Low/No Code AI Tools
There's been a lot of interest in Low and No Code tools. These are usually (web) apps that can be configured through a UI and work well with other apps to perform a custom task workflow. I use this definition in contrast to an AI first tool such as a spreadsheet like web app that helps anyone do advanced analytics. So, imagine using a google form to populate a spreadsheet and then having each entry analyzed and then emailed to someone in your organization based on the analysis.
The integrated workflow tasks can be done with custom software, and many people would prefer the control and efficiency of doing it that way, but it can also be accomplished with low/no code tools. You might be wondering, if you could use custom software, why would use use a limited low/no code gui approach. It comes down to:
- Development speed - often it is easier for someone to string multiple tools together than it is to find a developer and explain what is needed.
- Accessibility - a domain expert that is comfortable with technology can create something that works without extensive programming experience.
- Flexibility - the low/no code workflow can easily be changed and updated as needs change.
- Implementation cost - a low/no code approach may have a low monthly fees associated with each tool used but this is usually much less than the expense of having a developer create a custom solution and then maintain it.
However, as mentioned, some of the drawbacks include the tendencies for these types of solutions to be difficult to maintain and "version control" (that is keep a backup copy of each workflow configuration), are often not well tested, and tend to be slower or not suited for all workloads. Consequently we must carefully consider when and where the advantages out weight the drawbacks and risks. As well as when a low/no code workflow should be converted to traditional software to improve its resiliency and performance.
What about the AI Part?
Now, lets talk about what AI is. A definition for AI from PWC states:
In our broad definition, AI is a collective term for computer systems that can sense their environment, think, learn, and take action in response to what they’re sensing and their objectives. - PWC
I prefer to focus on the idea that not all problems can be solved/answered directly and that AI is the computational approach to finding good decisions with incomplete data for intractable problems.
Area of study focusing on systems that can adapt, learn, and reason to make (good) decisions even with ambiguous, uncertain or incomplete data and unsolvable (intractable) problems. - Me
It used to be that systems that could play (simple) games like checkers and tic-tac-toe were AI if they were not programmed directly. Then it was chess but now we take that for granted. Then go, etc. There is an old joke that
AI is whatever hasn't been done yet. - Tesler’s Theorem
I bring this up for two reasons:
- Low/No code AI tools may look like regular tools with the system using AI behind the scenes
- AI outputs may not be optimal or consistent but rather a system's best effort at providing a solution given the existing constraints.
The first point addresses the idea that we take some apps for granted. For example take generating a delivery route given a list of addresses that must be visited. There are many different ways that the deliveries can be ordered. It may be possible to generate the shortest route if there are not to many stopping points but if you need to consider traffic and weather predictions it is less clear that you can get to an "optimal" solution. In this case, you may want to use various AI techniques to generate a good solution.
The second point addresses the fact that you may get different routes even if you ask the same question in the exact same way, either because there is some randomness in the process or because conditions and intermediate predictions change from invocation to invocation.
These are both important issues, different from traditional software, to keep in mind when trying to automate a task with low/no code AI tools.
Example Use Case: Automating Background Removal in Photos
As an example, suppose you operate a consignment shop selling items on various online marketplaces. Good product photos can positively impact the sale price. A typical photo editing workflow could involve several steps, ranging from loading images into Photoshop to resizing them for each platform. So your current workflow may look like:
- Load photo into Photoshop
- Use magic wand to outline item
- Cut out the background
- Resize versions
- Save new photos
An AI workflow may be:
- Load photo into purpose built (background removal) app
- Press the process button
- Resize versions
- Save new photos
A No-Code workflow might be:
- Put photo in specific directory
- Wait while system processes the photo
- (There's no step three.)
Some other use cases where you can start thinking about using low/no code tools to automate processes include, automatically generating (sales, status, KPI) reports, researching and scoring sales leads, analyzing customer feedback and reviews, and optimizing daily operations
How do you create this magic?
A typical low/no code workflow combines both AI and non-AI tools. The primary components include:
- An orchestrater
- Webhooks and AI and Non-AI APIs
- Macros, scripting & keystroke recording functionality
An orchestrater is an app that is responsible for figuring out what step(s) need to be performed next and kicking off that process. On the web there are apps like Zapier, Bubble, IFTTT, Make.com, etc.. iOS/MacOS have a system/app called Shortcuts. I'm sure there is something similar for Windows, Android and Linux though I don't know them by name.
And finally many people are trying to use LLMs as an orchestrater with mixed results. This idea of using LLMs is getting a lot of attention right now and is usually what people mean when they speak of LLMs as "Agents". Personally I think that LLMs may not be the right tool for planning and reasoning but can work well when used together with other tools specifically designed for those purposes.
The once the the next step(s) is identified it is kicked off, usually through an HTTP (or local) API (Application Programming Interface) call. An API is list of functionality for programmatically interacting with an application. So for example the orchestrater may POST the image to API that removes the background or generates a caption. Often the APIs can take some time to complete so will accept a "webhook" which is another API that you (or your orchestrater) host so that it can get notified when the step is complete.
Another important set of functionality is apps that let you record macros and key strokes and let you do simple scripting. The tools let you automatically drive almost any web app and include it in your workflow. They're often called Robotic Process Automation (RPA) and they can exist as part of the web app itself, as part of the OS, or as a stand alone tool that you use to control other apps.
Several platforms like Zapier, Bubble, and IFTTT are commonly used orchestrators, while Google Forms, Google Sheets, and Airtable are frequent workflow components. I won't mention specific AI tools here, other than to say the major LLMs have APIs as will many AI tools, because there are many being developed and the ones to consider seriously are changing all the time. Perhaps check a site like There's an AI for That or others that come up
Summary
Low/no code tools offer powerful capabilities but come with their set of challenges. Keep the following in mind:
- Explore creative solutions while acknowledging real-world constraints.
- Integrating traditional approaches with AI apis can enhance this even further.
- Start with a simple demo or proof of concept; it could lead to remarkable outcomes.
By understanding the benefits, limitations, and the ever-evolving role of AI, you can make informed decisions about incorporating low/no code tools into your workflow.