SaaStr 842: The 90/10 Rule for AI Agents: What to Build vs Buy with SaaStr's CEO and CAIO

What It Covers

SaaStr CEO Jason Lemkin and CAIO Amelia Ibarra outline their 90/10 rule for AI agent strategy: buy 90% of needed tools off-the-shelf and only build the remaining 10% when no adequate solution exists. They demonstrate this framework through two recently vibe-coded internal tools built using Claude, Replit, and Claude Cowork.

Key Questions Answered

• •The 90/10 Build vs. Buy Framework: Default to purchasing existing AI tools for 90% of needs, reserving custom development for three specific scenarios: no adequate off-the-shelf solution exists, proprietary data cannot be fed into third-party tools, or an existing paid tool lacks AI features entirely. SaaStr operates with three humans, one dog, and 20-plus agents using this discipline to avoid unsustainable maintenance overhead.
• •Vibe Coding Readiness Threshold: Attempting to build production-facing apps with single sign-on, external customer access, or complex integrations requires prior experience shipping simpler tools first. SaaStr built multiple websites, calculators, and internal agents over nine months before tackling their sponsor portal. Setting a hard time limit — one day to solve the hardest technical problem or revert to the existing tool — prevents sunk-cost spirals.
• •Claude Cowork as a Force Multiplier: Claude Cowork, running locally on a machine with browser access, processed all sponsor contracts in roughly one hour, extracted pass counts and signing dates, then autonomously generated unique registration promo codes inside a third-party event platform. A task previously requiring a full human workday was completed in under two hours, demonstrating agent-level automation for document-heavy operational workflows.
• •Zero-AI Products Face Immediate Churn Risk: SaaStr canceled a $10,000 annual sponsor portal subscription after building a superior replacement in roughly two days. The trigger was the incumbent tool having no AI features whatsoever in early 2026. Founders should audit every product in their stack: if a non-technical operator can vibe-code a better version in 48 hours, that product is at immediate displacement risk and losing customers silently.
• •The Jaw-Drop Test for Product Survival: SaaS founders should use their own product and ask whether it produces a jaw-dropping reaction comparable to using Claude or a top AI agent. Products that originally won customers by being dramatically better — not 1% better — must recapture that gap through AI. Companies failing this test are experiencing decay, and the market is already pricing that in, with strong-growth public SaaS companies trading at four times ARR.
• •Agent Maintenance Is the Underestimated Cost: Running 20-plus agents and eight custom-built apps creates compounding maintenance overhead that consumes significant founder time. Monthly deep security audits conducted by the coding agent itself surface dozens of issues, each taking roughly a week to resolve without breaking functionality. Teams should realistically assess how many people can manage agent infrastructure before adding new tools — SaaStr considers itself at maximum capacity despite being a lean operation.