Skip to content

Checklist: AI Readiness Assessment

Is your company ready to go AI-first?

Use this checklist before committing to an AI-first transformation. Work through each category with your leadership team to identify gaps and blockers. Items left unchecked represent areas that need investment before -- or during -- the transition. Not every box must be checked to start, but you should have a plan for each unchecked item.

Derived from the AI-First vs AI-Enabled and 7 Mental Models of AI-First frameworks -- Chapters 1 and 2.

Strategic Readiness

  • Applied the Kill Test: you can articulate whether removing AI would eliminate your product or merely degrade a feature
  • Determined whether you are building AI-first (AI is the product) or AI-enabled (AI enhances an existing product) -- and committed to one approach
  • Defined your AI value proposition in one sentence that doesn't include the word "also"
  • Assessed whether your pricing reflects AI as core value (consumption-based or hybrid) rather than a premium add-on
  • Identified which of the 5 signs of AI-first apply to your company: founded after modern AI, data-driven architecture, distributed AI expertise, AI-core pricing, "built on" positioning
  • Evaluated whether AI-enabled is actually the better choice for your context (large incumbent, heavily regulated, hardware-first, or human-judgment product)
  • Established a quarterly re-evaluation cadence for your AI strategy as the landscape shifts

Technical Readiness

  • Adopted an agent-first design posture: APIs are structured for AI consumers with explicit error handling and structured responses, not just human-operated interfaces
  • Designed systems for probabilistic outputs: confidence levels are surfaced, not hidden -- uncertainty is treated as a feature
  • Built or planned abstraction layers that allow swapping models without rewriting applications
  • Evaluated the build vs. buy inversion: checked whether building with foundation models is faster than procuring, integrating, and customizing vendor solutions
  • Established infrastructure for compound iteration: automated evaluations, fast feedback loops, and regression test suites for AI outputs
  • Defined a permission spectrum for AI autonomy: low-stakes actions run autonomously, high-stakes actions require human approval

Organizational Readiness

  • AI expertise is distributed across teams, not siloed in a single "AI/ML team"
  • Leadership understands the distinction between AI-first and AI-enabled at the strategic level
  • Product, engineering, and business stakeholders are aligned on which approach you are pursuing
  • Team structure supports rapid iteration (hours-to-days cycles, not quarterly sprints) for AI features
  • Roles are defined around human-AI collaboration: AI handles execution ("how"), humans provide judgment, taste, and direction
  • A plan exists for retraining and upskilling existing staff rather than only hiring new AI talent

Data Readiness

  • Identified whether your data architecture drives the product (AI-first) or merely supports features (AI-enabled)
  • Product interactions are structured to naturally generate training signals -- usage data is treated as a product, not a byproduct
  • A data flywheel is designed or planned: more users generate more data, more data improves the product, a better product attracts more users
  • Data freshness requirements are defined for all critical data sources
  • Data is accessible across departments rather than fragmented in silos
  • Data pipelines exist (or are planned) that can feed continuous model improvement without manual intervention

Cultural Readiness

  • The organization treats AI outputs as probabilistic rather than expecting deterministic perfection
  • Teams are comfortable shipping AI features that are "good enough" and iterating, rather than waiting for 100% accuracy
  • There is organizational tolerance for the risk profile of AI-first: model failure means business failure, not just feature degradation
  • Marketing and communication use "built on AI" language rather than "now with AI" -- the framing reflects genuine commitment
  • Post-incident analysis is standard practice: every AI failure is treated as a training example, not a blame event
  • The company designs for augmentation, not replacement -- AI tools make people more capable rather than surveilling or displacing them

Next step: For items you checked, validate with specific evidence. For items you left unchecked, assign an owner and a target date. Revisit this assessment quarterly.