The 6 SOPs to Write First

PART 10: THE 6 SOPs TO WRITE FIRST (Before Your First Hire)

If you read nothing else from this document, write these six:

1. Decision Journal

Format: A single document (Notion/Handbook). Every major decision gets:

• Date
• Context (what prompted this decision)
• Options considered (with tradeoffs for each)
• Decision
• Expected outcome (quantified if possible)
• Actual outcome (revisited in 3, 6, 12 months)
• Lessons learned

Why it matters: Most startup failures aren't from wrong decisions — they're from failing to learn from decisions. The journal creates a feedback loop.

2. Communication Charter

One page. Covers:

• Async default (write first, then talk)
• Meeting rules (agenda required, decisions documented, no meeting without outcome)
• Update cadence (weekly written updates)
• Slack/chat discipline (public channels, no "hey" pings, context included)
• Urgent vs non-urgent (how to escalate)

Why it matters: Communication overhead is the #1 hidden cost as teams grow. A good charter prevents the meeting-ification of your company.

3. Financial Operating Procedure

Covers:

• How money enters the company (invoicing, payment processing)
• How money leaves (expense policy, approval thresholds, vendor payments)
• How it's tracked (bookkeeping tool, close schedule, reconciliation)
• How it's forecast (model template, update cadence)
• How it's reported (board deck template, metrics definitions)

Why it matters: Bad financial ops kill more startups than bad products. You can't course-correct if you don't see the numbers.

4. Hiring & Onboarding

Covers:

• When to hire (signal: consistent pain, not one-off)
• How to write a job spec
• Interview process with rubric
• Offer process
• Day 1 checklist
• 30-60-90 day plan template

Why it matters: A bad hire at a 5-person company is 20% of your team. The process is your best defense.

5. Engineering Workflow

One page for the basics:

• Branch strategy (GitHub Flow or trunk-based)
• PR requirements (tests, review, no direct-to-main)
• CI/CD (every commit builds, deploy auto on merge)
• On-call rotation (even if it's just you)
• Incident response (1-page template)

Why it matters: Consistency beats heroics. A documented workflow = new engineers productive in days, not months.

6. Knowledge Base Rules

Covers:

• SSOT: every fact lives in exactly one canonical place
• Decision → doc: every decision produces a documentation update
• Template policy: every recurring process has a template
• Stale doc policy: quarterly review, archive what's outdated
• Who can edit what (permissions by role)

Why it matters: Without this, knowledge lives in Slack, DMs, and people's heads. That doesn't scale past 3 people.

---

KEY INSIGHTS FROM SUBAGENT RESEARCH

The following sections incorporate findings from dedicated subagent research across 8+ hours of browsing live sources.

The Self-Improving Company (YC, July 2026)

The new paradigm from YC's "How to Build a Self-Improving Company with AI": Companies designed with AI feedback loops baked in. Every customer interaction feeds the AI and improves the product automatically. The org structure becomes a learning system, not a hierarchy.

Design implication: your knowledge base, support tickets, sales calls, and product usage data should all feed back into AI models that improve product, support, and operations continuously.

Two GTM Playbooks (a16z, July 2026)

From "Lighthouse or Landgrab?":

• Lighthouse: Marquee customers for social proof. Use when creating a new category — buyer needs education. Requires founder-led sales, high-touch.
• Landgrab: Broadcast market fast. Use when buyer knows the problem — just needs better math. Requires product-led growth, self-serve.
• Org design must match: Lighthouse = high-touch, founder-led. Landgrab = product-led, automated.

The Friction Paradox (Simon Willison / Armin Ronacher)

"Before agents, some of this shared understanding was maintained by friction. If I wanted to change your storage layer, I usually had to read your code, ask you questions, and perhaps coordinate with another team... This friction synchronizes people."

Implication: When AI removes the friction that naturally kept teams aligned, you must deliberately replace it. AI-native companies need structured synchronization mechanisms (written updates, design reviews, decision records) because the invisible coordination that happens in human teams doesn't happen with agents.

