Your AI Isn’t Your Advantage. Your Business Logic Is.

Who should we trust when AI starts running the business?

By Gerald Schlechter, Founder and Chief Strategy Officer, enosix

October 9, 2026

I have been thinking a lot about trust lately, probably more than at any other point in my career.

There’s a remarkably small group of people building technology that I think will shape how we work over the next ten years more than anything I’ve seen. Maybe it’s eight people, maybe it’s ten. The number doesn’t really matter. They’re moving incredibly fast, and the rest of us are building more of our businesses around whatever they make.

I have great respect for what they’re doing, and I use these tools every day. But would I let those people run my business? Probably. They certainly know how to scale one. But would I let their AI tools run my business yet? No, because those tools don’t know how my business works, and frankly, they shouldn’t have to.

I wrote a while back that AI is getting smarter and we shouldn’t let it make our businesses dumber. I’m even more convinced of that now, because the conversation has moved on from how smart these systems get to what happens once we let them act.

Every executive is going to run into some version of that question soon. The frontier AI companies are building the intelligence while Microsoft, Salesforce, ServiceNow, SAP and others put it to work, and all of them are moving from AI that helps somebody understand something toward enterprise AI agents that do the work themselves, creating orders, processing returns and making commitments without a person driving every step.

The clock keeps speeding up

Early in my career doing something meaningful for a customer took about a year, and we all thought that was normal. Then it was a hundred days, then weeks, and now an agent can go from what someone wants to acting on it in a few minutes. The time between an idea and its impact on the business has pretty much collapsed.

That’s amazing, and honestly, it should make all of us a little uncomfortable. When things took a year there were plenty of places to catch a bad decision along the way, and in minutes there aren’t. If trust isn’t built into the transaction itself, I’m not sure another corporate AI policy gives you much protection.

Which brings me to something I’ve believed for a long time, that intelligence is not authority. Being smart enough to make a decision and being allowed to act on it are two different things, and business has always understood the difference. I don’t see why AI should change that.

Intelligence is not authority: an AI agent can be right about what to do without having the right to do it.

Can it give the 8%? Why AI agents can’t own pricing and commitment decisions

Here’s a normal situation. A customer wants 2,500 units, another 8% off and a slightly different configuration, and they need all of it by the end of the month.

An AI agent may understand that conversation perfectly well, and it might even handle the negotiation better than most people on the team. But can it give the 8%? Can it promise the 2,500 units, will that configuration ship by month end, does the customer’s credit support the order, does somebody need to approve the discount?

Those aren’t AI questions, they’re business decisions, and most large companies already have the rules to make them. Pricing logic, available-to-promise, credit checks, contracts, approval thresholds, it’s decades of decisions about how the company operates, and for many companies much of it lives in SAP.

I haven’t seen anything yet that convinces me we should let AI negotiate those commitments on its own. I want it to understand the customer, work through the options and help people move faster, but helping negotiate a deal and having the authority to close it are not the same thing. If a company has spent thirty years building its pricing logic, why would we want a model approximating it? And if an agent is going to promise inventory it needs to ask the system that knows what can be delivered right now, not a copy somebody moved somewhere else three hours ago.

I’ve worked with companies where even a small percentage of incorrect transactions would be unacceptable. When you’ve worked hard to earn a customer’s confidence, one bad commitment can undo a lot of it, and I think about that every time someone talks about letting AI transact on its own.

There’s plenty of talk about putting guardrails around AI and I agree with it, but many of the most important ones already exist. A pricing engine doesn’t remind somebody about pricing policy, it enforces it. Available-to-promise determines what can be promised, and credit logic decides whether an order moves forward. Companies have built those controls into the way they operate, and once AI starts transacting, I’d argue they become some of the most important governance a company has.

The catch is that the agent has to reach those rules before it makes the commitment, not after. If it tells the customer they can have the 8% and SAP rejects it five minutes later, the system may have caught the problem, but you’ve already made the promise.

Where should the authority live? AI agent governance and a single source of truth

Gartner, Deloitte and even SAP are increasingly focused on AI agent governance: how to govern agents as they move from making recommendations to acting. I agree with that direction. But there’s a question I don’t hear asked nearly enough. Where does that control get its truth?

There isn’t going to be one agent running the enterprise. There will be agents in Salesforce, Microsoft and ServiceNow, plus SAP Joule, Claude, ChatGPT and plenty of things nobody has built yet. What we can’t afford is each of them carrying around its own version of the company, its own copy of inventory and its own interpretation of pricing.

