Someone pitched you last week, and the AI automation reseller vs technical distinction was nowhere in the slick deck, confident delivery, or promises about AI transforming your operations. You walked away with a nagging question you could not quite put into words: do these people actually build anything, or are they reselling a Zapier template with a markup?
That question is the single most useful filter you can apply before you spend a dollar. Get it wrong and you pay four figures a month for something you could have set up yourself in an afternoon. Get it right and you get a partner who can build the thing that does not exist yet.
I am going to give you seven tests. Ask them on the next call. You will know inside ten minutes.
Key Takeaways
- The AI automation reseller vs technical divide is not about tools. Both use the same platforms. The difference is what happens when your problem falls outside a template.
- Ask what happens when the API has no endpoint for what you need. A reseller changes the subject. A builder describes a workaround.
- Resellers sell you a workflow. Builders sell you a system that knows your business, then hang workflows off it.
- Request a screen share of a live build, not a demo video. Recorded demos are sales assets; a live workspace is evidence.
- 95% of AI projects deliver no measurable return, and the ones that work start with the process, not the tool.
- If the vendor cannot name what they will not automate, they have not thought hard enough about your operation.
- Reselling is not automatically bad. A reseller at reseller pricing is fine. A reseller at builder pricing is the problem.
- Ownership is the sharpest test: ask who holds the accounts, the data, and the logic if you cancel next month.
Why This Question Matters More Than It Used To
Two years ago, most AI vendors were technical by necessity. The tools were rough. If you wanted an AI system doing anything useful in a business, someone had to write code, stitch APIs together, and handle the mess when things broke at 2 am.
That has changed. Platforms got good. A person with no development background can now assemble a working automation in a weekend using drag-and-drop tools, a template library, and a YouTube tutorial. Which is genuinely great. It also means the barrier to calling yourself an AI automation company dropped to roughly zero.
So the market filled up. And the outside of a reseller and the outside of a builder now look identical. Same website structure, same testimonial wall, same language about transformation and bandwidth. You cannot tell from the marketing. You have to test.
Here is why it costs you real money. A reseller can install what already exists. That is a genuine service, and there are plenty of businesses where an off-the-shelf lead-response workflow is exactly what is needed. The problem starts at the boundary. Your business has something odd about it, because every business does. A booking system nobody else uses. A quoting process with four approval steps. A client who needs their data handled differently from everyone else’s.
When you hit that boundary, the reseller has two options: tell you the platform does not do that, or tell you it does and then quietly deliver something that does not. Both of those cost you months.
Test One: Ask What Happens When the API Has No Endpoint
This is the fastest test I know. Describe a real thing you need, then ask: “What if the software we use has no API for that?”
A reseller hears an obstacle. They will redirect you to what the platform does support, suggest you change your process to fit the tool, or tell you that integration is on the roadmap. Watch for the pivot. It is always a pivot away from your problem and back toward their catalogue.
A builder hears a puzzle. They will start thinking out loud. Is there a webhook? Can we scrape the page? Does it export a CSV on a schedule we could pick up? Could we sit a small script in between? They might conclude it is not worth doing, and that is a fine answer, but they will get there by reasoning about your specific case rather than by consulting a feature list.
You do not need to understand the technical detail of their answer. You are listening for the shape of it. Curiosity and specifics versus deflection and brochure language.
The follow-up that seals it
Ask: “Have you had to do that before? What was the situation?”
Anyone who has actually solved an integration problem remembers it, because it was annoying and it took longer than expected. You will get a specific story with a specific client and a specific tool. Resellers give you a general answer about their process and their team’s capability. General answers to specific questions are the tell.
Test Two: Workflow Thinking Versus System Thinking
Ask the vendor to describe what they would build for you. Then listen for whether they describe a workflow or a system.
A workflow answer sounds like this: when a lead comes in, we send a text, then wait an hour, then send an email, then create a task for your team. That is a sequence of steps. It is useful. It is also, fundamentally, a recipe that exists in a thousand template libraries.
