Key Takeaways

  • AI coding agents break the traditional roadmap because writing code is no longer the scarce resource in software development.
  • Product teams must adopt “durable convictions but disposable features,” anchoring their strategy to long-term beliefs while treating individual implementations as temporary tests.
  • Claire Vo distinguishes between two mindsets: “Good stubborn is you stay with the problem. You revise the solution.” In contrast, "bad stubborn" uses cheap tokens to move the goalposts and ship endless iterations that avoid market reality.
  • Product managers must drop emotional attachment to code, treating software generation like a high-speed factory designed to intersect reality immediately.
  • Teams can replace static delivery schedules with Vo's Conviction-Led Product Operating Method.

The Vo's Conviction-Led Product Operating Method

Step 1: Build Your Convictions

Define the long-term direction and what you believe the future looks like 1 to 2 years out, rather than mapping 3-to-9-month feature timelines.

Step 2: Determine Evidence

Define upfront exactly what customer evidence would prove the conviction true, as well as clear failure criteria that would dictate stopping.

Step 3: Run the AI Factory

Use rapid AI building capacity to intersect reality as quickly as possible, creating testable software without emotional attachment to the code.

Step 4: Allocate Resources Based on Truth

Evaluate real-world feedback and allocate engineering tokens, investment, and effort only toward validated convictions, discarding failed solutions.

When This Works (and When It Doesn't)

This method applies directly when your team uses high-velocity AI coding tools, moving engineers from estimating delivery dates to rapidly testing market hypotheses. When code is cheap, your main risk changes: you are no longer limited by engineering hours, but by your ability to face reality when an experiment fails. As Vo notes, “what you want to see is clear progress against your convictions, but zero ego about your solutions.”

This model breaks down in heavily regulated or safety-critical environments where disposable software creates compliance liabilities, such as medical devices, core banking ledgers, or aerospace control software. In those domains, testing unverified code directly against customers introduces severe legal and operational risks. It also struggles when enterprise buyers require contractual feature guarantees before signing multi-year agreements.

What to Do With This

Take the primary feature planned on your team sprint board this week. Instead of tracking its delivery date, write out the 18-month market conviction behind it. Define the single metric or customer action that proves this conviction right within fourteen days of launch, along with the specific failure number that forces you to delete the code.

Use AI coding agents to ship that test within forty-eight hours. If the test misses your threshold, delete the feature branch immediately rather than scheduling follow-up patches. Reinvest your engineering tokens into an entirely different approach to the same customer problem.