The real product in the AI era

When offset printing democratised typographic production in the eighties, many predicted the end of graphic agencies. When desktop publishing put Photoshop^® and Illustrator^® in the hands of anyone with a computer, the prediction was repeated. When the internet made templates, stock photography and website builders available to all, the talk of disintermediation in professional communication returned once more.

The agencies did not disappear. The market redefined itself, services evolved, but the figure of the professional who transforms a communication objective into a measurable result maintained its position. Not because the new tools did not work, but because the competence did not reside in the tools.

Generative artificial intelligence poses the same question, with greater intensity. And the answer follows the same logic.

The Commoditisation of Output

If anyone can generate a convincing image with a text prompt, the value of that image as an isolated product tends towards zero. Not in the sense that no competence is required to obtain it — it certainly is — but in the sense that the technical barrier to access has been lowered in a radical and structural way.

The same applies to texts, videos, synthetic voices, graphic layouts. Generative AI has shifted the threshold of access to content production in a way that is structural, not transitory.

In a market where output is commoditised, value inevitably shifts to what the output alone cannot guarantee: the strategy that preceded it, the context in which it is placed, and the **responsibility** associated with its use.

In a market where output is commoditised, value inevitably shifts to what the output alone cannot guarantee: the strategy that preceded it, the context in which it is placed, and the responsibility associated with its use.

If anyone can generate a convincing image with a text prompt, the value of that image as an isolated product tends towards zero.

The Invisible Package

When a company engages a communications professional to produce materials intended for the market, it purchases something that goes well beyond the file delivered at the end of the project.

It purchases the certainty that the images used are correctly licensed for the intended use. It purchases the verification that the fonts employed do not violate distribution terms. It purchases the guarantee that the photographic content — whether depicting people, places or objects — does not expose the company to disputes over unauthorised use of likeness. It purchases the responsibility of someone who stands behind the choices made.

This “Invisible Package” — composed of verifications, basic legal competence, licence knowledge and assumption of responsibility — has always been an integral part of professional service, even when it was not itemised as a separate line in a quote.

In the AI era, this package becomes even more relevant, because the variables to manage multiply.

Why a company chooses to pay for an official API

The question is legitimate: if open weight models are freely downloadable, with performance often comparable to commercial services, why should a company pay for a subscription or API access to an official platform?

The reasons are various, and not all concern the technical quality of the model.

• Traceability of Rights

  Commercial platforms document, at least partially, the training conditions of their models and the licences on outputs. In a legal context, this documentation can make the difference.

• Indemnification Guarantee

  Some platforms — as in the already cited cases of Getty or Adobe Firefly — offer explicit coverage for commercial use of outputs. A model downloaded and installed locally offers none of these guarantees.

• Service Stability

  A commercial API guarantees continuity, updates, technical support and defined SLAs. Managing a local model internally requires infrastructure, expertise and ongoing maintenance.

• Data Separation

  Enterprise versions of the main platforms offer guarantees on the isolation of user data. A company uploading confidential documents — briefs, strategies, materials not yet public — needs to know precisely what happens to those inputs.

In short: the cost of the API does not pay only for computational capacity. It pays for a set of guarantees that a free model does not include.

Risk Transfer as a Service

There is a more direct way to describe what a communications professional sells in the AI era: Risk Transfer.

The client does not want to concern itself with verifying whether an AI model’s licence permits commercial use of the outputs. It does not want to assess whether a generated image might infringe third-party rights. It does not want to wonder whether content uploaded as a prompt will remain confidential or end up in a training dataset.

It wants materials that work, that can be used without risk, and someone who is accountable if something goes wrong.

This is precisely the model that made the communications professional indispensable when the tools were analogue, and that continues to make them indispensable now that the tools are generative. The technology changes; the logic of the service does not.

What remains of Creative Work

There is an underlying question, one that many ask without formulating explicitly: if the AI generates the output, what does the professional produce?

The answer is the same as always: the professional produces the decisions. Which image, in which context, with which message, for which audience, with which rights, on which channels, with which tone. The generation of the file has become faster — as printing did with offset, and retouching with Photoshop.

But the work that precedes and follows the generation of the file remains human, contextual, strategic.

Generative AI has not eliminated the need for those who know what to produce. It has only made less visible, to those outside the industry, the distance between knowing how to use a tool and knowing how to build effective communication.