A system answer sounds different. It starts with what the AI needs to know before it does anything: who you are, what you sell, how you price, which clients get handled differently, what your team is called and who does what. Then the workflows hang off that shared context. The lead-response automation, the quoting automation, and the follow-up sequence all draw from the same understanding of your business rather than each being configured in isolation.
The difference shows up six months in. A collection of separate workflows gets harder to maintain as it grows, because every change means touching four places. A system with a shared context layer gets easier, because each new piece of work inherits everything the system already knows.
This is what I mean by an AI brain and an AI workforce. The brain is the part that holds context: your business, your data, your priorities. The workforce is the set of things that act on it. Most vendors sell you the workforce and skip the brain, because the workforce is what demos well.
How to phrase it on a call
“If I asked you to add a second automation in three months, how much of the first one gets reused?”
A builder will talk about shared context, shared data, and how the second one is faster because the groundwork exists. A reseller will describe it as a separate project with separate setup, because for them it is.
Test Three: Ask for a Live Screen Share
Not a demo video. Not a recorded walkthrough. A live screen share of something actually running.
Recorded demos are sales assets. They are cut, rehearsed, and built to look good. That does not make them dishonest, but it makes them useless as evidence of technical capability, because a reseller can record a demo of a template they bought.
Ask to see the inside of a live build. The configuration. The logic. The place where the error handling lives. Ask what happens when a step fails at 3 am, and ask them to show you.
A builder will do this happily, because their own system is their best proof. I show people my actual workspace. Not a mockup, the real thing, folders and all, including the parts that are ugly. Anyone who has genuinely built something has an instinct to show it off.
A reseller will find reasons. Client confidentiality is the most common one, and it is sometimes legitimate, so give them an out: ask to see something they built for themselves. Every technical operator has built things for their own business. If they have not automated their own operation, that tells you plenty on its own.
Test Four: What Will You Refuse to Automate?
This one catches people off guard, which is the point.
Ask the vendor what they would not automate in your business. A reseller has no answer, because their pitch is that the platform handles it. They will say something vague about complex judgment calls and move on.
A builder has strong opinions and will give them to you unprompted. They will tell you that anything with a legal or financial consequence needs a human approving it. That the first version of anything client-facing should be reviewed before it’s sent, until you have watched it for a few weeks. That automating a broken process just breaks things faster. That certain conversations, the ones where a customer is upset or a deal is in the balance, need a person.
Strong opinions about limits come from having watched automations fail. That experience is exactly what you are trying to buy.
There is a version of this that matters even more in New Zealand and Australia: compliance boundaries. If you are in finance, real estate, health, or anything licensed, ask directly what the AI is not allowed to say. A vendor who has built in a regulated vertical will have a clear answer about where the line sits and how the system stays on the right side of it. A vendor who has not will tell you it is fine.
Test Five: The Ownership Question
Ask this exactly: “If I cancel in month two, what do I keep?”
The answers spread out fast.
Some vendors will tell you that you keep everything, because the accounts are in your name, the data is yours, and the logic sits in a place you control. That is the good answer. It means they are confident enough in the ongoing value that they do not need to hold your business hostage.
Some will tell you it all sits on their platform under their account and you would be starting again. That is not automatically disqualifying; plenty of legitimate software works that way, but it changes what you are buying. You are renting access, not owning a system. Price it accordingly and know that your switching cost grows every month.
The answer that should worry you is confusion. A vendor who has not thought about what happens when a client leaves has not had enough clients leave to learn from it.
The related question about your data
Ask where your business data actually lives and who can read it. This is a technical question with a technical answer. Anyone who builds gets specific quickly: it sits here, it is encrypted like this, these people have access. Anyone who resells gives you a reassurance instead of an answer.
Test Six: Look at the Pricing Shape
Pricing tells you more about a vendor than their website does, because the shape of the price reveals the shape of the work.
