Four ideas from an enterprise keynote that a 20-person business can use on Monday.
At its Think conference in May, IBM announced what it called its most comprehensive expansion of enterprise AI and hybrid cloud capabilities to date. The headline wasn't a product. It was a claim: that running AI in the enterprise requires a new operating model.
IBM's own framing of the problem is blunt. Many enterprises have invested heavily in AI, but only a few believe it is paying off. Its CEO's argument is that the companies pulling ahead aren't deploying more AI. They're redesigning how their business operates.
That's the same shift behind the "AI operating system" language coming from Microsoft and others, which we covered in Microsoft Says AI Is the New Operating System. IBM's version is worth reading closely because it breaks the idea into four parts. Each of them has a small-business equivalent.
The four systems in IBM's model
IBM describes the operating model as four integrated systems working together:
- Agents. Coordinated AI that executes and adapts across the business.
- Data. Real-time, connected information that gives teams a shared view of what's happening.
- Automation. End-to-end workflows that scale across processes.
- Hybrid. Operational independence for sovereignty, governance and security, so AI can run consistently and with controls.
IBM's point is that each of these is a separate priority enterprises are chasing, and the value comes from treating them as one system.
Idea 1: Agents need a control plane
IBM says the challenge shifts as organisations go from a handful of agents to thousands, built by different teams on different platforms. The problem stops being building agents and becomes keeping them governed and auditable. Its answer is a next-generation watsonx Orchestrate, described as an agentic control plane with consistent policy enforcement and accountability.
The SME version: you won't run thousands of agents, but you will run several: one reading email, one chasing invoices, one answering questions. Without a single place that says what each may do and logs what it did, you have the same problem at a smaller scale.
In Clara, that place is built in. Nothing goes out until a named person approves it, messages go from the person's real address, and every action is recorded with who, what and when.
Idea 2: Agents are only as good as the data they can see
IBM's second pillar is data. It notes that for most enterprises data is siloed and without meaning, and it pairs a real-time data foundation with a context layer so AI can reason reliably over business data, with governance enforced and decisions made explainable.
The SME version: your data is siloed too, just in smaller silos. The client request is in a mailbox, the order in a spreadsheet, the invoice in accounting software, the contract in cloud storage. An AI that can only see one of them will give confident, incomplete answers.
This is why connecting your existing tools comes first. Clara reads across Microsoft 365, Gmail, Xero, Sage, Manza, OneDrive, Google Drive and Dropbox, builds one picture of your clients, orders and documents, and cites its sources when it answers. If you want to see how this plays out with finance data, read Clara Now Connects to Xero, Sage and Manza.
Idea 3: Move from monitoring to coordinated action
IBM's Concert platform is pitched as a move from passive monitoring to coordinated, intelligent response. Where traditional tools capture metrics, it correlates signals into a single view and moves from insight to action with built-in governance and human oversight. IBM also points to a pain familiar to anyone who has run a small team: humans serving as the connective layer between systems never designed to work together.
The SME version: a dashboard that nobody opens is passive monitoring. What helps is something that notices an overdue invoice, a stalled request or an unanswered lead, prepares the next step, and brings it to a person.
That's how Clara works. It inspects your whole business every night, reports each morning on what needs you, and prepares the follow-ups for approval.
Idea 4: Governance and sovereignty aren't optional
IBM's fourth pillar is about running AI in sensitive environments, with regulated data and cross-border jurisdictions. Its line is that compliance can't be a configuration choice, and its Sovereign Core product embeds policy at the infrastructure level.
The SME version: you don't need infrastructure-level sovereignty, but you do operate under POPIA, which applies to any AI that touches personal information. The practical questions are the ones from our piece on POPIA and AI agents: where is data processed, who can see it, can you produce a log, and does a person approve consequential decisions?
What IBM's model gets right, and what to leave behind
Keep the structure. Agents, data, automation and control really are the four things that determine whether AI works in a business. Treating them as one system rather than four purchases is the right instinct at any size.
Leave the scale. A programme with private previews, hybrid cloud management and sovereign infrastructure assumes an IT department and a long timeline. Most South African SMEs have neither, and shouldn't wait for either.
| IBM's operating model | What it means at SME scale | Clara |
|---|---|---|
| Agents with a control plane | A few automations, with one place that governs them | Mandatory approval, real senders, audit log |
| Connected, real-time data | Email, accounting and documents readable together | Integrations with Microsoft 365, Gmail, Xero, Sage, Manza and storage |
| Insight to coordinated action | Overnight checks and prepared follow-ups | Nightly business inspection, morning briefing, automatic follow-ups |
| Governance and sovereignty | POPIA-aware handling and role-based access | Built to support POPIA, role-based permissions, encryption |
The takeaway
When IBM, Microsoft and the WEF independently describe the same shape, it's a good sign the shape is right. The useful move for a small business is to take the structure and drop the enterprise machinery. Ask of any AI tool you're considering whether it connects your data, does the work, keeps a person in control and tells you where the data goes. To compare vendors against each other, use the checklist in Agent Washing: How to Tell If Your AI Agent Is Just a Chatbot.
See the four ideas on your own work
Book a 30-minute demo and we'll show Clara handling the requests, orders and invoices your team deals with every day.