AI isn’t an instant on/off switch for improved marketing performance. Even as the models themselves improve, using them effectively is more challenging than assumed.
That’s why, as of 2026, 88% of orgs use AI in some form, yet only 6% observe strong financial results from it.
Edwin Choi, the founder and CEO of Jetfuel.agency, knows this feeling all too well.
He spent the last year rolling AI out across his 35-person team, and for a while it looked like a runaway success, with insights, ad copy, and reports flying out the door faster than ever.
Then he looked closer. Output was up, but quality had fallen, and the team spent their days cleaning up after the tools.
“AI doesn’t fail because the stack is bad,” he told the audience at this month’s AI for Marketers Summit. “It fails because the team can’t adopt it.”
Amen to that. Read on to learn about the system he built to close that gap.
1. Build the Vision Before You Touch a Tool
The vision isn’t to simply use more AI. Not every task needs or benefits from AI at all.
The way Choi draws that line is with a single question: What is the core of your company’s DNA?
For his agency, that’s being the client’s right-hand advisor on the hard calls and turning raw data into storylines worth acting on. For another team, it might be sales, or creative, or anything else. Whatever strikes to the core of that DNA stays human. Everything outside it, in his words, is “ripe for AI.”
He sets this as part of a wider vision – mapping where the team should be in 6 to 12 months, breaking that into goals each sub-team owns, with one senior person accountable for each.
2. Set the Guardrails First
With the vision agreed, the next task is fencing it in. AI strategy always needs guardrails so it doesn’t become sprawling and chaotic. Choi runs three guardrails:
- Focus on one thing at a time: It’s far too easy to end up running numerous half-built automations that are all equally mediocre, so each team commits to improving a single task well before moving on to the next.
- Collaboration on AI learning: AI brings different strokes for different folks, so different team members might bring different specialisms to the table. Choi has found that people learn by watching others rather than by reading instructions.
- Protect what’s human: Anything that calls for human-level strategy, judgment, or critical thinking is off-limits to automation. It’s a conscious, deliberate decision that protects the company’s DNA.
As Choi said himself, “We all collectively decided that any task or anything that we do that requires high-level strategy, critical thinking, or a sense of judgment will be human only. They will not be touched by automation.”
3. Have Everyone Write Their Perfect Work Day
One leadership exercise Choi has found surprisingly revealing is asking everyone on the team to write down their perfect day, the eight hours of work they’d happily do, every day, for the next five years.
The answers vary enormously. One person’s perfect day is spent ideating high-level creative concepts, while another’s is spent deep in data. Either way, the response tells him what people dearly want to hold onto to keep their work meaningful and enjoyable.
4. Build a Shared AI Brain
This is the centerpiece of the whole system, and the part Choi thinks most teams get wrong.
The common version of “sharing AI knowledge” is a folder full of prompts, but that only goes so far. He talks about a shared system that combines persistent memory, live context, a library of reusable skills, and tools layered on top.
Choi explains, “People are reinventing the wheel every time. When they start a conversation, when they start work, it starts over from scratch every time, and the prompt can only do so much. So this unified brain kind of compounds and unifies everyone’s work and efforts together.”
His build, which the agency calls Jetfuel HQ, involves three key layers:
- A central data layer that pulls every metric, KPI, and client detail into one source of truth. When a new campaign kicks off, the system reads the briefs, meeting transcripts, Slack threads, and emails.
- An MCP wrapper is essentially a secure pipe that lets platforms like Klaviyo, Meta, and Shopify talk to the AI, so any skill or agent the team launches can access that data layer.
- Claude wired in with fresh context loaded at the top of every session, so nobody ever opens a blank slate.
The payoff is shared memory across every team, reusable skills, and a master change log that the skills themselves learn from over time, with built-in KPI tracking.
This can be decisive in how companies leverage AI better than others. Harvard Business Review found that once everyone is running the same models, the advantage comes down to context, not tools.
5. Prioritize With a Value-and-Time Matrix
Once you can build almost anything you can imagine, the temptation is to build all of it, and that’s a fast track to burnout.
