Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Angle
A Product Owner position placed between whoever designs a product and the team that ships it is judged on the line it installs, not on the quality of the work done there: as soon as thinking the product and carrying it through to production live in two heads, the person designing no longer sees the arbitrations that decide what the customer will meet, and the person making them has not been given the mandate to own them. The three supports of that separation give way without any need to invoke artificial intelligence: the truth of a feature is what runs in production and not what was specified; the backlog is a consumable medium, which can ground a function but not a profession; the title distributes the mandate ahead of any negotiation, so that the word chosen decides what its holder will be able to settle. Automation does not make this argument true — it collapses the cost of the plumbing that paid for the position, and removes the last practical reason to go on dividing.
Synthesis
Taken idea by idea, the material looks like a quarrel of vocabulary between two job titles. Put end to end, it describes a method for examining interface roles, applicable well beyond product.
The starting point is a fact of manufacture: between the specification and the release, compromises, edge cases and technical constraints redefine the shipped behaviour, and those arbitrations do not find their way back into the initial document. There is nothing polemical about this fact — it simply makes untenable the position of whoever stops at the intention. They are not working on the product, they are working on a description that expired at the first iteration.
The separation is then paid for twice. On the design side, it prevents the loop from closing: nobody comes back to tell the designer where their wording was ambiguous, and the cost of their approximations is paid by the team that discovers them. On the decision side, it dissociates the choice from its consequence, exactly as a soft decision dilutes responsibility across meetings and documents — except that here the dilution is written into the org chart, and therefore permanent.
What remains are the justifications, and each falls for a different reason. The backlog justification falls on the lifespan of its object: a medium consumed on delivery can keep someone busy, it lays down nothing on which an expertise accumulates. The vocabulary justification falls on its own principle: if the title did not count, it would not have served to partition responsibility without granting the power to arbitrate. The domain expertise justification falls on observation: knowledge of the domain is found in support, among Customer Success Managers, among long-serving developers, and it can be acquired.
Automation only enters last, and through economics. The division had a quantifiable motive: translating, documenting and formatting cost enough to warrant paying for a dedicated position. That cost collapses, and a division of labour deprived of its motive holds up only through habit. Two reservations prevent drawing too broad a conclusion: what collapses is a volume ceiling, not a decision bottleneck — no arbitration becomes any easier; and what no system can be delegated is answering for a bet in front of other people, six months later.
What the whole thing brings into view: the question put to an interface role is not "does it do it well?" but "what does it separate, and who pays for that separation?". It admits an answer other than abolition — a team working on regulated healthcare software kept the position and redistributed its content — and it applies first of all to whoever asks the question, since they are often the beneficiary of the division.
Tensions / contradictions
The main tension bears on the title. The angle holds that the job title manufactures the mandate; several practitioners answer that the mandate alone counts, and that a Product Manager locked into operations, with no strategic latitude, will be busy without producing anything useful. The two positions are not settled here: the title steers the attribution of the mandate without guaranteeing it, and insisting on vocabulary can mask the fact that the mandate is the real subject.
A second tension, about the way out. Removing the transmission belt and redistributing its content are not the same gesture, and the second is the only practicable one in a regulated domain with a legacy codebase. Nothing makes it possible to say in advance which of the two routes holds in a given organization.
A third tension, more awkward: whoever disputes the division is also the one who benefits from it, and the argument they carry must be examined with the same wariness as the one they are fighting.
Questions
- What criterion distinguishes, in a given team, a distribution of product work from a separation between design and execution?
- How do you verify that an arbitration mandate has actually followed a job title, other than by observing after the fact that the arbitrations were not held?
- Up to what size of organization can the same person design and accompany through to production, before the loop necessarily has to run through formal devices?