ARP: the successor to ERP that we started building before AI agents existed
An ARP (Agentic Resource Planning) is management software with artificial intelligence agents that don't just store business data: they work with it. What it is, how it differs from an ERP, and how we've been building it in ShephERP since 2022.
For thirty years, ERP has been the same thing: a program where you log everything that happens in the company. Sales, invoices, stock, customers, staff. It's indispensable, but it's also patient to a fault: it does nothing until someone logs in, finds the right menu, and types. Now that artificial intelligence can use tools, a word has emerged for what comes next: ARP.
First, what an ARP is
ARP stands for Agentic Resource Planning. It's the evolution of ERP (Enterprise Resource Planning): a management system that, in addition to storing business data, works with it. Inside, there are AI agents that query, propose, and execute tasks on the same database used by the team, with each person's permissions.
In short: in an ERP, you work for the program. In an ARP, the program works for you.
| ERP | ARP | |
|---|---|---|
| How it's used | Menus, forms, and lists | By asking in words (via chat, voice, or phone) and also through menus |
| Who does the work | The person, screen by screen | Agents prepare and execute; the person decides |
| When it acts | When someone logs in and clicks | Also beforehand: it notifies, proposes, and reminds |
| What it returns | Data and reports | Completed tasks and prepared decisions |
| Who it serves | Only the team | The team and, if you want, also your customers: by phone or on your website's chat |
By the way, don't confuse it with the ARP in computer networks (Address Resolution Protocol), which has nothing to do with it.
What an ARP is not
Today, almost every management software claims to have “AI.” In most cases, it's a chat stuck on top that summarizes data or drafts an email. Useful, but still an ERP. To call something an ARP, three things must be true:
- It acts on real data. It doesn't just answer questions: it creates a product, moves an appointment, prepares a purchase order, or takes an order, with the permissions each team member has.
- It gets ahead. It doesn't wait for you to ask. By morning, it already knows what's on your schedule, what's stalled, and what should be done first.
- It knows when to step aside. If it doesn't know something, it doesn't make it up: it says so, asks for confirmation, or hands the matter over to a person. And everything it does is logged: who requested it, what it did, and when.
The third is the hardest to achieve and the most important. An agent that makes up a price in front of a customer does more harm than one that doesn't exist.
What an ARP looks like inside: ShephERP
ShephERP is the management platform we develop at SoftCollie, in Valladolid. It has what you'd expect from an ERP: catalog and stock, point of sale, online store, invoicing with VeriFactu, customers, appointments, team, and time tracking. Each business activates only the modules it needs, and they all share the same data. That last sentence is what makes it an ARP. You notice it in day-to-day use; the screenshots are from our demo environment, with fictitious businesses and data.
1. The day starts prepared
When you log in, you don't see a dashboard full of charts. Instead, you see the “Today” screen. Shepherd, the assistant, has already reviewed what was closed yesterday and what's still pending, the schedule, and the tasks, and has laid out the day: first, the quote that expired yesterday; then, the supplier order due today. It fits everything into the free slots between appointments. If you're only going to do one thing today, it tells you which one.
And it doesn't just look at what's yours. It also sees the team: who's overloaded, who owes you something, which task has been stuck for days (and offers to unblock it), how collections are doing compared to last week, and what's worth keeping an eye on before it becomes a problem.
2. You ask it like a colleague
Instead of opening the inventory, filtering by stock, and building the order item by item, you just ask. The assistant decides which ERP tools it needs, uses them, and responds with real data. But it doesn't get ahead of itself: the supplier order is prepared as a draft and isn't sent until a person confirms it.
3. Automations you create with words
For repetitive tasks, there's no need to program anything. You describe what you want, and ShephERP turns it into an automation: it figures out when it should run, which tools it will use, and explains its plan in plain language before starting. By default, it checks with you before acting. Every execution is saved step by step, with what it decided and what it did. And if you ask it to do something it doesn't know how to do yet, it tells you instead of pretending.
4. It also answers the phone
The final piece is voice. ShephERP's phone assistant answers the business's calls with the business's own data, such as the catalog, schedule, or stock levels, takes orders or appointments, and logs them in the system. When it doesn't know something, it hands the call over to a person or takes a message. Since October 26, it has been answering SoftCollie's own phone: we are the assistant's first user.
The story: an ARP before it had a name
SoftCollie was founded in Valladolid in May 2022. ChatGPT wouldn't arrive until November of that year, and it would be a long time before anyone talked about “agents.” What we were clear about was the problem: small businesses pay for three or four different programs (one for invoicing, another for the online store, another for the calendar, and a spreadsheet for the warehouse) that don't talk to each other. And every piece of data that doesn't talk to the rest is manual work.
That's why ShephERP was designed as a single platform: one data model for all modules, and every action in the program as an operation with its permissions. In 2024, we launched the current architecture, with microservices and the same data types shared across the entire system. We didn't do it with artificial intelligence in mind, but with order in mind. But that order was exactly what made it possible to take the next step.
When language models capable of using tools appeared, we didn't have to bolt a chat onto an old program. It was enough to give the AI the same operations the application already used, with the same permissions. In early 2025, the first assistants connected to each business's data arrived. In September of that year, we adopted MCP (Model Context Protocol), the open standard that AI models use to work with external tools. In 2026, invoicing with VeriFactu, the call center, and, in May, the current version of the platform in production arrived, with real businesses selling, invoicing, and managing their day-to-day with it.
We didn't add AI to an ERP. We built an ERP that AI can use just like a person: with permissions, with context, and leaving a trail.
What's next
ARP doesn't replace anyone. It removes repetitive work so the team can focus on what only a person can do: serving well, deciding, and growing the business. The next pieces are in that direction: outbound calls (“your order is ready”), phone orders connected with delivery, and more specialized agents by sector, from neighborhood shops to clinics or workshops.
If you want to see it in action, call 983 460 815 and talk to our assistant, or try ShephERP. And if your business needs something tailored, tell us about it: that's what we've been doing since 2022.