Reseller pricing is usually a flat monthly fee with a small setup charge or none at all. This makes sense for them. Setup is loading a template, so there is no real cost to recover, and the monthly is where the margin lives. If the setup fee is small and the monthly is large relative to it, you are probably looking at a subscription to someone else’s software with a markup.
Builder pricing usually front-loads. There is real work up front: mapping your processes, writing the parts that do not exist, testing against your edge cases. That work has a cost, and it shows up as a meaningful project fee. Then the ongoing sits at a level that reflects genuine ongoing effort: training the system as it hits new situations, optimising what is not working, implementing the next piece.
Neither shape is wrong. What is wrong is the mismatch: reseller work at builder pricing. Someone charging a significant project fee for what turns out to be a template install and a logo change.
So ask what the setup fee buys, specifically, in hours and deliverables. “What am I paying for in week one?” A builder itemises it without hesitating. A reseller talks about onboarding.
There is a useful piece of external research here. MIT’s NANDA initiative found that around 95% of enterprise AI pilots delivered no measurable return, and the projects that succeeded shared one trait: they started with the process, not the technology. The tool-first vendors are the ones producing that 95%. You are trying to filter them out.
Test Seven: Ask About the Failure
Every real build has a failure story. Ask for it.
“Tell me about a project that did not go the way you planned. What happened and what did you do?”
A builder has several of these and tells them without much discomfort, because failure is ordinary in technical work. You will hear about the integration that turned out to be undocumented, the client whose data was messier than anyone realised, the automation that worked perfectly and that nobody on the team would use because it changed their routine.
Notice that last one. The hardest part of this work is not building the thing. It is getting people to use it. Anyone who has delivered more than a handful of projects knows that development is a fraction of the job and adoption is most of it. A vendor who only talks about the build is telling you they have not got far enough into a project to hit the real problem.
A reseller either has no failure story, which means they have not done enough, or produces a sanitised one where the client was at fault. Blame-shifting in a sales conversation is a preview of blame-shifting in a delivery one.
What Good Actually Looks Like
If you run the seven tests and the vendor comes out on the technical side, here is roughly what you should expect the engagement to look like.
It starts with process, not tools. Before anyone talks about what software to use, someone maps how your business actually runs. What you do daily, weekly, monthly. Which tasks eat time. Which decisions route through you and only you. This mapping stage is where most of the value gets decided, because building the wrong thing well is worse than building nothing.
Then it starts small. One automation, chosen because it is high-frequency and low-risk, so you can watch it run and build trust before anything important depends on it. Not a platform rollout. One thing, working.
Then it compounds. The second build reuses the context from the first. The third reuses both. By the sixth, the system knows enough about your business that new work gets faster rather than slower. That compounding is the actual product. It is also the thing a template library cannot give you, because templates do not learn your business.
And there is a human in the loop where it matters. Anything with money, legal exposure, or a delicate relationship attached gets reviewed before it goes out, at least until you have watched it for long enough to trust it.
If you want to see what the underlying architecture looks like before you talk to anyone, I have written about what an AI automation agency actually does and about recovering from a bad AI vendor, which covers what to do if you have already been through one of these.
A Note in Defence of Resellers
I do not want to leave you thinking that reselling is a scam. It is not.
Plenty of businesses need exactly what a well-run reseller provides. If your problem is that leads sit for four hours before anyone calls them, and a standard lead-response workflow fixes that, then a standard lead-response workflow is the right purchase. You do not need bespoke engineering for a solved problem. Paying for custom work you do not need is its own kind of waste.
The rule is simple. Match the vendor to the problem, and match the price to the vendor.
A reseller solving a standard problem at reseller pricing is a good deal. A builder solving a genuinely unusual problem at builder pricing is a good deal. The two failure modes are paying builder prices for template work, and hiring a reseller for a problem no template covers and then spending six months discovering that.
