GenAI and low-code: What companies should consider when employees prompt their own apps
Digitalization, automation and simplification of daily work tasks: this is the objective of many companies looking to make processes more efficient and reduce the workload of their employees. But who is in charge of introducing new solutions? Can - and should - specialist departments use LLMs and low/no-code tools to create their own applications without any support from IT? What do you need to bear in mind?
“ChatGPT, write me some working code for an app that automates our approval process!” Could this simplify and accelerate the automation of processes or even the creation of entire apps? Not only would this relieve the burden on IT, it would also reduce costs and bring companies many other benefits.
For some time now, there have been various providers and solutions for no- or low-code programming that no longer require users to have a command of complex programming languages. Alternatively, they can digitally map and automate logical workflows, e.g. approval processes, using drag & drop. Whereas low-code still requires a basic level of manual programmatic intervention, no-code approaches rely exclusively on visual elements. The combination of no/low code and large language models (LLM) is now gaining momentum and opening up further possibilities.
No-code and LLM – when technologies clash
Currently, for example, a Microsoft research team is experimenting with a visual low-code approach with two LLMs in the background. A special feature here is the separation of tasks. At the beginning, the workflow is planned. This step is accompanied by the first LLM. Users can verify whether the process proposed by the planning LLM meets their objectives. During the subsequent execution of the workflow, the second LLM appears and provides content-related answers. A workflow can be created, for example, that supports employees in customer service or can also be used for a corresponding chatbot.
This kind of combination of several LLMs, each specialized for a specific task – such as planning and execution – combines the advantages of no-/low-code programming with those of language models and complements the increasing possibility of creating no-/low-code applications via prompting.
The simpler the operation of these services the more likely it is that employees will discover this option for themselves – even without official guidelines or regulations. And this creates opportunities for the business environment. As “citizen developers”, the entire workforce can develop automation solutions as prototypes from the bottom up, and IT can then support the roll-out to the entire company.
It can’t work without governance
As the saying goes: cleanliness is next to godliness. If every employee actually creates their own applications and processes with LLM and low/no-code platforms, an uncontrolled sprawl is almost inevitable. IT would have to get a whole host of apps under one roof. The risk of data protection and compliance breaches would increase rapidly – not least due to the use of as-a-service LLMs.
It is therefore strongly recommended that the issue be addressed at a corporate level.
Consider the following questions:
- Which people in the company should be able to design and publish solutions?
- In which scenarios and cases is this even an option?
- How are the relevant processes anchored and documented in IT governance?
- Which GenAI solutions and other IT products may be used?
- Do these solutions fulfill regulatory and compliance requirements?
- Do specific rights and roles have to be set up for the use of the applications or certain functions?
- Are the existing solutions sufficient technically or do they need to be expanded or modernized?
- Do the solutions need to be scalable if additional users are added in the future?
- What role does IT play in all of this?
- Which other stakeholders need to be involved?
- How are employees trained so that they can use GenAI in compliance with regulatory requirements?
- Do certain company-wide standards also need to be implemented for “citizen-developed” apps?
Safely exploiting the opportunities of LLM and low/no code
Given the potential for combining different technologies, further applications are likely to emerge in the future. Many different use cases can be mapped, for example through Copilot in the various Microsoft products or options for companies to operate their own LLMs or use GenAI solutions from third-party providers.
We will work with you to find the best possible solution for your company’s requirements. Please feel free to contact Nils Barkawitz and his colleagues: You can contact us here.