There is a particular pitch making the rounds right now: hire the agency, get a self-service AI tool. Log in, type a prompt, out comes the content, the campaign, the report. The agency takes a margin on the platform and calls it transformation. It demos beautifully. It also misunderstands where the value in this work actually lives.
Strip the branding off most of these offerings and the same thing is underneath: a login to GPT or Claude with a logo on top and a markup on the invoice. The client could buy the raw tool for a fraction of the price. What they are actually paying the agency for, in theory, is the wrapper, a slightly nicer prompt box and the comfort of a familiar vendor. That is not a service. It is arbitrage on a price the client will eventually discover for themselves.
The reason it falls apart is that the hard part of this work was never producing the output. AI is fast, fluent, and tireless, and it will confidently generate something that looks on-strategy while being quietly off-strategy in ways a non-expert cannot see. It has no stakes in the outcome, no memory of what failed last quarter, and no judgment about which of the ten plausible directions is the one that actually moves the number. Hand a self-service tool to a client and you have not given them leverage. You have handed them a faster way to produce confident, mediocre work.
That is the divide. On one side, AI replaces the thinking: prompts go in, volume comes out, and nobody senior is accountable for whether any of it serves the strategy. On the other side, AI executes under human judgment: an experienced operator decides what to build, why, and in what order, then uses the model to get there ten times faster. The inputs look identical. The results are not even in the same category. Speed amplifies whatever direction you are already pointed, which means handing a fast tool to a weak strategy just gets you to the wrong place sooner.
Our thesis at Principal Harbor is that AI belongs in the second column, never the first. We work alongside our clients to deliver fast results with AI, not to replace good strategy with bad strategy dressed up as automation. The model drafts, mines, tests, and ships at a pace no team could match by hand, but the calls that determine whether the work matters, what to prioritize, what to cut, when the output is confidently wrong, stay with people who have skin in the outcome and the experience to recognize the difference.
In practice that means you get the velocity everyone is promising and the oversight almost nobody is including. Faster does not have to mean worse, and strategic does not have to mean slow. The agencies reselling a wrapper are betting that clients cannot tell the difference between output and outcomes. We are betting the opposite, that the gap shows up in the results, and that it is the entire job.
