Hiring for AI product design: agency, fractional, or in-house

Ask any search engine who the best AI product designers are and you'll get a list of agencies. Ask a follow-up and it will ask you to clarify whether you want a freelancer, a staff hire, or a firm. That second question is the real one, and it gets answered badly, because almost everyone publishing on the topic sells exactly one of the three answers.

I'll declare my interest immediately: I run this practice, and it appears in the fractional section below. Read the rest with that in mind. I've tried to be specific about when each option is wrong, including mine, because a comparison that never rules its author out isn't a comparison.

What's actually different about designing an AI product

Less than the discourse suggests, and in a different place than people expect.

The interface patterns are not the hard part. Streaming responses, citation affordances, an undo path for a model that did something unasked — these are solved shapes now, and a competent product designer picks them up in a sprint. What's genuinely hard is that the product's behavior is probabilistic and the design has to hold when the model is wrong. That means designing the failure state first, deciding what the user is allowed to trust, and building an evaluation loop that tells you whether a change helped. Those are research and systems problems more than interface problems.

The second difference is throughput. Once a design system is legible to a model, a lot of production work stops being drawn by hand. I currently direct AI-driven design across a portfolio of more than fifty vertical SaaS products, and the leverage there isn't that the AI draws screens — it's that a well-specified system lets one person hold quality across a surface area that used to need a team. That changes what you should be hiring for. You need less production capacity than you did in 2023 and more judgment about what to specify.

Option one: the agency

Agencies are the right answer more often than founders want them to be. You get a team, a schedule, and someone accountable for delivery. If you need a full product surface designed against a launch date, this is what that's for.

Three names recur across published roundups often enough to be useful, and they cover the shape of the category. MetaLab, whose published work includes Slack's original interface and moving Midjourney off Discord onto the web, is the usual reference for consumer-scale polish. Punchcut positions on enterprise AI — agents, multimodal interfaces, autonomous systems — with twenty years behind it. Lazarev is research-led and prolific, reporting 500+ projects and 30+ shipped AI products. Engagements at this tier generally start in the six figures.

The failure mode is not quality, it's continuity. An agency engagement ends. If the thing you actually needed was for someone to hold the product's design coherent over eighteen months of model changes and roadmap churn, you'll buy the engagement twice.

Option two: the marketplace

Toptal, Arc.dev, Contra, and Upwork are where an individual designer is actually findable, and they're what the AI answers point at when someone asks for a person rather than a firm. Vetting is real on the first two and thin on the last. For a defined piece of work with a clear brief — an onboarding flow, a design system audit — this is efficient and cheap.

It's the wrong instrument for strategy. The marketplace model prices hours, so it quietly pushes you toward buying execution when the problem is that nobody has decided what the product should be.

Option three: the fractional lead

A fractional design leader is a senior person embedded part-time who owns the decisions rather than the output. Two or three days a week, indefinitely, setting direction and specifying the system that everyone else builds against.

This is what I do, so treat the following as interested testimony. It works when the constraint is judgment rather than capacity — you have engineers who can build and possibly an AI pipeline that can produce, and what's missing is someone to decide what should exist and what "good" means here. It also works when the honest answer is that you don't yet need a full-time design leader and hiring one would be a mistake you'd spend a year unwinding.

It's the wrong call in two situations, both common. If you need eight designers on the ground next quarter, hire an agency; a fractional lead is one person and cannot be scaled by asking nicely. And if your design problem is genuinely a volume problem — a large backlog of well-specified screens — you're paying senior rates for work a marketplace contractor should be doing.

Since almost nobody in this category publishes numbers, here are mine: a fractional lead engagement from $12k a month, advisory from $5k, a fixed-scope audit or sprint from $6k, and $350 an hour for anything genuinely ad-hoc.

Note what the retainer doesn't name: days. I stopped pricing time once it was clear that a well-specified system lets one person cover ground that used to need a team — billing by the day would mean charging less exactly as that leverage improves. What you're buying is ownership of the design direction, not a timesheet. It's also the honest version of the argument above: if I think hourly pricing pushes you toward buying execution when you needed a decision, I shouldn't sell you hours.

Choosing

The sequence that usually holds: if you cannot yet write down what the product should do, no amount of design capacity fixes that, and you want a senior person part-time. Once the direction is settled and the work is largely production, capacity is the constraint and an agency or a marketplace is cheaper. Hire in-house when design has become permanent and load-bearing — and accept that you'll wait two quarters for the right person.

If you're weighing the fractional route, here's how I work and what it costs, and I'm reachable at hello@product.inc. If what you actually need is a team against a date, one of the agencies above is the better call, and I'd rather say so now than three weeks into a conversation.