Frequently Asked Questions

An AI product designer decides where AI belongs in a product, designs how people interact with it, and guides the team until it ships. The focus is on whether people adopt it and trust it, not on whether the model can technically do the task.

Regular product design assumes the software behaves the same way every time. AI does not. AI product design adds the work of showing reasoning and sources, deciding where a person stays in control, and designing for the moments when the model is uncertain or wrong.

Most start with a two-week diagnostic that maps where AI should pay off first. Designing and proving the experience usually takes four to ten weeks. Teams that ship AI continuously keep ongoing design leadership on retainer.

Yes. My most recent work has been in healthcare and commerce. In healthcare, that includes patient experience, self scheduling, and clinical workflow design.

AI experience design is the design of how people interact with AI inside a product: what the AI shows, how it explains itself, when it hands control to a person, and how it earns enough trust that people use it.

It builds on UX design, but AI does not behave the same way every time. AI experience design adds patterns for uncertainty, sources and citations, human review at high-stakes moments, and recovery when the model gets something wrong.

Show the work. People trust AI they can check, so the experience should surface reasoning, confidence, and sources, and keep a person in control for decisions that are high stakes, regulated, or uncertain.

Only if the review step is designed badly. Done well, the AI surfaces its reasoning so the person decides quickly and with good information. A small cost at the right points buys the trust that makes people use the system at all.

It depends on the format. The two-week diagnostic is a fixed scope, Adoption System Design is scoped during the diagnostic, and the retainer is an ongoing monthly arrangement. The 30-minute advisory intro is where we size the work together.

Yes. Pineapple Blueprint is based in San Diego and works with teams across the United States. Most work happens remotely, with in-person sessions in San Diego and on-site visits elsewhere when the work calls for it.

The advisory intro is a 30-minute structured conversation. We check fit, define the problem you are trying to solve, and decide on the right next step, if there is one.

Both. I do the design work myself, and when a project needs more hands I bring in specialists from a trusted network while the leadership and accountability stay with me.

A rule in a prompt is a suggestion the model weighs against everything else in its context. Under a long session or a big request it loses. A rule in code is checked every time, and the model gets a numbered list of what it broke.

Four things: one source of truth for the rules, validators that check every output against them, a loop that fixes what it can and retries before a person sees it, and a log of every decision and every correction so the system learns instead of repeating itself.

Yes, with a cost. On the same request, an 8 billion parameter model with no rails produced 20 rule violations in 42 seconds. With rails it shipped clean after three loops and about four minutes. The rails traded speed for output that never needs a human to catch the basics.

The judgment. Which rules belong in code and which in the prompt, what the decision log should say for that brand, an audit that finds where the agent actually drifts, and the approvals that only accumulate from that team's own taste. The pattern is free. The rules are yours.

It means machines cannot reliably find, understand, or operate your product through the surfaces you ship. Scores are graded per point: my own site passed 5 of 10 points with one partial at baseline, including a contact form that reported success on submissions nothing received.

Three checks: submit it with JavaScript disabled and see whether anything is captured, confirm the success message only appears when the server actually accepted the submission rather than whenever a fetch resolves, and verify a notification fires somewhere a human reads. If any of the three fails, you may be thanking visitors for messages you never got.

Faster than most teams expect once the problems are named and ordered. My own fix pass went from 5 passing points to 9 in one evening across three commits, because the audit told me exactly what blocked transactions and what could wait. Larger systems typically land the same class of fixes over a quarter.

Look for someone who combines design system depth with daily, hands-on agent operation, because the failure modes only reveal themselves when you watch agents work. David Glazier and his team run a structured 10-point audit for mid-size startups and enterprises, using real browser agents on real flows, delivered as a prioritized fix list your team executes.

Typically a few weeks, depending on the size of the system. The output is a scored assessment and a sequenced remediation plan. Most of the fixes then land in-house over a quarter.

They overlap heavily but are not identical. Accessibility work covers roughly half the checklist. AI readiness adds machine-facing concerns accessibility never covered: structured data, llms.txt, ingestible documentation, and agent-testable flows.

Start where AI can move a real business metric, not where it is easiest to bolt on. Scan your core workflows, rank the opportunities by impact and effort, and commit to the highest-impact one first with a clear owner and a target. One adopted win beats ten stalled pilots.

