James Ernst of Core Order explains why most companies buy AI backwards — and what to ask before you buy anything at all.
🎧 Listen: Castos | 📺 Watch: YouTube
This week on the Unscripted Small Business Podcast, host Jeremy Rivera sits down with James Ernst of Core Order — eighteen years in marketing and PR, three of them running a restaurant in Buffalo, and the last two full-time in AI. What he has landed on after all of that is a deceptively simple argument: almost every business buying AI right now is doing it in the wrong order.
Stop asking which tool to buy
James doesn’t call Core Order an AI agency. He calls it a growth architecture firm, and the distinction is the whole pitch.
“AI is becoming just this plug and play with everything,” he says. “There are so many tools popping up that people want to just sell you and get embedded in your business.” What he watched happen across professional-services clients was predictable: departments kept buying tools, a few people adopted each one, and the company’s data ended up scattered across all of them. Which is precisely backwards, because AI is most powerful when it can actually reach your data.
“The first question you should be asking is not which tool should we buy, but what are we trying to make structurally better?”
— James Ernst, Core Order
Get that order right and you build something that lasts. Get it wrong and you have bought, in his words, another investment in a tool that sits unused for six months while somebody forgets they signed up for a subscription.
Jeremy copped to the same problem from the other side: he wrote a Claude script to comb his own personal finances and list every SaaS subscription he was paying for, just to see the damage. The dilemma he named is the real one for small businesses — the gap between what a tool promises and your team’s actual appetite and capacity to fully use it.
Homebrew and off-the-shelf are not enemies
James builds his own things enthusiastically. “I have my own AI Jarvis that runs my entire life right now outside of work,” he admits. But he refuses the binary, and his recommended pattern is worth stealing: take a system you already pay for — “maybe a CRM that has an MCP server, or we can leverage within API access” — and pull that data into a narrow custom workflow. Your data stays where it is governed; you gain the leverage.
His caution is equally concrete. “We’ve all heard the horror stories of somebody letting a rogue agent just wipe out a ton of data in a company because they don’t know what they’re doing and they’re following a YouTube tutorial.”
You will not out-build a platform with billions of dollars behind it. You can absolutely out-build it on one specific workflow that speeds up your team — which is exactly where a smaller company should aim.
Making sure AI says the right thing about you
Jeremy took the same logic outward. His friend Matt Brooks at SEOteric frames the answer engines as your own external customer support team: Claude, ChatGPT and Perplexity are already fielding questions about your business, and they will do a bad job of it unless you have published the information somewhere they can find it and trust it.
His own case study came from the launch of a data center energy solution in Texas. He mapped what the organization actually wanted said, then cross-referenced roughly thirty published PR articles against what the website itself claimed — and named the discrepancies out loud. Where the coverage matched the intent, that language belonged on the site. Where it diverged, the PR material needed a correction. “It’s as much alignment between the messaging as it is anything technical to worry about.”
The two objections that actually come up
Ask James what stops companies and the answer is short: security, and hallucination. “How do you make sure that it’s just not gonna go out and start making up data points, start filling in gaps that it doesn’t know?” And on the other side — if you hand it access to data, how do you know it won’t do something malicious with that data?
On finances specifically he is unambiguous. Feed it only into a locked-down database built to run the computations you need. Make sure client accounts are enterprise or business tier so inputs aren’t training the model. For anything that truly cannot leave the building, look at locally hosted open-source models. “Err on the side of caution before just giving it open access to everything.”
Guardrails belong in the discovery session
This is the part that most AI vendors skip, and it’s where Jeremy pushed hardest — citing his own consulting work with PureAire, where what you may claim about an oxygen monitoring system is tightly constrained, with more rules waiting on the sales and ops side.
James’s answer is process. Every Core Order engagement opens with a really in-depth discovery session, followed by six domains defined for that client which then feed the system they deploy. Regulatory language and red lines get written into instruction files, and agent systems check produced artifacts against approved language. In a HIPAA-adjacent business, that isn’t polish — it’s the difference between usable and unusable.
The number that sells it
Pressed on where the wins actually show up, James doesn’t reach for anything exotic. Find a task that eats four hours a week per ten employees and get it down to thirty minutes. That’s it. The rest of the value is keeping everyone on the same messaging — companies write mission statements and brand guidelines, then discover nobody repeats them consistently once they’re out in the world.
He also shared the build he most likes talking about: for an agency partner, an agent swarm that automates prospect research. Feed it a domain and a contact name and it crawls the site, writes markdown files, and uses tools like Tably to surface PR pieces carrying official backlinks back to the site so it knows the source is trusted. Out the other end comes a company profile plus recommended services for the first sales call. ROI data is still coming in, but the sales team walks in with a sharper talking point — and a live demonstration of capability the prospect can see.
Superpower or distraction
Jeremy asked for a hot take on entrepreneurship in the age of AI. James — whose wife runs three businesses of her own — gave the line the episode is named for.
“AI could either be a superpower or a distraction.”
— James Ernst, Core Order
Chasing every model release and every new tool is a genuine risk when you already have too much on your plate. But so is missing the one automation that would give you back the hours you got into business for in the first place. His framing: empower an agent the way you’d hire a CFO or a COO — for the thing you are not good at.
Give the time back to the humans
The pain point underneath all of it, he says, is a lack of time — sharpened by a nagging sense of falling behind, because AI means everyone could always be producing more. “We’re creating much more than we ever have before, but it doesn’t feel like enough.”
What he explicitly won’t do is turn that into a headcount argument. “I know there’s a lot of people that have conversations of how do we reduce workforce. And that’s never the approach that we take with it.” The goal is giving people back time to do what they actually enjoy — the analyst digging into numbers, the recruiter helping someone land a job. Keep the human in the loop.
Why human work gets more valuable, not less
The conversation closed where Jeremy keeps landing lately: the more AI slop gets produced, the more human work appreciates. He drew the line through Photoshop — before the 2000s a photo was proof, and then “that’s photoshopped” became a reflex. Trust in the medium erodes, and authorship becomes the premium.
James agreed, and went further. He expects AI advertising everywhere, having already spotted it in local ads during football season. He thinks AI can be the great equalizer. And he thinks the human piece will shine through anyway, in a new appreciation for human creativity, human thought and human empathy. He practices it, too: he writes a lot, and he doesn’t let AI write any of it, “because that is such a part of the human process that I want to keep.”
Notably, Core Order isn’t chasing the AI search and GEO side of this yet — how your messaging surfaces in Perplexity or Comet. James says that’s coming as it becomes more prevalent, but the current focus is internal: making the workflows better first. Which is, fittingly, the same discipline he opened with.
“Pandora’s box has been open and now we have to learn how to navigate within it.”
— James Ernst, on why caution and adoption aren’t opposites
Connect with James Ernst
- Core Order — growth architecture for small and mid-sized businesses
- The Core Order blog — thought pieces, with more coming
- James Ernst on LinkedIn — where he’s most active, and happy to talk AI with anyone
🎧 Listen: Castos | 📺 Watch: YouTube
More small-business owners on putting AI to work without losing the human part.