# The Product Trio Doesn't Need to Exist Anymore

Published: 2026-07-23
Canonical: https://polyviam.com/writings/product-trio-doesnt-need-to-exist-anymore/
Excerpt: AI turns three deep specialists into one wide operator, backed by specialists who polish the last 20%. Here's the Initiative-Owner model, and why the trio doesn't survive it.

The more I use AI, and the more I observe teams using it, the more convinced I am that the product trio is dead. And if you're an org building software in this way then you are behind.

The trio was the default unit for building software. It assumed people were T-shaped: deep in one discipline, passable in the others. The PM couldn't design. The designer couldn't ship code. The engineer couldn't do discovery. So you put three deep spikes in a room and made them collaborate.

AI dissolves that paradigm. It makes people U-shaped - competent at everything (assuming reasonable competence with AI). Not perfect, of course. Roughly the average of the parameters it is given. A shallow U, perhaps, but a U nonetheless.

A competent operator with AI now produces a decent spec, a decent design, decent code, with decent copy. Breadth has stopped being scarce. Unless you operate within the top ~quartile of your discipline, your depth has been repriced down to token-cost.

Here's how that looks visually. The Ts in the product trio model overlap to achieve the necessary depth and breadth. Now, they are replaced by one U. Underneath the U are now the Ts, or even more extreme - Is (true specialists).

![Two diagrams. Fig. 1: three overlapping T-shaped beams labelled PM, Designer, Engineer, showing the trio](/img/writings/product-trio-doesnt-need-to-exist-anymore-shape.svg)

_Fig. 1—2: the product trio's three overlapping Ts, replaced by one U carried to an 8, finished by I-shaped specialists._

The consequence for teams is blunt. If anyone competent with AI can produce roughly 8-out-of-10 work across the whole lifecycle - spec, design, copy, code - on their own, does this paradigm hold up?

I've now run an AI-transformation assessment with several client teams, and we inevitably end up circling the same op model discussion. I call it the Initiative-Owner (IO) model.

An IO takes one initiative end-to-end: spec, design, code. User research, copywriting, mockups, prototypes, front-end, back-end architecture, documentation, market research - all of it, one person.

They're supported by specialists: designers (UX, UI, motion, brand), engineers (front-end, back-end, devops, secops), marketers (organic, paid, copy, brand).

The strength of the model is that handoffs collapse. Context stays in one head. The product is coherent because one person holds the goal, the strategy, and the execution at the same time. Nothing is lost in translation between a spec author, a design author, and an implementation author, because they're the same person.

The IO isn't expected to ship (although maybe they can in certain circumstances). They take an initiative from idea to "as close as possible to shipped". This happens when AI can no longer move it forward; when the next increment needs judgment or skill the model doesn't have. Then the specialists come in and take it from an 8 to a 10. Motion polish. UI tuning. Copy edits. Back-end re-architecture.

Specialists have a second job, and it's actually the higher-value one: they build the assets and internal infrastructure that make IOs faster and better. Design systems. Coding standards written into repo docs. MCPs. Analytics access. Documentation. Skills. The better these are, the closer an IO's end output gets to a 10, and the less polish it needs at the end.

The obvious objection is that everything now lives in one head. (That's the point.) You manage that the way you always did: documentation, review, the specialists who touch the work. But you don't reintroduce three-way handoffs to fix it. The handoffs cost more than the risk.

Not everyone can be an IO. Three traits are non-negotiable, and they're mostly innate:

1. **Agency.** The most important one. The ability and the appetite to move something forward without being told to.
2. **Taste.** Without it, an IO can't tell good from mediocre, which means they can't tell when they've hit the ceiling of AI, because what they've actually hit is the ceiling of their own taste. Taste is the stopping function.
3. **AI-interest.** Genuine, restless curiosity about what the tools can do and how to get more from them. Tolerance isn't enough.

Four more traits matter, but less, because they can be developed. Roughly in order, they are:

4. Structured thinking
5. Resilience
6. Learning by doing
7. Domain fluency

Most organisations will keep the trio and bolt AI onto it. That's the mistake. The trio was a workaround for a constraint: that no one person could be good enough at everything. Now, the constraint is gone. When the reason for a structure disappears, the structure should go with it.