Every marketing agency now has an answer for how AI makes its team faster: less time on production, a prototype built in an afternoon.
Fair enough. But that tells you surprisingly little about whether the work is getting any better.
A more useful question is whether your agency understands your business better than it did six months ago, and what it's doing with that understanding. It's the conversation Major Tom's president Lyn Bryan, web and creative director Darren Maher and innovation architect Olu Osunrinde have in the first episode of our podcast.
Jump to the conversation
The advantage AI gives a good agency is understanding: how much it knows about your business, and what it does with that knowledge. That depends on context the models don't come with, like why your past decisions were made, including the ones that didn't work. Most of it walks out the door when someone leaves either side of the relationship. An agency that keeps it can bring you answers to problems you never briefed, and show you a working prototype before you commit budget to building anything.
Every team can use the same AI models, but each one feeds them different context. An AI that knows your business and the decisions already made will give better answers than one handed a folder of documents with no sense of how they connect or why they exist.
Olu Osunrinde, Major Tom's innovation architect, has the best way of picturing it. Take the brightest intern from MIT and put them in a mechanic's workshop. Ask them to fix a tire. They know how in theory. What they don't have is any idea how this particular shop does it.
"You can have the most intelligent model, but it doesn't mean you're going to get everything done the best way." — Olu Osunrinde, Innovation Architect, Major Tom
A team-wide AI subscription gets you the intern. Giving the intern more to read doesn't fix it either. Upload a gigabyte of PDFs and start asking questions, Olu says, and you get "garbage in, garbage out", because a document can't tell the AI what led to it or what happened next. More output without that context is how marketing gets easier to run and harder to explain. Olu tells the intern story at 13:21.
Darren Maher, who leads web and creative at Major Tom, tells the story of a developer who picked up an existing build after the original developer had left. The new developer could see exactly what was in front of them:
"I know what is built. I know how it was built. I have zero context on why it was built that way." — Darren Maher, Director, Web & Creative, Major Tom
The person on the client side who had made that call had moved on too. Everyone assumed the feature was there for a reason. Nobody could say what it was.
Your team almost certainly has a version of this: a campaign rule or a page on the site that someone decided on three years ago and nobody can now explain. That knowledge of why past decisions were made is your institutional memory, and most of it lives in people's heads.
"I don't mean it in a disparaging sense. Organizations have the institutional memory of goldfish." — Lyn Bryan, President, Major Tom
Sometimes people haven't even left, Lyn adds. They've been promoted well beyond the decisions they used to make.
The records most teams keep don't fill the gap. Olu points to the log a project manager keeps of decisions and risks: "Why did we decide against that? The RAID log doesn't cover that." Meeting notes are worse. Lyn says the team has come to rely on transcripts, because typed notes were "completely full of their own bias and what they decided to hear in that moment." (Think back to your own meeting notes. Is she wrong?) It's a big part of why marketing gets so hard to explain to anyone who wasn't in the room. The goldfish moment is at 19:40.
It looks like one shared record of the whole relationship, from the first sales conversation onward, that the agency team and your team can ask questions of. At Major Tom that record is the Client Growth System, and it keeps the reasons behind decisions alongside the decisions themselves.
It draws on the tool Major Tom's teams do their day-to-day work in, so the reasoning behind a recommendation, including the options that were debated and dropped, is kept alongside the recommendation itself. Whoever picks up the work next starts with that understanding. Olu describes the aim as working on "one side" with clients, rather than an agency side and a client side. For you, that changes three things.
A new person on your team starts with the history. If the person who signed the agency up is on parental leave by the time the work starts, Olu says, the agency can now be the one briefing their replacement, with everything they missed since the first sales call.
You can look at the losses as easily as the wins. You probably see your agency's view of the work as a deck every few months, and there's only so much a slide can hold. A shared record lets your team ask about the work directly. In Lyn's words, you can "query the wins and query the losses."
It remembers what didn't work. A project management tool will keep the data, Darren says, but it "isn't going to tell you, 'by the way, you tried that five years ago and the results weren't so hot.'" A record built around decisions can flag it before anyone repeats it. Lyn is looking forward to the reverse as well: tracing a great result for a client back to the conversation, two and a half years earlier, where the idea started.
It helps while everyone's still in the room, too: with everything in one place, Darren says, "you can start seeing the disparities between what one team is doing versus what another team is doing."
The fair objection is that a system only knows what you choose to put in it. Lyn's answer is to put everything in and have people check it: "We put in all of the data and we use human insight to sift through pieces that are valid and not." A joke in a meeting transcript isn't a decision, and someone has to know the difference. A record of the highlights would just be a longer deck, and it wouldn't hold up when someone examines the work closely. Olu explains how the system is put together at 20:23.
It should bring you answers to problems you haven't briefed. AI now handles a large share of first drafts and production, and the time that frees up is only worth something if it goes into listening harder to your business and spotting what you haven't asked for.