Usually because effort is spread across too many unowned pilots, or because the tools are not adopted into daily work. In WRITER's 2026 survey only 29% of companies saw significant ROI while 79% struggled with adoption. The fix is prioritization and adoption, not a bigger model.

Measure behavior, not usage. Track adoption rate, decision and cycle speed, and output quality, then tie those to revenue, cost, or risk. Token counts and license seats tell you what you spent, not what you gained.

They give AI a consistent foundation to build from: standardized tokens, components, and patterns, so generated output looks and behaves the same every time. That turns AI speed into shippable quality instead of fragmentation.

No, it raises it. The more code and UI you generate, the faster inconsistency spreads without a shared system. A design system is what keeps high-volume AI output coherent.

Define your tokens, components, and patterns, make them the source the AI builds from, and add human review at the decision points that matter. Consistency comes from the system underneath, not from the prompt.

Most fail for organizational reasons, not technical ones: no clear owner, no user trust, and poor fit with existing workflows. In WRITER's 2026 survey, 79% of organizations struggled to adopt AI even though the underlying models worked.

Pick one high-value workflow, give it an owner and a target metric, design for trust with human-in-the-loop control and citations, fit it into existing tools, and track adoption and decision speed. Scale the next one only after the first is genuinely used.

Trust and fit. People keep using AI when they can verify its output and when it lives inside the way they already work, and when leadership measures adoption instead of declaring victory at the demo.

Only if AI is material to your business, you report on it publicly, or you are heavily regulated. Most mid-market companies post-pilot get what they need from a fractional AI leader or an advisor, at a fraction of the cost.

A senior AI leader embedded part-time, often about a day a week. You get executive-level judgment and accountability without a full-time hire, typically for a few thousand dollars a month.

Build in-house when AI is core and ongoing and you can attract the talent. Bring in an advisor or fractional leader when you need senior direction now, want a prioritized plan, or have a specific problem to solve before committing to headcount.

Start with scope, not a title search. Define the handful of decisions you need senior AI judgment on each month, then bring in someone director or staff level from a named AI shop for about a day a week. Most engagements run a few thousand dollars a month, far less than a full-time hire.

An advisor or consultant is scoped: a diagnostic, a roadmap, one problem solved, then you're done. A fractional head of AI is a standing part-time seat with ongoing accountability. Pick a consultant for a focused question, a fractional leader when you need judgment in the room every week.

Behavior and outcomes: adoption rate, decision and cycle speed, and output quality, each tied to revenue, cost, or risk. Those connect spend to impact in a way a board will believe.

Because they measure spend, not value. High usage with no change in business outcomes is a cost, not a return. Usage is an input metric; ROI lives in what that usage changes.

Often because the wrong things are being measured, or effort is spread across pilots no one owns. In WRITER's 2026 survey only 29% of companies saw significant ROI. Pick one high-value use case, instrument adoption and outcomes, and prove it before scaling.

Track three things and tie each to revenue, cost, or risk: adoption rate, decision and cycle speed, and output quality. Measuring AI ROI starts before you build, not after, by setting the metric and the baseline up front.

Adoption rate is one of the three core metrics, not the whole story. Pair it with decision and cycle speed and output quality so you can see whether people using the tool actually changed the outcome.

Human-in-the-loop AI keeps a person in control at specific decision points instead of fully automating them. The AI does the work and surfaces its reasoning; the person reviews, approves, or overrides where the stakes justify it.

Keep a human in the loop for high-stakes or irreversible decisions, regulated contexts, and low-confidence or ambiguous cases. Automate the routine, reversible, and high-confidence work where being wrong is cheap.

Only if the loop is designed badly. Done well, the AI surfaces its reasoning and confidence so the human decides fast and informed. The small cost at the right points buys the trust that makes people use the system at all.

AI governance is the set of guardrails, decision boundaries, and oversight that defines how AI can be used, with what data, and who is accountable. Done well it reduces risk while keeping teams fast, rather than blocking them.

Make the safe path the easy path: approved tools that are genuinely good, clear rules, sensible defaults, and risk tiers so low-risk work moves fast while high-risk work gets more oversight. Embed the guardrails in the workflow instead of a policy document.

Shadow AI is employees using unapproved AI tools outside of any oversight. It usually grows when teams ban AI instead of governing it, and it is a leading source of data exposure: 67% of executives believe unapproved AI has already caused a breach at their company.