The seven tests are not there to catch bad people. They are there to tell you which of those two deals you are being offered, so you can decide whether it matches what you actually need.
How to Run This on Your Next Call
You do not need to run all seven like an interrogation. Three will usually do it, and the conversation stays friendly if you frame them as curiosity rather than cross-examination.
Start with the API question, because it is the fastest signal. Follow with the live screen share request, because it is the hardest to fake. Finish with the failure story, because it tells you who you will be dealing with when something goes wrong, which it will.
Write down what you hear. Not the content so much as the shape: specific or general, curious or deflecting, opinionated or agreeable. Agreeable is not a virtue in a technical partner. The best ones will tell you that one of the things you asked for is a bad idea, and explain why.
Conclusion
The AI automation reseller vs technical distinction is not really about code. It is about whether the person across the table can think about your business as a specific thing with specific quirks, or only as a candidate for something they already have on the shelf.
Both kinds of vendor exist. Both have their place. The expensive mistake is not knowing which one you are talking to, because the pitch is identical and the outcomes are not.
Underneath all seven tests is one question: are they selling you a workflow, or building your business a brain? A workflow does one thing forever. A brain holds your context, sees your numbers, and gets more capable every time you add to it. That is the difference between a purchase that plateaus in month three and one that compounds for years.
Ask the questions. Listen for the shape of the answers. You will know.
Ready to Work Out What You Actually Need?
If you are not sure whether your problem needs a template or a build, that is worth thirty minutes of conversation before you spend anything. Book a 30-minute discovery call, and we will talk through what is actually eating your time and what the right shape of fix looks like. No deck, no pitch, just a straight conversation about your operation.
Frequently Asked Questions
What is the difference between an AI automation reseller and a technical builder?
A reseller installs and configures pre-built automation templates on platforms they resell, usually with a markup on the software licence. A technical builder writes and adapts the parts that do not exist yet, handles integrations without documented APIs, and builds a shared context layer that multiple automations draw from. Both use similar tools. The difference shows up the moment your requirement falls outside a template.
How can I tell if an AI automation company can actually code?
Ask them to screen share a live build rather than play a demo video, and ask what they would do if a tool you use has no API. A technical operator reasons out loud about webhooks, scheduled exports, or intermediate scripts. A reseller redirects you to what their platform already supports. Also ask for a specific failure story: real build experience always produces them.
Is it bad if my AI vendor uses Zapier or Make?
No. Those are legitimate tools that technical teams use constantly, because rebuilding solved problems is a waste of money. The question is not which tools they use but what happens at the edge of what those tools do. If the honest answer to every unusual requirement is “the platform does not support that”, you have a tool limitation being presented as a project scope.
How much should AI automation cost for a small business?
It depends entirely on whether the work is a template install or a genuine build. A standard workflow on an existing platform should cost less than a part-time hire and carry a small setup charge. Custom work that involves mapping your processes and writing new integrations front-loads the cost into a real project fee. The warning sign is builder pricing attached to template work.
What questions should I ask before hiring an AI automation agency?
Seven are worth asking: what happens when an API does not exist, how much of the first build gets reused in the second, can you show me a live system, what would you refuse to automate, what do I keep if I cancel, what does the setup fee buy in hours, and tell me about a project that failed. Three of those will usually be enough.
Why do most AI automation projects fail?
Research from MIT found around 95% of enterprise AI pilots produced no measurable return, and the common factor was starting with technology instead of process. A vendor who leads with the tool rather than with a map of how your business runs is following the pattern that fails. The projects that work map the process first, pick one high-frequency task, and expand from there.
Do I own the automations if I stop working with the vendor?
Ask directly, because the answer varies widely. Some vendors set everything up in accounts you own, so you keep the logic and the data if you leave. Others build entirely inside their own platform, meaning you are renting access rather than owning a system. Neither is automatically wrong, but the second changes what you are buying and grows your switching cost every month.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.