Darren puts the gain at a head start of 60 or 70 percent on a lot of deliverables. The part he cares about is where the rest of the time goes:
"What excites me more than anything is what you do with that 60, 70 percent of your time that you didn't have before. We're more open to listening more intently to clients and trying to solve problems they're not coming to us with." — Darren Maher, Director, Web & Creative, Major Tom
His own example is the point in a website project when the new site needs its content. Many clients don't have a copywriter on the team, so the writing lands on whoever is in marketing, whatever their background. Darren's team can now build a pipeline for it, where AI writes briefs from the brand guidelines and voice, and researches competitors to find gaps worth writing about. "Your website doesn't stop once it launches," he says, so the same pipeline keeps suggesting content long after the launch work is done. Nobody briefed it, but it solves a problem the client would have hit soon after launch.
It's also why Lyn calls planning agency work in hours "a ridiculous metric". Two people can spend the same hours and produce completely different work. The better starting point is the business problem and what it takes to solve it. Darren makes the point at 05:29.
A prototype lets you judge an idea by using it, before you've committed budget to building it properly. A proposal asks you to picture the result from a description. A working version shows you whether it solves your problem and what it would take to finish, while changing course still costs very little.
This is where retained context and AI speed meet. Olu's example: two years ago, a team suggests an idea and the client says they don't have the data to support it yet. The record keeps that conversation. When the relationship moves on and the client now has that data, the system surfaces the old conversation and a person on the team decides it's worth acting on. They don't send a proposal. They come back with something that works.
"You're no longer selling your client a dream. You're showing them the experience." — Olu Osunrinde, Innovation Architect, Major Tom
Building that version used to mean a queue. Darren remembers knocking on developers' doors to ask how long something would take; now his team can have a prototype running in less than a day, often without developer hours at all. "People are tangible," Lyn says. "They want to be able to touch and feel it."
What changes for you is the order of decisions. You see the idea working with your own content and your own problem before anyone writes a budget line for it. The people in your team who'd have argued over a description can react to the same thing on a screen. And a no is cheap. "If that prototype doesn't work, you can move on to the next one," Lyn says. "You haven't invested a year in building it."
A prototype is still a test, though. Darren is clear that developers come in afterwards to check it properly before anything goes live. The conversation on prototypes starts at 36:24.
Ask how much of your business it understands and what it does with that understanding. A good partner should be able to explain the reasons behind your past decisions, point to ideas it raised that you never briefed, and show you working versions of its best ones.
These five questions come out of the conversation. Take them into your next agency review.
| Ask your agency | A good answer sounds like |
|---|---|
| What do you do with the time AI saves you? | Specific problems they've raised that you never briefed. |
| If our main contact left tomorrow, how would the next person learn why we made the calls we made? | A shared record of decisions and the reasons behind them, rather than "they'd read the handover document". |
| Can we see what didn't work, as well as what did? | Yes, and they can show you where it's recorded. |
| When you have an idea, do we see a prototype or a proposal? | A working version wherever it's feasible to build one. |
| Who checks what the AI produces before it reaches us? | A named person who reviews it, and knows a joke in a transcript isn't a decision. |
The best question isn't how quickly your agency can use AI. It's what your agency now knows, notices, and does that it couldn't before.
If you'd like to see what those answers look like on a real account, here's how we work with marketing teams.
AI is taking over a growing share of production work, like first drafts and research summaries, but it isn't a substitute for judgment about your particular business. That shifts what an agency is worth to you toward what it knows about you and what it does with that knowledge. When you review a partner, ask how its team and its costs have changed since it started using AI, and where the time has gone.
Olu sees three common causes. Projects get mandated from the top without asking the people who do the work what slows them down. They skip the foundation work of turning documents into something a machine can search by meaning. And costs grow faster than planned. Major Tom started by asking every team what got in their way, then ranked the list with RICE: reach, impact and confidence, divided by effort.
It varies by project, but the costs teams most often underestimate are the foundation work of organizing their data and the volume of use. On the API you pay for what the model reads and what it writes, which Anthropic's pricing lists as separate charges, so sending your whole archive with every request adds up. Olu's approach is to build a focused brief from internal records first. If you copy it, check usage limits, data terms and pricing, which differ between API and subscription plans.
No. "You need to just make a start," as Darren puts it, "rather than trying to get everything perfect." Waiting for perfect data is one of the most common reasons AI work never begins, and sorting out half-organized material is something AI handles well. Get your material into one place, start on one narrow problem, and let that first project show you which gaps actually matter.
It can, when everyone feeds the same models the same generic prompts. Lyn's view is that different context produces different ideas: "We're talking about bringing in the entire context that is unique to a situation to then get different ideas." Ask your agency what it gives the AI about your business before it asks for ideas. If the answer is "the brief", expect ideas anyone could have had.
Search visibility and showing up in AI answers overlap a lot, though they aren't identical, and strength in one tends to help the other. Olu's advice is to stop treating SEO, AEO and GEO as three separate projects with three acronyms attached, and ask your agency one question instead: "How can I show up in the right places?" We've written more about how your website is being read by AI before your customers get there.