Choi has been there himself, with seven Claude Code windows open across his monitor and 14-hour days to match. “It’s just too little friction between what I can think and what I can execute,” he said.
His fix is to score every possible build on two axes: the value it creates and the time it takes, then let the resulting grid make the call.
- High value, high time. Where the team spends the bulk of its energy, on deep research, heavy data pulls, and the strategic synthesis that genuinely moves client outcomes.
- High value, low time. The quick-win pile, worth grabbing whenever there’s a spare hour.
- Low value, any time. Whether it takes five minutes or five days, it gets left alone for now.
It’s a deliberately crude filter, and that bluntness serves to divide tasks appropriately.
6. Run an Education Plan
None of this infrastructure is worth much if people can’t use it well, so Choi built an accessible four-stage education plan, and everyone starts at stage one on their first day. It involves:
- Baseline skills: The fundamentals of how LLMs work, why they hallucinate, prompt engineering, and time management. Everyone takes a prompting course, no exceptions.
- Infrastructure: Setting up your own local environment and troubleshooting it when it breaks. Needing your hand held through this is a red flag for the team, because the road ahead is a series of self-solved roadblocks.
- Specialized tracks: Here, the paths diverge, with SEO, paid media, creative, ops, and finance each splitting off into their own skill sets and agents.
- KPIs and growth: AI fluency becomes a pillar of performance reviews, anchored by a 90-day benchmark. By the end of it, you should be able to ship one brand-new automation from start to finish on your own.
As Choi described the result, “By the end of the 90 days of the AI education plan, you should be able to build one net new automation, on your own, from start to finish, and get it to a state where it’s workable and can ship out good work.”
7. Spend More Time Improving Than Building
Building an automation is the easy part, and the day you get it working is only step one.
The valuable work is ensuring it’s transparent and trustworthy. Choi’s rough rule is to spend a day building an automation, then much longer improving it: running it again and again, pressure-testing, fact-checking its output, and tightening the logic until the results hold up every single time.
He explains, “We would want you to spend 2, 3, 5, or 10 days improving the loop and feedback and the next versions of that, because that’s where the human power comes in. You need to fact-check your skill or automation, you need to make sure that you’re using your critical thinking judgment, that the output is sound and reliable and predictable.”
He says even when the AI executes and delivers something, a person still has to take it the final mile. Over time, this is precisely what allows people to spend less on execution and more on the strategic side.
8. Emphasize Culture
Choi’s last point, and the one he says matters most, is that the tools and skills don’t stick without the culture around them. The trick is being honest about how the team feels. AI is a lot to keep up with, and he’s clear that it lands on leadership as hard as anyone.
“All of us at JetFuel, we often say, hey, we are overwhelmed ourselves. But let’s stick to the plan, and keep in mind you’re still early, and here’s the vision for the company.”
The bigger worry underneath that is whether AI makes people expendable, and Choi meets it head-on.
“There’s a lot of anxiety around, well, will I be replaceable? And the answer to that, and it’s part of our vision, is that we are building individuals here who will not be replaceable.”
The rest are a few working habits. Everyone builds in the open and shares what they make, so progress spreads. Wins are celebrated. And new tools aren’t simply dumped on everyone without adequate time to test, experiment, and train.
Build Your Top-Level MarketingOps System
Adoption keeps climbing, and so does the failure rate that shadows it. Gartner expects 60% of projects to be abandoned through 2026, almost always for reasons unrelated to the technology itself.
If you take only three things from Choi’s playbook:
- Build the brain: It’s the connective layer that lets everything else compound, and you can begin with a single shared workflow.
- Choose culture: Tools and agents date in months, while culture is what carries a team through every model update.
- Protect human skills: Keep protecting your team’s judgment, taste, and critical thinking, because that’s the part that stays yours no matter how good the models get.
Want to dive deeper into building a world-class AI operating system your team will use?
Register for our upcoming Virtual AI for Marketers Summit!
Here are some similar articles you might enjoy:
- Top 10 AI Marketing Leaders to Follow in 2026
- How to Build Ethical, Brand-Safe AI Guidelines for Your Marketing Team
- AI Maturity in 2026: Building and Enabling the Modern Marketing Team
