Osman Sief.
Articles

Low-code and AI: why the platform matters more in year two

The first AI use case is easy to fund. The platform choice decides whether the tenth one is.

Osman Sief · 2026-11-19 · 4 min read

In year one of an AI program, almost any architecture works. There is one use case, one sponsor and a lot of attention. The team can hand-build integrations, run a separate security review and accept some manual steps. The pilot goes live and everyone celebrates.

Year two is where programs succeed or quietly die. Now there are five use cases in the queue, each with a different owner. Every one needs identity, audit logging, integration with the same core systems and a user interface people will actually use. If each use case is built as its own point solution, the cost and the review burden grow with every addition. Momentum fades.

Why a shared foundation wins

This is where enterprise low-code earns its place. On a platform like OutSystems, the parts every use case needs, such as screens, workflows, roles, integrations and logging, are built once and reused. The AI component, whether it is a hosted model, a local model or a service such as IBM watsonx, becomes one module among many rather than a separate system to secure and maintain.

Point solutions

  • New stack per use case
  • Separate security reviews
  • Integration rebuilt each time
  • Cost grows with every case

Shared low-code foundation

  • One governed platform
  • Security reviewed once
  • Reusable connectors
  • Cost per case falls

In practice this changes the economics. Our first use case on a new client foundation usually carries the setup cost. The second reuses most of it. By the third or fourth, we are working from a template and the conversation shifts from can we build it to which one next.

Governance gets easier, not harder

Government clients worry, rightly, that more AI means more risk. A shared foundation answers much of that. Data residency, role-based access, audit trails and approval workflows are configured once and inherited by every new use case. Security teams review the platform pattern, not every application from scratch. Compliance with PDPL and NDMO guidance becomes a property of the foundation.

Use case 1: foundation
→
Use case 2: reuse
→
Use cases 3–5: template
→
Portfolio: run & govern

What to decide in year one

If you are planning your first AI use case now, make three decisions with year two in mind. Choose a delivery platform your own team can maintain, not just your vendor. Put identity, logging and integration in the foundation, not inside the first use case. And agree a simple intake process for the next use cases, so the queue is ranked by value rather than by who asked loudest.

Where the AI actually runs

A shared foundation also makes the hosting question easier to answer. Some use cases can call a hosted model. Others, because of the data they touch, need a model that runs inside the organisation's own environment or a national cloud. When the AI component sits behind a common service layer, the same application can switch between hosted and local models without being rebuilt. That flexibility matters as regulations and model options keep changing.

Building the team that runs it

The platform is only half of year two. The other half is people. By the second year, the client should have a small internal team that can configure screens and workflows, monitor agents and handle first-line changes without calling the vendor. Low-code makes that realistic, because business analysts and junior developers can work on the platform after weeks of training rather than years.

We usually plan this from the start: client staff join the delivery team on the first use case, lead parts of the second under supervision, and own the routine work by the third. The vendor stays for architecture, complex integration and new AI capabilities. That split keeps costs predictable and keeps the knowledge inside the organisation.

Signs your program is ready to scale

Three signs tell me a program is ready for its next phase. New use cases reuse more than half of what already exists. Security and compliance reviews take days rather than months. And business owners are asking for the next use case before the current one goes live. When all three are true, the platform has done its job.

A note on cost

Sponsors often ask whether low-code licences are worth it compared with open-source tools. The honest answer depends on the horizon. Over a single use case, a hand-built solution can look cheaper. Over five or ten, the licence usually pays for itself through reuse, faster security reviews and a smaller team needed to run everything. I encourage clients to compare total cost over three years and at least five use cases, not the price of the first project, because that is the scale at which the decision really plays out.

The first use case proves AI can help. The platform decides whether it becomes part of how the organisation works.


Articles

← Back to home