D-015Infrakit[?] open
date:
org:
Infrakit · Head of product team
category:
Team design
title:

Organise the product team around customer jobs, not around who builds

context:
AI made building fast. Scoping and understanding customers did not get faster with it. Developers waited on coordination for small UX calls, and roadmap trade-offs tended to go to whoever pushed hardest.
decision:
Proposed a team model where developers own small UX through weekly dogfooding and monthly customer contact, dedicated project coordination is kept only for rehauls and new products, and every customer job sits in a state matrix with a fixed number of coordination slots.
hypothesis:
Leadership makes its trade-offs openly against the matrix. Small UX ships without waiting for anyone. Rehauls and new products actually finish because coordination is not spread thin. Recurring bug clusters stop coming back once they are treated as rehauls.
confidence:
75%
review_date:
reversibility:
one-way
related:
D-016
tags:
proposal, team-design, ai-era, prioritisation, customer-understanding
/decisions/d-015.mdwrite-up: below

Building got cheap fast. A developer with AI tools can turn a clear idea into working code in a fraction of the time it took a year ago. What did not speed up was knowing what to build. Scoping, understanding the customer’s actual job and deciding what not to do became the bottleneck. Two symptoms showed it. Developers waited on someone else for small UX decisions they could have made themselves. And bigger trade-offs were often settled by the loudest voice in the room, not by which customer problem mattered most.

So the proposal moves decisions to where the knowledge is. Developers own small UX. They earn that by using the product themselves every week and talking to customers every month. Dedicated coordination is kept for the work that really needs it, which is rehauls and new products. Everyone’s time is split visibly, so it is clear how much goes to building, to bugs and to customer contact.

The state matrix

Rows are customer jobs, not product modules. A module view hides which job is hurting. Each row is in one of four states: backlog, iterative, rehaul or new product. Iterative work is handled by developers through dogfooding. Rehauls and new products get a coordination slot, and there is a fixed number of those slots. A job only moves up if another moves down. That turns the leadership conversation from a wish list into a trade-off.

Two more rules keep it honest. Only signals from customers that match our ideal customer profile count. Everything else is filtered out by default. And bugs are read by pattern. If bugs keep coming from the same root cause, that job becomes a rehaul candidate. If they are scattered, the answer is more bug capacity, not a rehaul.

Alternatives considered

Cost and trade-off

Developers spend real time on dogfooding and customer contact that used to go to building. Some customer jobs will sit in backlog with nobody driving them, on purpose. Some vocal customers outside our target profile will hear no more often. Small UX decided by many people will be less consistent than one person deciding. I think those costs are cheaper than the current one, which is building the wrong thing fast.

Status

Proposed Nov 2026, not yet decided or rolled out. Leadership reviews it first, then the team. I call it one-way because once people’s roles and expectations change, walking it back costs trust even if the structure itself can be undone.

What would change my mind

If developers skip the dogfooding and customer contact once delivery pressure rises, the model falls back to the old one with less coordination. Same if leadership keeps adding rows to rehaul without taking one off. I will look at both before the review date.

Lessons so far

Writing the matrix down made the trade-offs uncomfortable in a useful way. It is easy to agree that we focus. It is harder when the slot count is visible and something you care about lands in backlog.

What I can’t share

Which customer jobs sit in which state, the number of slots, the capacity split per person, who is on the team, and the internal discussion around the approval. Anyone inside the company will recognise this as our team. That is unavoidable for a card about my own team.

<- Back to the log