Most of us have spent a good part of our careers cleaning up after enterprise systems that held different versions of the same information, and giving those inconsistencies the power to make decisions and execute transactions doesn’t strike me as progress.

So here’s where I land. Let the intelligence come from wherever the best intelligence comes from. Models, agents and applications are going to change, probably more than once in the next few years, and the authority shouldn’t move every time they do. It should stay with the business, and an agent that needs to act should reach the real business logic at the moment of decision, not a copy or a set of rules somebody rebuilt for one platform.

Connecting an agent to SAP is getting easier every month. Getting it to do the right thing once it’s there is the hard part, and it’s the part we’ve spent more than a decade working on at enosix.

We built our SAP process virtualization technology because we didn’t think companies should have to replicate their SAP data and logic across other systems just to make those systems useful. AI makes that far more important, and it’s a big reason we built arnold® AI by enosix.

arnold® isn’t another AI model. There are companies spending billions building those, and I’m happy to let them compete on who has the smartest one. What we’ve built is a governed execution layer that lets those models act against live SAP processes using the company’s own rules, rather than a second version of those rules living somewhere else.

Through virtualization, arnold® works with live SAP processes in ECC and S/4HANA. When an agent wants to price something, check availability or create an order, it runs through the same logic the company already relies on. The agent works within the permissions and business rules that govern the transaction in SAP. Where approval is required, the request can go to the right person before anything is committed, and the transaction remains traceable through the system of record.

What I care about most is that a company doesn’t have to rebuild those controls every time another AI application comes along. Whether a request starts in Microsoft Copilot, Salesforce Agentforce or ServiceNow, or in a frontier model like Claude or ChatGPT, the logic stays where it belongs. I don’t think a company’s governance should depend on which AI platform ends up winning. Let them compete on intelligence, just don’t hand any one of them the keys to your business.

Trust gets earned: a framework for expanding AI agent autonomy

None of this is an argument for slowing down. If anything, I worry more about companies that spend so much time talking about AI governance that they never give AI anything meaningful to do, and you don’t learn much that way.

I’d give agents real work, starting where the rules are clear. Let them execute inside those rules, send the exceptions to the people who should make the call, and widen their responsibility as they prove they can handle it. That’s what I mean by experimenting with permission to transact. It doesn’t mean putting a human in every loop either, because if somebody has to approve every transaction an agent touches we haven’t accomplished much.

Gartner predicts that by 2027, 40% of enterprises will demote or decommission autonomous agents because of governance gaps they only found after something went wrong in production. I don’t read that as a reason to avoid autonomy; I read it as a reason to build trust into the work from the start.

So, who do you trust?

I trust the models to reason, and I trust agents with a growing share of the work. I trust people with the decisions that need judgment. And I trust the rules we’ve spent decades developing to decide what the business allows.

Everyone is going to have access to extraordinary intelligence. Companies can license the same models, and I suspect the differences between them will matter less over time than most people expect.

What’s still going to be yours is the way your company works, your pricing and customer agreements and configurations and approval logic, and the thousands of decisions your people have made about how to serve customers and protect the economics of the business. My bet is the companies that do well over the next few years will be the ones that make that expertise safely usable by whatever AI comes along next.

That’s why I don’t think your AI is going to be your advantage. Your business logic is.

So don’t teach every new AI its own version of it. Let it come to the business, then let it go to work.

Frequently asked questions

Who should you trust when AI agents act for your business?

What does “intelligence is not authority” mean for AI agents?

Intelligence is not authority: an AI agent can be right about what to do without having the right to do it.

What is AI agent governance, and why does it matter?

Gartner, Deloitte and even SAP are increasingly focused on AI agent governance: how to govern agents as they move from making recommendations to acting.

How much autonomy should an AI agent get?

About Gerald Schlechter

Gerald Schlechter is Founder and Chief Strategy Officer of enosix. With more than two decades of experience in SAP environments, he has advised Fortune 500 companies on complex ERP landscapes and enterprise transformation, with a consistent focus on helping companies use the business logic they’ve already built rather than recreating it every time a new application or technology arrives.

About enosix and arnold®

enosix connects AI and modern applications directly to live SAP data, process, and business logic, in real time, without replicating it, migrating off it, or rebuilding what SAP already does well. Its technology works with SAP ECC and S/4HANA environments as they exist today, so companies can put their real SAP data, real SAP process, and real SAP business logic to work in whatever interface or AI experience people are using, without duplicating that logic anywhere else. arnold® extends that same real-time approach to enterprise AI, letting agents and AI experiences act on live SAP data, process, and logic while preserving the business rules, permissions, governance, and human judgment that consequential enterprise work still requires.

See arnold® in action