Series brief: AI workflow platforms & tools
AI got to work.
Well, to be clear, AI has “gotten” to work for us as it now permeates enterprise applications with its automations, accelerators and ancillary augmentations that we hope will make our human work experiences better… and, if anything, AI has “got” to work, now that we have invested so much development time working through the evolutions of plain old AI, AI with retrieval augment generation (RAG) domain-specific alignment, the current era of agentic AI…. and the inevitable next killer AI development cycle which we will no doubt come by Christmas this year, or thereabouts.
All of which (extensive preamble) brings us to the question of AI workflows as a multi-level entity.
But are AI workflows the working methods and processes we use to build AI, or are AI workflows the best way to describe human job functions that are now augmented with AI?
Spoiler alert, it could be both.
In reality, we’re probably talking about the impact of AI on what used to be traditional working tasks, methods, roles and jobs.
Remember jobs?
They came before workflows when people just went to an office. We can reasonably suggest that now (especially in an era of machine-to-machine communication and now agent-2-agent interconnection and orchestration) we need to consider AI workflows as any element of an organisation’s total operation fabric that can be improved via the use of AI services.
AI workflows: a definition
Let’s agree that an AI workflow is a systematically structured process that relies on a degree of AI services to align, optimise, streamline, analyse, manage and steer a workplace function, role or task. Because we have the benefit of AI to draw upon, we can also say that (unlike a traditional work role, which might be comparatively static) AI workflows learn and improve over time as they are exposed to more data to gain higher levels of validated experiences. As such, AI workflows are (arguably) even more flexible than humans because they are engineered and architected from the start of adaptability and change.
Elements of an AI workflow could include data collection, data ingestion, data deduplication as well as data processing, pre-processing and preparation.
Components of AI workflows
At the model level, AI workflow engineering would include the selection (or creation) of the machine learning or deep learning model employed in the AI workflow process itself. A subsequent (and critical) stage in any AI workflow moves the software system in question to the point of action execution, which might be nothing more than an alert or output of some kind, or could be a more complete business-level decision instruction.
We can see the deployment of AI workflows in potentially any workplace function.
Key areas here to highlight include customer service (AI chatbots and now more extended agentic services), employee onboarding inside HR systems (everything from workplace security badges to laptops and issuing company cars) and financial services (where correlation and pattern recognition is always useful for fraudulent activity detection) and many more besides.
Vendor selection
Who creates and makes AI workflow technologies?
This is a technology attack surface and topography that spans every tech vendor from CRM and HR platform companies… onward to ITSM vendors, through to real-time data players, database firms and onward to neuro-symbolic AI purists… and everyone in between.
Usual suspects would of course include the three major cloud hyperscalers (Microsoft is known for Power Automate… and although Redmond defines this as a process automation platform, the proximity to our discussion here in solid) and – after Google and AWS – we can also look to major vendors such as SAP, Salesforce and Snowflake.
Then there’s always Oracle, aiming to label itself as a hyperscaler as it does, or a provider of “strategic cloud platform services” as magical analyst house Gartner likes to quadrant-ize it. Oracle has its Oracle AI for Fusion Applications in this space.
Then there are specialist workflow vendors such as Zapier, UiPath, Automation Anywhere, Workatoa… and AI purists too such as Jasper (for marketing use cases) and Notion (for workspace and productivity tools with integrated AI) and others such as Cflow (big in finance), of course Appian (always strong in regulated enterprise environments and broader business use cases), lesser known names such as Tallyfy (strong on recurring workflows) and let’s also mention SnapLogic.
CWDN series
We now embark on the Computer Weekly Developer Network editorial series dedicated to examining AI workflow technologies. We want to know what code and data elements go into AI workflow platforms and tools, how they work, how they are orchestrated, what obstacles they face and what they can really do for employees in the workplace.
What are the most important factors to consider when looking at the AI workflow technology market?
Gemini thinks it’s ease of use, ability to integrate (do we even need to say “seamlessly” or can we take that as a given?) with other applications and APIs, the ability of any given AI workflow technology to exhibit great data management control, its ability to offer specific (cool) functionalities like natural language processing (NLP)… and of course ability to scale.
Business considerations in the market will always come back to so-called speed (or time) to value, total cost of ownership and (perhaps more interestingly) an AI workflow platform’s capability to oversee adoption and change management in the first place.
All of these factors need clarification… as does the need to showcase safety, security and a healthy portion of anti-bias and hallucination control.
Engineers, advocates and evangelists, we want your opinions and (actionable, if you must) insight.
Image credit Google Gemini
Image credit: ChatGPT