How Gen AI will reengineer workflows over the next 12 months.

After prompt engineering and adding a Gen AI copilot, what will happen next?

1. Prompt engineering

We’ve all spent time exploring how GPT-4, Gemini and other Large Language Models (LLMs) can help our productivity and creativity by crafting prompts that squeeze just the right amount of breadth, depth and detail from an LLM.

You know how it is, giving lots of context at the start, getting the LLM to use stages, adding loops and check points to give it more guidance, pushing it to give links we can check on and worrying about hallucinations that we have not spotted.

Prompt engineering feels like a mix of giving directions to a team member and writing software. And OpenAI has cleverly created a market for off-the-shelf prompts with its GPT Store. Even more cleverly, like any platform, it generates valuable data on what people want and what can be done for them … a valuable radar in a fast-changing industry.

2. Adding a Gen AI copilot

Adding a Gen AI copilot to your flagship product is a textbook strategy for using emerging technologies, which is why Microsoft bought all those OpenAI shares. Less affluent software firms have just rented the use of LLMs from the models’ owners, e.g. OpenAI, Gemini, Claude, Perplexity.

Many service firms just use LLM capabilities by adding Gen AI helpers for external customers and partner support. The market for Gen AI-based services is mushrooming.?

Also, firms with lots of rare data can use it to help build and optimise LLMs in the first place, or for searching through and accessing specific information to support internal decision-making. Good decisions depend on access to specific information and let’s face it, using information is all about being specific.

Some firms are benefiting from sitting on a gold mine of rare data, like Reddit and O’Reilly Media. Their job is to figure out what sort of decisions the rare information contained in it can support.

Lots of consultancies are keen to help you use RAG for this. Retrieval-Augmented Generation (RAG) is a way of getting up to date data that was recorded after an LLM was built, or which contains rare data. For example, market-level data is really useful, customer segment data is even better, but if you can get it, the gold standard is specific customer profile data and an app that can ask real-time follow-up questions.

3. What will happen next?

Human organisations are built around constraints, like all systems. Constraints are what gives them form and boundaries – remove a production bottleneck and you get a downstream bottleneck; fires burn at the rate they receive oxygen and fuel.

Human limitations are what we build organisations and invent technologies to exceed. By helping more people to work together smoothly to increase capacity. Or by recruiting talent with specific skills to increase capabilities. And by using flexible software to add more capabilities, to automate some tasks and to connect the work of people and machines together into workflows.

Every useful new technology exceeds the limitations of past organisational designs and LLMs are not an exception. Some good ideas are out there but it is unclear how this will unfold with Gen AI.

This is partly because LLMs are a black box of a ‘solution system’ and partly because of the complexity of the ‘problem system’. The exact shape of LLMs solution capabilities is still emerging – the technology is fundamentally complex and it is developing fast. Also, different firms present different problems, they have different arrangements of people and machines in different workflows, all with different input and outputs.

But some firms are in a unique position to see the future. For example, saas firms that provide applications which power human workflows (e.g. Zoho), or that provide the capabilities to connect other applications together (e.g. Zapier) or that provide human connection and collaboration tools (e.g. Slack).

Before I was an academic, I worked with Coca-Cola, Danone, Motorola and other firms on business transformation projects – and I think we are just seeing the start of a massive end-to-end transformation of company workflows. Organisations will look very different in a few years’ time.

AI Workflow Reengineering starts now

I’m researching for TRU in Canada and Nottingham University in the UK on how Gen AI will change the mix of humans and AIs in workflows over the next 12 months:

• AI versus human capabilities – what can we swap out?
• Changing human roles and tasks – which is best for different tasks?
• Human in the loop – where do humans have to be even when AIs can do a task?
• Augmentation and automation mix – what’s the balance?
• New ‘workflow architectures’ – enterprise and ecosystem levels
• Market level reconfigurations - disruption

Please get in touch if your organisation has started to think about AI Workflow Reengineering – I’d love to hear your views about the near future.