Someone said “agentic” 4 times in your last leadership meeting.
And every time you look at LinkedIn, someone’s talking about the 6 AI agents they just deployed.
Meanwhile, everyone wants to know what you’re doing with AI besides just drafting briefs and decks faster.
If your AI strategy = a prompt library, you’re being asked to debate build vs. buy for a category you aren’t even IN yet.
Let’s talk about it.
The Agentic Crossover
If you’re a Marketing leader trying to figure out AI agents, you’ve probably been handed the same 2 options as everyone else:
Build your own. Or buy something off the shelf. End of menu.
Pick A or pick B, present it in the next leadership meeting, and keep it moving.
The data supports this:
Kana just surveyed 225 CMOs, CAIOs, and CDOs at enterprise companies. (That’s Chief AI Officers and Chief Data Officers.) Most of them claimed the same thing: agents in production, moving fast, handled.
But in that SAME SURVEY, they also flagged 2 specific blockers: data governance wasn’t in place, and their teams hadn’t been trained to actually run the agents day to day.
Confidence and readiness aren’t the same thing. 1 of them was there. The other 1…really wasn’t.
Today, we’re talking about what each option REALLY costs you. Then (and only then) I’ll help you pick a lane.
The spoiler is: I think the framework itself is wrong.
BUT FIRST: hold your horses. Before we even get to “build vs. buy,” there’s a step this framework skips entirely. And it’s important.
1st: Are You Even Past Chat AI Yet?
Saying this gently, but if your team’s “AI strategy” is a shared doc of good ChatGPT or Claude prompts, you’re debating build vs. buy for a category you haven’t actually entered yet.
Typing a prompt, getting a draft back, editing it, moving on, that’s useful! It’s also not agentic anything.
Here’s the difference, quick & dirty:
- Chat AI is a conversation. You ask, it answers. You’re the one triggering every single step and collecting + uploading all the context it needs. Write this email, summarize this report, draft three headline options. Nothing happens unless you type the next prompt.
- Agentic AI is a system. A system that executes a multi-step task with less of you in the loop at each step. It drafts the email, yes, but then it pulls the right audience segment, checks it against send frequency rules, schedules it, and updates you on the results. You set it up, it runs the workflow autonomously.
A customer service chatbot with a good prompt library isn’t a Marketing agent. It just feels like one because it’s fast.
That’s the huge divide Kana’s survey exposed: leaders using “agentic language” to describe regular, generative, chat AI. And then, in the next breath, admitting that governance and training for real agents wasn’t actually in place. Even if they DID have “agents” already in production.
Here’s how you actually cross that gap:
- Find the WORKFLOW, not the individual task. Chat AI is great at single tasks. Agentic AI earns its keep on a workflow with multiple steps that currently require you to manually move between three or four tools. (And if you can’t name the full sequence of steps, start to finish, stop right there: you’re not ready to agentify it yet.)
- Map WHO checks the output today. Whatever human review happens in that workflow now has to be designed INTO the agent. “Agentic” doesn’t (and shouldn’t) mean “unsupervised.” It just means the supervision is built into the system instead of living in your calendar.
- Start with 1 workflow. The teams that stall out are the ones trying to agentify everything, everywhere, all at once. If you want to make the agentification leap successfully, pick one (1!!) repetitive and well-understood workflow, get it right, and expand from there.
- Decide, honestly, whether this needs a standing agent at all. Sometimes the answer is just a better prompt or skill file, not a permanent system. I know the pressure to leverage this can be intense, but not every task justifies an agent watching your data around the clock.
Okay, so let’s say you’ve now identified a real workflow worth agentifying.
NOW we’re ready to pick a lane.
Let’s review those options.
OPTION A: Build it Yourself
This is the option that sounds the most impressive in a board deck. “We’re building our own AI agents in-house.” Bold. Ambitious. Very LinkedIn.
Option A gets you real customization, no question.
The problem is what it takes to get there:
- A data science or engineering team dedicated to building, testing, and shipping the agent
- The infrastructure underneath it: an LLM, a way to sequence multi-step tasks, and integrations into every tool it needs to touch (CRM, CDP, ad platforms, warehouse)
- Ongoing maintenance – every time your stack changes, the agent needs updating too
- Security and compliance review, on a system that’s now touching customer data
- Someone whose actual, for-real job is watching this thing and correcting it when it’s wrong
No data science or engineering headcount sitting idle, waiting to take on your project? Bummer. Because that means the “build it yourself” option just became “you spend the next 6 months learning to be an amateur AI engineer.” On top of the Marketing job you were actually hired to do. Congrats!!
