The Second Opinion Positioning Play: Framing Your AI as a Check on Experts — Not a Replacement for Them — Is the Fastest Path to Trust.

The Play

Antoine ran 266MB of DICOM MRI files through Claude Code using Opus 4.8 and asked it to read his shoulder scan independently. His doctor had diagnosed a Grade III partial-thickness tear. The AI found no tear. An arbitration pass concluded: "Evidence favours Reader A (moderate-to-high confidence). Mild insertional tendinosis; NO discrete partial- or full-thickness tear identified."

The AI also flagged that his prescribed treatments — shockwave therapy without calcification present, and a Traumeel injection — had questionable clinical backing.

The post hit 276 points on Hacker News.

What made it resonate was not the AI capability. It was the framing.

Antoine did not say "AI replaced my doctor." He said he used AI to get a second opinion. That framing — so simple it barely registers as a product decision — is the single highest-leverage positioning move available to any AI product going to market in 2026.

Why "Second Opinion" Works Where "Replacement" Fails

When you tell a user your product replaces an expert, you are asking them to accept three things simultaneously:

1. The expert was wrong or overpriced

2. The AI is more reliable than the expert

3. They should trust a product they have not yet used over a relationship they already have

That is a hard sell. It requires the user to be already dissatisfied, already skeptical, and already willing to take the reputational risk of acting on AI output over professional advice.

When you tell a user your product gives them a second opinion, you are asking them to accept one thing:

1. An additional data point before a high-stakes decision is useful

That is not a sell. It is a statement almost everyone already believes.

The user who would push back against "AI replaces your doctor" will immediately recognize the value of "AI reads your MRI before you decide whether to get surgery." The decision being informed is the same. The framing determines whether the user sees your product as a threat to their existing relationships or an addition to them.

The Output Design That Makes It Work

The second-opinion framing only holds if the output reinforces it.

Antoine's experiment produced findings — but the most actionable thing it generated was a list of questions: why was shockwave therapy prescribed without calcification? What is the clinical evidence for the prescribed injection?

That output design is not an accident. It keeps the human expert central to the process. The AI does not conclude that the doctor was wrong. It generates questions for the next appointment. That framing makes the patient feel equipped rather than defiant — and it protects the product from the "AI playing doctor" objection.

Design your output as preparation for a human conversation, not a replacement for it. The question list is the product.

Where the Demand Already Exists

The communities where people are already seeking second opinions manually are the distribution channels for this framing.

Chronic illness forums. Rare disease patient groups. Orthopedic recovery communities. Legal question subreddits. Financial planning Facebook groups. Any space where people post complex professional documents and ask strangers with adjacent expertise to help them understand it.

These users are already doing the behavior your product enables. They are already seeking a second opinion. They are already treating peer input as a check on professional advice. Your product is a better version of what they are already doing — and the framing of "second opinion" makes that immediately legible.

Applying This to Your Product

The second-opinion play is not limited to healthcare. It applies to any domain where:

• An expert produces an output the user cannot fully evaluate
• The user has a high-stakes decision to make based on that output
• A second read adds information the user can act on

Legal documents. Architectural plans. Financial projections. Code security reviews. Marketing attribution reports. Contract redlines. Lab results. The professional produces something. The user does not know what to do with it. Your AI gives them a read before they decide.

The pricing logic follows from the decision, not the compute. Antoine's second opinion was worth doing because the alternative was committing to a treatment that might be unnecessary. Price your product relative to the cost of getting the decision wrong, not relative to the cost of running the model.

The Trust Flywheel

The second-opinion framing creates a trust flywheel that replacement framing does not.

When the AI output matches the expert, the user feels validated and trusts both. When the AI output conflicts with the expert, the user has a reason to go back with better questions — and the conversation that follows either confirms the original recommendation or surfaces something worth knowing. Either outcome is a win.

Replacement framing has no equivalent flywheel. If the AI is right, the expert looks bad. If the AI is wrong, the product looks bad. There is no scenario where both parties come out ahead.

Second-opinion framing is additive by design. It creates a world where AI and experts coexist, where the user is more informed, and where the product earns trust through every use case rather than only the ones where it wins.

Antoine's final line says it better than any product copy: "My hope is that in a couple of model generations, we'll trust AI to review MRIs the way we trust it to proofread our emails."

Proofread our emails. Not replace the person writing them.

That is the positioning. Build toward it.

Source: https://antoine.fi/mri-analysis-using-claude-code-opus