Disclaimer: The opinions represented here are those of the individual and do not necessarily represent those of their current or former employer.
AI is changing how creative and performance teams produce work, and it is happening fast. The biggest change is that we have a brand new way of working, where you brief an agent and wait for it to come back instead of clicking every button yourself. That shift rewards teams who can think clearly and brief well. It punishes teams who hope a clever prompt will save them. To unpack what this means for growth and creative teams, we sat down with Kate Lovejoy, Head of Customer Success at Poolday.ai, an AI video editing agent. Kate spends her days coaching hundreds of companies through this exact transition, and she has a clear view of what separates the teams that scale from the teams that stall.
Watch the full interview below, or read on for a selection of key takeaways.
Key Takeaways
|
The first change Kate sees is less about features and more about how people are doing the work. For years, using a tool meant doing the task yourself or setting up rules that ran the same way every time. Working with an agent breaks that pattern: you hand off the task, it goes away and does real work, then comes back with something for you to react to. The mechanics are easy. The adjustment is not, because you are giving up the click-by-click control you are used to and learning to trust a loop you cannot watch the whole way through.
For performance and creative teams, this reframes the job. Your value moves from making each asset by hand to directing the work and judging the result, which makes the brief and the taste behind it the real craft. McKinsey describes the same shift at the company level, with people moving from executing tasks to owning and steering outcomes as agents take on more of the work. It is also where creative volume comes from: a team that briefs well gets more shots on goal and more variations to test, without losing the human judgment that makes any of them good.
Letting an agent run unchecked is expensive in two ways. It burns tokens, and it burns something more precious: the hours your team spends fixing work that went the wrong direction. The trap is that an agent will commit fully to a wrong reading of your brief and hand back a polished, finished batch before you ever get the chance to course-correct.
"In this shift to turn-based working, your time becomes very precious, and those interventions become very precious. So what we coach customers through is designing gated experiences where you're not just letting the agent run. You're saying, show me the plan before you build it. Or, show me three creative directions and I'll choose the best one. You apply your taste and your craft and decide which direction to go before you proceed."
The fix is to build checkpoints into the workflow. Ask for a storyboard, a plan, or a few directions before the agent fans out into dozens of finished creatives. Each checkpoint is a cheap chance to catch a wrong turn before it gets expensive. For performance teams, this maps straight to cost control and ROAS: you spend compute and hours only on directions you have already approved.
For a while, the winning move with AI seemed to be a clever prompt: the right magic words, often copied from someone's viral thread, that would unlock a great output. Kate thinks that moment has passed.
"I think of right now as the death of prompt engineering. There was a time when folks were trading clever recipes, and it was supposed to be a magic recipe for an amazing output. We've passed that. The folks most successful with AI are able to think surgically about what am I giving the AI, what exactly do I want back, and what do I care about. The simple formula we give customers is: tell the system your inputs, outputs, and hard constraints."
Clear thinking beats clever wording. State plainly what you are feeding the system, what you want back, and the rules it cannot break. Kate is not alone in calling time on clever recipes: IEEE Spectrum went as far as declaring prompt engineering dead, citing research where models wrote better prompts than the humans hunting for the perfect phrase. She also warns about the opposite problem: bloated prompts. She spends a lot of time helping teams untangle long system prompts and .md files that have grown until they contradict themselves. When a prompt asks for snappy pacing in one place and slow transitions in another, the model has to guess, and people blame the AI for the mess they fed it. Often the best fix is to scrap the old file and rewrite it in clean, plain language.
The fastest way to get useless output is to ask for something vague. The strange part is that most people already know how to give a good brief: they do it every time they hand work to a teammate or an agency. But something changes when the partner is AI. The same person who would never tell a designer "make it pop" will type exactly that into a tool and expect magic. Kate sees this gap constantly, and she says it is one of the easiest things to fix once you notice it.
"You would never tell a creative agency just show me something cool, or I want an awesome 30-second video. And yet sometimes that's how people brief their AI. The creatives who are most successful set what really matters to them: here's the platform this will run on, here's my audience, here's my best-performing creative, and here are five creatives I love. That's what you'd give a creative agency if you wanted great results, and it's what you give AI for the best results."
A strong brief for AI carries the same things a strong brief for an agency carries: the channel, the audience, the constraints, and clear examples of what good looks like. Kate stresses that last point hardest. Her example: upload an hour of gameplay and ask for a 30-second highlight, and the system can guess at the best moments, but it will not match your taste until you show it what "best" means to you. This is the same discipline behind a strong creative brief, where a word like "professional" tells a team nothing until you show them an example. This is also why Poolday builds a shared creative system for each customer, with brand guidelines and system memory that hold steady across the company while each project layers its own rules on top. Define "best" once, and every brief after it gets sharper.
People love the idea of one perfect prompt that solves everything. Kate pushes teams past it toward something more durable: a system that gets smarter every time a human corrects it.
Treat every correction as training, not a one-off fix. Capture it in shared memory, brand guidelines, or a reusable workflow so the whole team inherits it. Over time this turns scattered prompts into repeatable processes and a shared voice, instead of every person solving the same problem from scratch. For teams running high creative volume, this is how quality compounds while cost stays flat.
The teams that get all of this right share a culture, not just a toolset. Kate sees two traits in the teams that adopt AI well, and they need each other: psychological safety, where people experiment and play and push the system to see where it can go, and a structured set of KPIs they are trying to hit. You need both. Safety means people can say what failed, not just show the polished result. Her advice is to build in public with your team, share the experiments and the misses, and keep the robots on robot work so humans can focus on taste and vision. In a field moving this fast, the teams learning out loud are outpacing the ones guarding their secrets. You can follow Kate Lovejoy on LinkedIn and see what she is building at Poolday.ai, where the team just shipped custom apps for interacting with their video editing agent.
Looking to drive more performance from your campaigns? Adikteev helps app marketers scale with advanced retargeting and incrementality-driven solutions. Get in touch to learn more.