Low-code platforms have been in enterprises for years. What has changed is that they now come with AI built in. A business analyst can describe a workflow in plain language, get a working app, connect it to a model and have it read emails, update records and send approvals — in an afternoon.
That is genuinely useful. It also means that systems touching customer data, money and decisions are being built outside the usual engineering process. Blocking it rarely works; people find a way around. Governing it well does.
What changes with AI
Classic citizen development produced forms, trackers and simple workflows. With AI added, the risk profile moves:
- Apps can make judgements, not just follow rules
- Prompts and model settings become logic that nobody reviews
- Connectors can send internal data to external model endpoints
- Agents can take actions across systems with the builder’s permissions
- Output quality can vary without anyone noticing
None of this is a reason to stop. It is a reason to know what exists and to put the right guardrails around the riskier parts.
Tier by risk, not by tool
One rule for everything either strangles small productivity apps or under-controls serious ones. A simple tiering works better:
- Personal productivity — used by the builder alone, no sensitive data; light controls
- Team tools — shared within a team, internal data; registration and basic review
- Business-critical or customer-facing — sensitive data, money, external users or automated decisions; full review, testing and ownership, often with engineering involved
Builders should be able to tell which tier they are in from a short checklist. When an app moves up a tier, the controls move with it.
Guardrails built into the platform
The best governance is the kind builders do not have to remember:
- Environments — separate development, test and production, with promotion steps
- Data loss prevention policies — which connectors may be combined, which data may leave
- Approved model endpoints — AI features only call models that meet your data handling and residency rules
- Managed identities — apps run with service accounts, not personal credentials
- Logging — who built it, who changed it, what it did
Most major platforms provide some of these controls. The work is configuring them deliberately and keeping them current.
Ownership and lifecycle
Citizen-built apps often outlive their builder’s role. When someone changes jobs, the app that their team depends on becomes orphaned.
Require for anything above personal tier:
- A named owner and a backup owner
- A short description of purpose, data used and who relies on it
- A review date, after which unused apps are retired
- A path to hand over to IT if the app becomes critical
An inventory of what exists is the foundation. Without it, every other control is guesswork.
Testing AI behaviour
Business builders are not usually trained to test model output. Give them simple tools: a set of example inputs with expected results, a way to rerun them after changes, and a clear rule that AI outputs affecting customers or money need human review until tested.
For higher-tier apps, engineering or a centre of excellence should help build a proper evaluation set before wider rollout.
A centre of excellence that enables
A small central team — platform admins, an architect, someone from risk — makes this work. Its job is to publish templates, approved connectors and patterns, run training, review higher-tier apps and spot when something should become a professionally engineered system.
If the centre of excellence is only a gate, people route around it. If it makes building safely easier than building unsafely, they use it.
Where to start
- Build an inventory of existing low-code apps and AI features
- Define three risk tiers with a one-page checklist
- Configure environments, data policies and approved model endpoints on the platform
- Assign owners and review dates for anything shared beyond one person
- Set up a small centre of excellence with templates and office hours
Incubics helps organisations set up low-code and AI guardrails, and takes over the apps that outgrow citizen development.