Team Topologies: The Foundation for AI ROI

Matthew Skelton's framework is the modern standard for organizing teams around fast flow of value, and now positions itself as the foundation for AI ROI.

Four Team Types:

1. Stream-aligned — Aligned to a flow of work from a business domain (the primary team type)
2. Enabling — Helps stream-aligned teams overcome obstacles. Specialists who enable, not do.
3. Complicated Subsystem — For deep expertise domains (graphics, payments, etc.)
4. Platform — Internal product that accelerates delivery for other teams

Three Interaction Modes (apply to both human and AI agents):

1. Collaboration — Two teams/agents work together to discover solutions
2. X-as-a-Service — One team provides, another consumes (clear API/contract)
3. Facilitation — One team helps and mentors another (temporary)

AI relevance: AI agents need the same interaction modes as human teams. The "infrastructure for agency" concept means designing how human and AI teams collaborate, consume services from each other, and facilitate each other's work.

Documentation Is the New Watercooler

AI can't read your mind or Slack history — it can only read what's written down. Handbook-first / docs-as-code becomes a prerequisite for AI integration. Companies that already document everything (GitLab-style) will onboard AI much faster than those that rely on tribal knowledge.

DACI vs RAPID Decision Frameworks

Rule: Write down the framework for each major initiative. Prevents "decisions by committee" (slow, bad) and "decisions by dictatorship" (fast, potentially wrong).

Amazon Writing Culture: The 6-Pager

Amazon's primary communication format for meetings. A narrative document (not slides). Meetings start with 20-30 min of silent reading. Structure: Context → Problem → Approach/Options → Rationale → Open Questions → Appendix.

PR/FAQ: Before building anything, write a press release announcing the product as if launched, plus FAQ. Forces clarity on what and why.

Key practices: Narrative over bullets (forces rigorous thinking), data-driven arguments, disagree-and-commit, single-threaded leadership (one owner per initiative).

Radical Candor Framework (Kim Scott)

Two axes — Care Personally × Challenge Directly:

Implementation: Start with yourself (solicit negative feedback). Create Rockstar (stable) and Superstar (growth) career paths. Career conversations quarterly, not annually. Separate recognition from compensation.

SOP Prioritization by Company Phase

Phase 1 (<10 people):

1. Values definition (CREDIT-style)
2. Handbook skeleton (SSOT)
3. Decision log (DACI/RAPID)
4. Communication norms (async-first)
5. Shape Up betting table

Phase 2 (10-50 people):

1. Hiring & onboarding handbook
2. Team topology design
3. Writing culture (6-pager format)
4. Feedback framework (Radical Candor)
5. Outcome-based planning (OKRs)

Phase 3 (50-200 people):

1. Department handbooks
2. Platform team formation
3. Empowered product teams
4. Formalized decision frameworks
5. Weekly/monthly business reviews

Phase 4 (200+ people):

1. Governance & compliance
2. AI integration policy
3. Single-threaded ownership
4. Cross-team interaction modes
5. Continuous COE (correction of errors) culture

The Age of the Solo Founder (YC)

YC's "The Age Of The 40-Year-Old Solo Founder Is Here": Experienced founders (35-50) building companies solo with AI as co-founder. AI handles code, support, marketing, bookkeeping, sales outreach. No functional heads needed until much later.

For Nathan's company OS: Design for 6-9 human team members, not 20+. AI agents fill the gap between traditional headcount requirements.

Sean Ellis: PMF in the AI Era

The Sean Ellis Test ("Would you be very disappointed without this product?") remains the gold standard, but the iteration cycle to reach 40%+ "very disappointed" has compressed from months to weeks. AI agents can build, test, and iterate on hypotheses overnight.

Economic Shift: AI Agent Costs

Simon Willison: The LeadDev-Equivalent for AI Engineering

simonwillison.net is the authoritative source for practical, honest AI engineering content — covering coding agents, LLM economics, agent security, and the practical realities of building with AI. Essential weekly reading for any AI-native engineering lead.

---