Jensen Huang said it on stage in February. The next big leap in AI isn't a smarter model. It's tool use.
Dharmesh Shah said the same thing, more bluntly. If your AI agent wants to know how many customers you signed up last month, you don't want it reasoning over Reddit. You want it asking the system that was built to know.
Both right. Both incomplete.
Here's the part that gets missed.
If you simply hand your database to an agent and call it a tool, you've given a stranger the keys to your warehouse and pointed at the inventory. Helpful, sure. Also terrifying.
A database isn't a system. A spreadsheet isn't a system. Even an accounting system, with all its tables and APIs, isn't quite a system. Those are records. Snapshots. Evidence of what already happened.
What an agent needs is something different. It needs a place where the rules live. Where credit limits get checked, batches get traced, prices get a sanity check, and someone (or something) has to ask permission before doing anything irreversible.
That place has a name. It's the system of engagement.
The system of engagement sits above the data. It's the layer that says: it doesn't matter if you're a human in the warehouse, an integration from Shopify, or an autonomous agent acting on a manager's behalf. The same rules apply. The same trail gets written. The same flags get raised when something looks off.
Records are passive. Engagement is active.
Most small and medium manufacturers don't have one. They have an accounting package, three spreadsheets, an inbox, a WhatsApp group, and a guy named Pieter who knows everything. The rules live in Pieter's head. When an agent arrives next year and asks to book a sales order, there's nothing to keep it honest. Pieter is on holiday. The agent will book the order. And the lot won't trace, the credit won't check, and the audit won't reconcile.
This isn't a problem the agent can solve. No matter how clever it gets.
The companies that will get the most out of an agentic workforce in the next two years are not the ones with the cleverest prompts or the biggest context windows. They're the ones who already did the boring work. They mapped their processes. They wrote down their rules. They built (or bought) a system that enforces those rules at the boundary, regardless of who's pressing the button.
The agent then becomes a very fast, very tireless coworker who happens to be governed by the same playbook everyone else is governed by. No special privileges. No backdoor to the database. No reasoning its way around the credit limit because nobody told it the credit limit existed.
That's the work for 2026. Not picking the right model. Not buying the right copilot. Building (or finally finishing) the system of engagement that sits between your business and anyone, human or otherwise, who wants to act on its behalf.
This is the bet I'm making with Floware.
We built Floware for brick and mortar businesses, a chemicals distributor who can't trace which batch of solvent went to which customer, and for a food manufacturer whose ISO auditor wants to know who changed a price and when. So every action in Floware (every quote, every purchase order, every shipment) goes through a typed, validated, audited layer. State machines. Event logs. Permissions. Flags that surface what needs attention before anyone has to go looking.
I thought we were building a traceability system. Turns out we were building a system of engagement. Same architecture. Different decade.
If you run a business on physical goods, here's the question worth sitting with this quarter:
When the agents arrive (and they will, faster than you think), what will they be acting against?
If the answer is a folder full of spreadsheets and Pieter's memory, no model on earth is going to save you.
If the answer is a place where the rules already live and the trail already gets written, the agent becomes the most reliable employee you've ever hired.
The work isn't picking the right AI.
The work is being a business worth automating.
Your business is unique, but your software is off the shelf? Ditch the workarounds and let's build your ERP systems to fit your teams.