In 2015 I was running the version of the agency that everyone is now predicting the death of.
Thirty people in Amsterdam. Seniors selling the work and holding the standard, mediums running the projects, juniors producing the volume. The margin lived in one specific place: the gap between what a junior cost us and what a junior billed out at. I did not call that structure a pyramid at the time. I called it the business, because it was the business, mine and every agency's I knew.
Predictions about this industry tend to be written so they can never be wrong. Directional, undated, hedged into fog. I want to do the opposite, because the vague versions are worse than useless: they feel like insight and commit to nothing. So these five are dated, specific, and written to be graded. Screenshot this page. In 2029, one of us gets to say I told you so, and I accept the risk that it is you.
Five predictions, written to be graded
One. Half the staff, double the output. By 2029, a strong independent agency at two million in revenue runs on twelve to fifteen people and produces more than the thirty-person version of itself did in 2024. Not the same output with fewer salaries: more output, better margins, smaller building. Revenue per head at well-run independents lands at three to four times the 2024 norm. The layer that thins is the one my own margin was built on, and the thirty-person shop at two million reads, by 2029, the way a seventy-person shop would have read in 2019: as a sign something has not been examined.
Two. Every serious agency runs on an agency brain. This is the prediction I hold with the most conviction and the most bias, in that order. By 2029, the center of a well-run agency is not the founder's head and not a project management tool. It is an expert system that holds what the founder's head holds today: the standards, the past decisions, the client history, the way this particular shop judges work. New staff learn from it. Delivery gets checked against it. The founder finally goes on holiday because the thing that made the agency the agency no longer commutes in one skull. Peter and I are building toward exactly this future, which means I am the man selling shovels while predicting gold. Weigh my confidence accordingly. Then notice that founder dependence has resisted every fix the coaching industry threw at it for twenty years, and ask yourself why this one would be different. My answer: every previous fix asked the founder to write things down for humans. This one makes the written-down things operational.
Three. SaaS goes homemade. The subscription stack shrinks, hard. Agencies stop paying for a tool with two hundred features to use nine of them, because building an internal tool with one main function takes days now, fits the workflow exactly, and costs nothing per seat. Single-function never means only one function: it means one main job, done the way this specific business works. By 2029 the default question flips from "which tool should we buy" to "why would we buy this instead of building it Tuesday". I say this as someone with SaaS products in the market, so this is the second prediction where I am betting against my own book, and the only defense I can offer is that I run my own operation on homemade tools already: the content calendar I publish from, the bookmark brain I mine, the measurement systems I check monthly. The prediction is just my practice with a date on it.
Four. The hour dies twice. Client-side first: production work, the layer AI accelerated, gets priced per outcome by 2029, and hours survive only at the top, where the client is buying judgment and attention rather than output. Labor-side second, and this is the part the pricing conversation keeps missing: once output stops mapping to time, the 40-hour, five-day container stops being the default way agency work is organized. Capacity gets planned around judgment availability, not seat time. An agency still quoting production in hours in 2029 is announcing a discount, and an agency still measuring its people in hours is measuring the wrong thing with confidence.
Five. Nobody pays for medior again. Used well, AI produces competent, medior-level work in nearly every discipline, on demand, at close to no cost. That becomes the floor, and everything at the floor becomes free, which is a polite way of saying worthless as a differentiator. The market splits at that line. Below it: infinite competent sameness. Above it: distinctive work, real judgment, a recognizable point of view, and the seniors who can produce those, made scarce by the same industry that dismantled its own apprenticeship to save money at the bottom. The panic of 2029 is a senior panic, and the premium of 2029 is a uniqueness premium. Quality was never positioning, as I learned expensively in my own agency. By 2029, quality is not even scarce. Being unmistakably yourselves is the last position left.
The shape underneath
Put the five together and the replacement for the pyramid is visible: a small senior core doing distinctive work, an agency brain where the middle layer was, homemade tools where the subscription stack was, outcomes where the timesheet was. Less a pyramid than a keel. Narrow, heavy in a different place, built for stability instead of stacking.
I hold these five with different confidence, and I have flagged the two where my bias runs strongest: I am building the brain and I am selling the SaaS the third prediction undermines. I would rather name both than pretend the view from nowhere. The one I consider closest to inevitable is the death of the hour, because it is already happening in every pricing conversation I hear. The one that will look most wrong or most right is the medior line, because it makes a claim about what AI will not do, and those age fastest.
What I keep coming back to is that the pyramid was never the point of the agency. It was just the container the margin lived in, so long and so comfortably that we mistook the container for the business. The margin is moving house. Everything above is one operator's guess about the new address, written down where it can be checked, because the industry has heard enough predictions that dissolve on contact with a calendar.
2029 is a Saturday away, 150 times over. See you there.