THE AI YOUR AGENCY WAS MEANT TO BUILD. BUT HASN'T.

among is an operating system for advertising agencies, built by people who spent twenty years inside them. It took years to learn how to write down the work: the order it happens in, the questions senior people ask without noticing, and the reasons ideas get killed. And we tested every piece of it on AI to see where it broke, which turned out to be most pieces, most of the time. What follows is what we learned. Some of it is about AI, but more of it is about agencies, which had problems long before AI turned up. Read it in order, or skip to whichever one annoys you most.

Polish used to be evidence that someone had thought hard, because finishing was expensive. Now a junior’s first draft arrives with the same structure, typography and confidence as a senior’s final, and the competent and the confidently wrong look identical until someone sits down and reads closely.

AI learned advertising from what the industry publishes, and the industry only publishes its endings: the case study, the awards entry, the LinkedIn post written months later. Nobody publishes the note a senior gives a junior, the insight routes ruled out before lunch, or the read that this client is scared and that one wants to be famous. What is online is hundreds of competing methods, so AI blends them until it has something to give you. The result looks like strategy and sounds rigorous, but it doesn’t follow any single line of reasoning from start to finish.

AI was meant to take the admin and leave people the thinking, and somehow managed the reverse. It writes the copy, makes the images and drafts the strategy, while the people it was meant to free are still chasing approvals and hunting for the file called FINAL_v7_actual. The most powerful tool we’ve ever built ended up pointed at the half of the job people came into advertising for.

Agencies sell outcomes and bill for inputs, and AI shrinks the inputs. Every hour the machine saves is an hour the agency can no longer charge for, and the client has already asked for the discount. So hours can’t be the product any more. What clients will keep paying for is judgement, the part AI can’t compress.

Most agencies can only grow one way: win more work, then hire more people to do it. Revenue and headcount climb together and the margin stays where it was, because every new fee arrives with a salary attached. That’s how a 150-person agency can be busier than ever and still be in trouble. The way out is to let senior judgement reach more of the work without needing more seniors, so revenue can grow faster than the team.

The industry’s big bet on AI is agents: give one a goal and let it work through the steps on its own. But an agent treats the steps as optional. If it can see a quicker route to something that looks like the goal, it takes it. In strategy that’s a real problem, because working through the steps is how you find out the goal was the wrong one.

An agent with latitude behaves like a brilliant junior. It will rewrite the brief because it thinks it found a better answer, and tell the client yes because yes feels good. It has all the brains and none of the experience, and the difference is that a junior learns while the agent is the same first-year every Monday morning. That’s why the promised future has people managing agents instead of people. But you only have to manage something that was never put on a rail. Give it a narrow task, a fixed sequence and a gate, and there’s nothing left to babysit.

You can read every book on driving and still stall on your first hill start. The instructor who can explain the bite point learned to explain it by sitting next to hundreds of learners who got it wrong. Agency judgement works the same way. It isn’t in any textbook, and the people who have it can’t put it into words until they watch someone miss it. We wrote among by watching AI miss it, again and again, and writing down what was missing each time. Every fix cut the mistakes down and uncovered smaller ones underneath. Measure a coastline with a one-metre stick instead of a one-kilometre one and it gets longer, because now you’re counting every rock. The gaps work the same way: they never run out, but each one matters less than the last.

among works like a team you’ve hired rather than a tool you’ve bought. Karen takes the brief, Gillian the strategy, Don the creative, Alfred keeps the work moving and Marty watches the money. Each has one job and one part of the method. They do the early legwork, put their thinking on the table and push back when something doesn’t hold up. People bring their own thinking in at every stage, and the work ends up further along, and further from the obvious, than either would have got to alone.

Most problems start in the brief, so Karen reads it the way a senior strategist would. She pulls out the objectives, checks they don’t quietly contradict each other, matches the job to one of 32 brief types, attaches KPIs that can actually be measured, and asks the questions nobody wanted to ask in week one. Gillian does the research before anyone writes a word, marking what’s fact, what’s only a signal and what’s still an assumption. Then she drafts and scores insights and builds brand and platform thinking from them, and she won’t pass an observation off as an insight. She runs on frameworks and rubrics from Gillian Rightford, and is named after her. Don turns the strategy into springboards that each open a different creative world, and develops ideas from safe to interesting to powerful, scored against the brief and for originality. At every stage the team works alongside them: adding their own thinking, asking for more, arguing with the scores, or taking it somewhere none of them went.

Execution is the most crowded part of the AI stack, and every agency already has favourites there. Meanwhile, the thinking that decides whether any of it is worth making happens in a general-purpose chatbot. among ends at the big idea and the creative direction, then hands over to whichever tools the agency already likes.

among is built by a sister and brother who came at the same problem from opposite ends. Genevieve Jansen van Vuuren spent twenty years as a strategist across research agencies, ad agencies and tech consultancies, watching good work depend on judgement nobody ever wrote down. Garreth van Vuuren built an integrated agency from nothing to R250m and 150 people in five years, then watched the agency model pull it apart. She knows how good work gets made. He knows what it costs an agency to make it, and how quickly that cost outgrows the business. among is the system he needed and didn’t have, built from the method she’d spent her career carrying around in her head.

The lessons behind the method took years, and most of what we tried first was thrown out. The method itself is written down one module at a time, and the build started in August 2026 with our development partner in London. The first agencies see it this October, timed for budget season, and the full launch is planned for January 2027. Until then we’re still writing things down, because every test turns up something we missed.