Not to mention, an agent isn’t a 1-and-done build. It has to be managed like a direct report that sometimes lies to you, not a tool you configure once & leave alone. Someone has to keep coaching it, correcting it, and updating its scope as the business changes, indefinitely.
Marketers didn’t sign up to become systems administrators or enterprise-grade infra builders.
OPTION B: Buy the Point Solution
Option B feels safer. You’re not building anything, you’re just buying a tool!
Plus, by now, every single vendor in your stack has bolted an “AI agent” onto their product. Your CRM has 1. Your ad platform has 1. Your content tool has 1. Your data warehouse has 1. I’ll spare you the rest of the list, LOL.
Individually, each 1 looks like progress. You bought a thing, the thing does a task, box checked.
Except, uncheck that box, because: none of these agents talk to each other.
You end up with a dozen disconnected agents, each 1 only seeing its own narrow slice of the business, not even aware the others exist.
That’s the difference between an individual agent and an enterprise agent…1 solves a task in isolation, the other operates across your real environment. WITH context.
The worst part is you might not even notice the fragmentation at first. Then suddenly it’s 18 months later, 3 agents are giving 3 different answers about the same customer, and untangling it costs more time & money than they ever saved.
To me, this looks like a re-run of the same trap we all fell into with bloated martech stacks. Buy a tool for every problem, forget the part where you make sure they all work together.
Because “teamwork makes the dream work” doesn’t just apply to humans.
(SURPRISE!) OPTION C: BUILD-WITH
Sorry for the fakeout. Yep, there’s a secret third option you might not know about.
You don’t have to choose between doing all the engineering yourself vs. settling for a fragmented tool that only sees part of the picture.
Build-WITH means an outside technical team makes custom, enterprise-grade agents alongside you, connected to YOUR stack and data.
That gets you:
- Customization like the build option, because it’s built around your specific data and workflows, not one-size-fits-all, and it can actually account for your company’s institutional knowledge and brand governance guidelines instead of ignoring them
- No engineering burden (unlike the build option), because the team that builds it also owns the maintenance, security, and governance.
- No fragmentation (unlike the buy option), because it’s designed to work across your entire stack instead of living inside 1 tool.
This is where Kana fits in. They’re the people who did that survey, and instead of acting as another tool in the pile, they absorb the technical lift so Marketing can stay focused on…Marketing.
Kana is tech-agnostic, meaning they build AROUND whatever you’re already running, rather than forcing you onto a new data warehouse or CRM just to get THEIR agent to work.
Why This Matters Right Now
A category this new lives or dies on who’s actually building it. So I looked into it, and Kana’s founders (Tom Chavez and Vivek Vaidya) aren’t new to this. Chavez was the co-founder and CEO of Krux (acquired by Salesforce) and Rapt (acquired by Microsoft). Vaidya was CTO of Salesforce Marketing Cloud and co-founder of Krux alongside him. These are two people who’ve built and sold category-defining infrastructure before, not 1st-timers testing out a new framework on your budget.
Most importantly, agentic Marketing isn’t really optional at this point. Every serious company is heading there, whether the internal roadmap says so yet or not. The real decision that’s left is WHICH path gets you there without a) burning out your team, b) fragmenting your stack, or c) lighting your budget on fire.
My Takeaway + Next Steps
You don’t have to build it yourself. And you definitely don’t have to settle for 10 disconnected point tools pretending to be a strategy. There’s a 3rd lane.
Here’s what I’d do this week: take 10 mins to review the Agentic Divide, and walk into your next leadership meeting with a real picture of where everyone actually is right now.
Then, join my event on August 19th. Me, the co-CEO of Sendoso, and the head of marketing at Kana are getting together to talk through how to actually get started agentifying your marketing, the practical steps that take you from prompting in a chat window to managing a semi-autonomous agent, the risks, and the rewards. Free & virtual.
👉 READ: The Agentic Divide
👉 REGISTER: Building Your First AI Marketing Agent
If you take 1 thing from this: remember you don’t have to have all the answers about Agentic Marketing right this second. You just have to be the person who’s paying attention, and asking the right questions while everyone else panics.
You got this.
