Engineering Journal
Build Software Agents Can Use, Not Just People
For most of software history there was one kind of user: a person with a screen. That is changing. AI agents now book, buy, compare, and coordinate on behalf of the people who run them. If your product can only be operated by a human clicking through a UI, you are invisible to a fast-growing class of buyer. At Mpaukwu, we design for both - the person and the agent acting for them - and we do it without handing the keys to anyone who asks.
An agent is a user with no hands and no patience
An AI agent cannot squint at a layout, guess what a button means, or wait on hold. It needs structure: a clear description of what your system does, well-defined actions it can take, honest error messages, and a way to prove who it is acting for. Give it that, and it becomes a tireless operator. Give it a pretty screen and nothing underneath, and it simply moves on to a competitor it can actually work with.
This is not science fiction, and it is not only a Silicon Valley concern. The same shift is coming to how a distributor finds suppliers, how a clinic restocks, how a buyer checks whether a business is real before sending money.
Why does this need a gateway?
You do not want every agent poking directly at your internal systems, and you do not want to expose raw data to something that could be impersonating a customer. That is why we built Africa Connect as a Model Context Protocol gateway: a controlled surface where AI agents like Claude, Gemini, and GPT can discover African businesses, check a trust score built from real verification inputs, and transact - without touching anything they were not explicitly allowed to touch.
The gateway is the difference between "agents can use us" and "anyone with a prompt can do anything." One is a feature. The other is an incident.
Identity comes before access
The first question is never "what can this agent do." It is "who is this agent acting for, and have they proven it." An agent booking a service is really a person or a business behind it, and that principal has to be verified before money or sensitive data moves. This is the same discipline we apply to SortAm, where verification and protected payments exist precisely because strangers are transacting - only now one of the parties is software. Trust score inputs, business verification, and scoped permissions are not paperwork. They are what stops the agent economy from becoming a fraud economy.
Guardrails are the product, not a setting
When you let an autonomous system act, the hard engineering is in the limits. Scoped access so an agent can read availability but not export a customer list. Audit logs so every machine action has a name and a timestamp. Policy checks so a request that is technically valid but obviously wrong gets stopped. Rate and value limits so a mistake is a small mistake. We build these the same way we build a hospital admin panel or an escrow release condition: assume something will go wrong, and make sure the system can see it, bound it, and undo it.
The same rigour, a new user
None of this is a departure from how Mpaukwu already works. We have always designed for the operator behind the app, for abuse before scale, for trust in the first architecture rather than bolted on later. An agent is just a new operator - faster, literal, and unforgiving of ambiguity. Products that were already honest about their state and careful with their permissions are the ones ready for it. Products held together by a human's judgement at every step are not.
The takeaway
Software is no longer only what a person clicks. Increasingly it is what an agent can safely be allowed to do on someone's behalf. That means a machine-usable surface, verified identity before access, and guardrails that hold when the caller never sleeps.
At Mpaukwu, we build for that world on purpose - because the businesses that agents can find, verify, and safely transact with are the ones that will still be discoverable when the front door is a prompt, not a browser.
Mpaukwu Trading builds founder-led SaaS products, automation systems, and production-ready platforms for African businesses, including SortAm, MediSeen HMS, and StoreBase. Read more from the Engineering Journal or start a project.