novel idea, map prior failed attempts in the space. The pattern of past failure is the constraint your design must avoid.
P-54: Two-Way Door / One-Way Door (Bezos)
Classify the decision before debating it. Two-way doors are reversible within ~30 days (feature toggles, copy, pricing experiments, hire of a contractor, small vendor swap) — make these fast, by a single owner, with no committee. One-way doors are irreversible or expensive to reverse (cofounder split, equity grants, major acquisition, regulated certification, brand rename, key engineering migration with data loss risk) — these get deliberate analysis and a wider consultation. Most founder paralysis comes from treating a two-way door as a one-way door, or — more dangerously — treating a one-way door as a two-way door (e.g., shipping a brand decision in a day, then discovering it cannot be unshipped). Anchored in Bezos's 2015 Amazon shareholder letter, still the canonical reference.
2026 AI-Era Patterns
P-60: Moat Is Distribution and Workflow, Not Model
Foundation-model access has commoditized at the API layer. In 2026, the durable moats are: (a) frontier compute scale (only a handful globally), (b) hypergrowth revenue with strong retention crossing $100M-$1B ARR, or (c) a vertical owned by deep workflow embedment in regulated industries (law, medicine, defense, insurance). If your pitch starts with "we use [model X]", the moat is missing. Source: Garry Tan, YC President — "AI strategy not tied to a specific workflow, KPI, and operating model is mostly theater" (ycombinator.com/rfs, Spring 2025 RFS release); Andrej Karpathy vibe-coding framing Feb 2025 — building is easy, distribution is hard.
P-61: Sell Completed Work, Not Tools
The winning AI companies sell outcomes (a closed ticket, a filed return, a coded medical claim, a settled case), not "AI-powered tools". "The customer is buying the outcome directly. The work budget in any profession dwarfs the tool budget, and autopilots capture the work budget from day one." Founders pitching capability without a measurable outcome, KPI baseline, and replaced budget line are pitching theater. Action: replace every "AI-powered X" sentence with "We do Y job, you pay per Y completed, here is the SLA." Sources: Sequoia, "Services: The New Software" (sequoiacap.com/article/services-the-new-software/) and Sequoia AI Ascent 2026 (sequoiacap.com/article/ai-ascent-2026/); YC RFS 2025-2026 (ycombinator.com/rfs).
P-62: Talk-to-Users Floor Holds Through AI Era
Vibe-coding lets one founder ship a working app in a week (Karpathy, Feb 2025), but the talk-to-users floor (P-02) does not change — if anything it rises, because every wrapper-style competitor will also ship. YC's core principle "make something people want" remains unchanged as the #1 failure mode. Easy shipping does not substitute for evidence of pull. Source: ycombinator.com/library ("YC's Essential Startup Advice").
P-63: Product-User Fit Precedes Product-Market Fit
Before claiming PMF, confirm the right user is pulling the product — not just any user. Founders often confuse a sharp sense of product-for-user with full product-market fit, and use that self-declaration to hire, fundraise, and scale prematurely. The correct sequence: nail the specific user → confirm pull → then expand market definition. Source: a16z, "Product-User Fit Comes Before Product-Market Fit" (a16z.com/product-user-fit-comes-before-product-market-fit/).
P-64: Sequoia Arc PMF Archetypes — Three Paths, Not One
Sequoia's Arc framework (2025) identifies three distinct PMF archetypes based on how customers perceive the problem: Hair on Fire (urgent known pain — requires both strong product and fast GTM), Vitamin (nice-to-have that founders must prove is a painkiller over time), and Hard Fact (customers don't know they need it — requires education before demand). Each archetype requires a different validation strategy and timeline. Founders who apply the wrong archetype's tactics to their PMF type stall. Source: sequoiacap.com/article/pmf-framework/ and sequoiacap.com/article/pmf-framework-2/.
Pattern Citation Format
Every piece of advice cites at least one pattern by ID:
"The bottleneck is user conversations. You haven't talked to ≥5 users this week (P-02). Below the talk-to-users floor, no amount of building closes the gap (P-01)."
Multiple patterns can stack:
"Three pivots in five months (AP-05) combined with no monthly metric reporting (AP-03) is a focus problem, not an idea problem (P-32, P-06). The action is committing to current direction for 8 weeks and instrumenting retention."
When the Pattern Fails the Founder
Patterns are priors, not laws. When the founder's evidence overrides the pattern, follow the evidence:
"Normally I'd say P-02 — talk to users. But you've done 30 user interviews this month and the bottleneck is downstream of demand. Let's look at activation."
Acknowledge the pattern, then update on evidence. Never skip the pattern silently.