I talked to Jess Cook, VP of Marketing at Vector, about how marketing teams use AI when they are small, moving fast, and short on room for mistakes. The public conversation tends to split into two camps.
- Camp 1: One side proudly refuses to touch AI “AI haters”.
- Camp 2: The other claims an agent runs their inbox, their calendar, and most of their life.
Jess sits in the middle, and I suspect most marketing leaders do too. What made the conversation useful was how specific she got about where AI earns a place on her team and where it stays out.
We covered her rule for deciding when a human needs to step in, then three projects where Vector leaned in hard: annual planning, ABM ad creative built from sales calls, and a full website rebuild.
Somewhere in the middle
Jess didn't start out comfortable. Around March, it felt to her like everyone made a mass move to Claude and started building wild things overnight, and she instantly felt behind. About seven months later, after a string of real projects, that intimidation is gone.
Some work still stays fully human at Vector. She and Josh outline every podcast episode together, record together, and edit it themselves.
The only AI involved is a summary for internal use. Events are the same story: outside of enriching an attendee list, she doesn't see how AI would handle the logistics and details.
What changed most was the question her team asks. Early on, it was how to spin up more blog posts. Now, as Jess put it, the question is "how do we amplify and accelerate the things that we're really good at, but in a way that just helps us put more good stuff out into the world?"
She also thinks refusing AI outright isn't realistic for a company trying to grow. I agree, though the other extreme has its own cost, and we hit it ourselves at Omni Lab.
We built so much into skills and our internal OS that people would show up to calls without really knowing what the analysis said, so we pulled back and added friction on purpose. Jess has seen the same thing: teammates handing her or Josh a 20-page document labeled as analysis, when, in her words, "you didn't do the analysis, AI did."
When should a human review AI-generated marketing work?
Jess has a rule she uses to decide how much human involvement a piece of work needs, and how often. She frames it with a baseball question: "How many at bats do you get with this thing? And if you only get one, if it's like a big report, humans should come in right away, should be involved at every step, should be editing and reviewing very in depth."
The flip side is work where you get many swings. Her example was programmatic SEO.
If you publish 20 blog posts at once, you don't need heavy human editing on every one, because the first results will show you which ones hit. Maybe it's three, and those three get more love and attention afterward.
So the short answer: the fewer chances a piece of content gets to land, the earlier and more often a human should review it. High-volume, testable work can launch lighter and get human attention after the data comes in.
That lens explains why our own content process at Omni Lab is as human-heavy as it is. Our content marketer interviews an SME (right now just me) each week about a pain point we've mined from sales calls, uses a Claude skill to draft from the transcript, edits it, and then I review at the end for anything that would make it feel more authentic.
Each of those pieces gets roughly one at-bat. It's slower than going fully programmatic, and I'm still not sure where the right line sits for us as we scale volume.
One caveat Jess raised matters here. Programmatic content can only run lighter because a human blessed the system first.
Vector works with an outside agency that has spent over a year building guardrails for tone, voice, messaging, competitive positioning, and what the product can and can't do.
Before Vector repositioned as a contact-level ABM platform, that system spent months building credibility around the term, mining Reddit threads, Quora, sales call objections, and even LLM answers to questions like which ABM platforms are best.
Building a 2027 operating plan with a thought partner
When we talked, Jess was about three weeks out from a leadership offsite in Florida. Each department leader will present an operating plan, and then the group will horse-trade: if your program is strong enough, maybe I give up something of mine so you can fund it.
Her deliverable is a storytelling deck with a high-level calendar, a budget broken into big buckets, the three things marketing has to get right, and a version of the plan for a smaller budget, including what she would cut.
She's feeding Claude a full year of history. That includes budgets, every program they ran, which events they attended, the pipeline and revenue each one produced, and how ads performed. New inputs go in as they arrive, like a prospectus from an event she's considering.
The strategy is hers. She wants one core sponsored event per quarter, because this year's events drove strong pipeline and revenue but all landed in what felt like a single month. She also wants a steady cadence where the product roadmap sets the narrative and the narrative sets the tactics, repeated each quarter and carried into the ABM program.
Where Claude helps is the math. Jess describes herself as a 50,000-foot planner, and budgeting and projections don't come naturally to her.
Claude tells her things like how much efficiency has to improve, and that if certain pieces aren't done by May, she won't hit the goal. She's 18 months into her first marketing leadership role, coming from a content background, and this is the first time she has a full year of data to work from.
I asked whether all this saves her time. She didn't claim it does.
What she said is that she'll walk in with a more solid story: what they did, what it produced, where to double down, what to stop, and where to reinvest. I'd guess plenty of marketing leaders are getting the same trade without saying it out loud.
Her working style is worth copying too. She's spent about three hours in conversation with Claude on the plan and hadn't opened the deck in the last hour of it.
When she checks every change, she gets pulled into fixing one sentence on slide three. So she treats Claude the way she'd treat someone on her team: give it the full context, add new information as it comes in, then sit down with an hour to review the output and give real feedback.
From 20 discovery calls to one-to-one ABM ads
This was the first AI project the Vector marketing team built together. Jess came back from the Exit Five Marketing Leadership Retreat, where the only topic anyone wanted to discuss was what they were building with Claude. Her takeaway was that the whole team would need to be good at this by the end of the year, so they should start building something together right away.
They picked the discovery calls from their 20 most recent closed-won deals. The workflow went into Fathom, pulled every pain point and objection from those first conversations, grouped them into themes with exact quotes, and recommended ad copy for each theme.
The pain points were real. The first rounds of copy didn't sound like Vector at all.
That's where Jess's 15 years as a B2C copywriter came in. She'd rewrite lines to be more fun, or find a pop culture reference that said the same thing better. Each batch netted out at 15 to 20 pieces of ad creative, each tied to a pain point a buyer had actually voiced.
Later they pushed the copy into Claude Design for first-draft visuals, then edited and commented until it looked right. New batches went out every month or six weeks as creative fatigued.
Each pain point usually got two versions. Take ad targeting, where a buyer is spraying and praying with native platform targeting.
One ad was pure brand awareness: a clever line about the pain point with only the Vector ghost and no product imagery. The second went to people who already knew Vector, with a product screenshot showing how to build ICP-only ad audiences.
The ads pointed to matching pages, blog posts, or new landing pages. Building those landing pages exposed content gaps the team hadn't noticed, where a real buyer pain point had nothing on the site to answer it.
Once the one-to-many version worked, they templatized it for one-to-one and one-to-few ABM. The template pairs the Vector logo and a handshake emoji with the target account's logo, and Claude swaps in each company's logo and name. Because the pain points came from sales calls, Jess knew they were real for the persona being targeted.
I pushed on whether those insights made it back into the sales process. Her honest answer was mostly reassurance: reps nodded along because they heard the same pain points on their own calls.
We had a different experience at Omni Lab. Mining our Fireflies calls through Claude led us to rework parts of our sales deck around the four problems we hear most from VPs of marketing, and prospects often tell us it feels like we understand exactly what they're dealing with. I don't think either of us has fully closed that loop yet.
Leaving Webflow in under 12 weeks
Jess was quick to add caveats here, because people push back when she talks about it. Vector is a small team with limited resources, and it has to test ideas fast and drop the ones that don't resonate just as fast. That context matters before anyone rips out their CMS.
The trigger was the repositioning. Vector needed persona pages and pages for every new ABM feature, and the traditional path looked like Figma designs, rounds of edits, a developer handoff, and more rounds of edits.
She estimated 16 to 20 weeks that way. They had fewer than 12.
At the same time, she kept hearing about teams moving off Webflow and WordPress to sites built with Claude Code and managed in GitHub. That scared her.
She had never touched Claude Code and had lived happily inside chat. So she talked to people who had done it, including Ben Reed at Exit Five, who showed her live how he asked Claude to update a page and watched the change go through.
Josh handled the technical foundation. He cloned the Webflow site using Claude Code and GitHub, hosted it on Vercel, and put the product details in the GitHub repo so the copy could be checked against what the product actually does. Sanity runs as the CMS for the blog and templated pages, and Claude Code can update Sanity fields too.
The build was still real work. Jess put in about 80 hours over those weeks on roughly 20 net new pages: feature pages, persona pages, pricing, competitive comparison pages, and the podcast moved onto the site.
Each one needed new copy and product animations, and none could feel templated. She posted a LinkedIn photo of Penny from Big Bang Theory, deep in a late-night gaming session covered in Cheeto dust, captioned as her every night since they decided to do this.
What she has now is a site the team can change whenever it wants. As she put it, "I still haven't written a line of code. I don't know how to do that. But I can talk to Claude, and I can tell it exactly what I need it to do."
Annalisa, her content marketing manager, now builds interactive elements into blog posts and reports, something she couldn't have done three months earlier.
The same repo feeds a design system the whole team uses in Claude. It covers the elements and modules from the site, and Jess is using it for slide decks and new LinkedIn banners for a campaign.
The win is brand consistency: sales and CS can create assets that look right when they leave the building, without routing every piece back to marketing for review. She was clear this doesn't replace a designer, and she's looking forward to the day she can hire one.
We run nearly the same stack at Omni Lab: Vercel, Sanity, and GitHub, with Cursor as our editor so we can switch models and run multiple agents on different branches. Once the system knows your brand guidelines and what a good page looks like, it hits the mark on the first try more often than I expected.
Final thoughts
What stood out across all three projects is that Jess's team didn't hand AI the parts they're best at. The copywriting instinct, the planning strategy, and the brand's personality stayed with people. AI took on the work that used to bottleneck them: reading 20 sales calls, running the budget math, waiting on a developer.
The at-bats rule is the piece I keep coming back to. It gives a team a way to decide how much review a piece of work needs before anyone argues about it, and it pushes against the 20-page AI analysis nobody actually did.
The harder part, for us at least, is focus. There's always another AI project that looks exciting. Choosing the few that replace real manual work, and making those run well, is something I'm still working on.
