Aussie AI

Chapter 1. Strategy

Book Excerpt from "Generative AI Applications: Planning, Design and Implementation"

by David Spuler

Chapter 1. Strategy

The Aussie Method

No, it’s not a new diet plan, a stock trading algorithm, or something else. Rather, it’s how to build an AI project according to Aussie AI’s best practices, and it goes like this with a two-step procedure:

1. Be the best, and

2. Have a few laughs along the way.

So, here, I’ll start you off:

Q: How many LLMs does it take to change a lightbulb?

A: One (56.48%), ChatGPT (25.71%), Dog (11.87%), What type of lightbulb? (8.39%)

Amusingly, I ran that question through ChatGPT, and got some great lightbulb jokes back. I’m saving them for the next edition, but feel free to ask it yourself.

The Aussie Method rules of AI include:

All AI project meetings start with a knock-knock joke or alternatively “An AI walks into a bar...” (Ouch!).
If your code crashes the AI engine, you’re bringing kolaches to the next meeting.
We need a leaderboard tracking who’s winning the game of count the sunrises.
Train your AI to give answers speaking like C3PO.

I hereby delegate to you the right to create extra Aussie Method rules for your AI project.

Be the Best

All jokes aside, I really mean what I say in setting the goals of your project. If you’re adopting AI in your organization, you should set your goal to nothing less than this: be the best!

How?

Well, firstly, I don’t think that you’ll get there by throwing money at AI vendors. You can get some short-term wins here and there, and drop a few percentage points off your cost base down to the bottom line. Maybe there’s a few ways to get some happier customers, and some increased revenue from doing better at sales and marketing. Nothing wrong with that!

But I think that all of these are a tactical gain, but it’s a strategic loss if that’s all that you do. Such plans are a good first step, and I’m not saying to avoid doing them, but they lack appreciation for what’s going to happen over the next few years. A seismic shift is going to happen in terms of both major areas:

1. External — you’ll need to engage more deeply with your customers (or lose them), and

2. Internal — address the overall workflow efficiency of internal operations.

To win in both those spaces, you’ll need to be ambitious. Here’s how to be the best at AI in your industry or organizational context:

Grow core competencies in AI subareas
Build IP
Improve staff capabilities
Revamp internal processes
Address AI governance

In order to do all this, you need to make a plan at a very high level, and then execute it throughout the organization.

Build the Best

To be the best at generative AI, you obviously need to make it a core competency in your organization by creating an AI team. But my point is that you need to be not just using and integrating this technology, but also building all of the generative AI applications that you need. There is much to gain from just installing vendor products, or using off-the-shelf models, whether commercial or open-source, but almost every company is working on this, and doing so won’t make your company the best.

If you’re not in the tech business already, your first thought for your business might be that you don’t want to build a core competency in building generative AI apps. You can buy that expertise from vendors, instead of belatedly becoming a tech powerhouse yourself. Whether you’re an airline or a bank, there are more important things to focus on than entering the AI software development business, right?

After all, you didn’t really need to become experts in coding applications for cloud architectures or building smartphone apps. There have been major IT trends before, and you didn’t need that capability in-house. Why now?

The answer is: business-level logic. Generative AI is the first technology that operates at a high level of business-specific logic. Cloud backends and smartphone front-ends were new ways to run existing technology, but generative AI changes the whole technology itself.

You’re right to be sceptical of this new trend, and also justified in not wanting to become a tech company. It’s unlikely to be a good choice to pivot your company to start designing GPUs and compete with NVIDIA. Similarly, I’m not suggesting that you become experts in all of the deep tech aspects of generative AI. Look up to the top, not down the AI stack.

The decision is one of focus. Instead of building a core competency in low-level AI capabilities (e.g., Transformer engines or huge LLMs), look at the business-level capabilities and how they are changing. My suggestion is to consider creating an in-house core competency in building vertical-specific applications in your core business area, no matter what business you’re in.

The underpinning reason for this recommendation of creating an AI development competency is to achieve business-aligned growth. The idea mainly applies if your overarching aim is top-line growth from generative AI. You probably won’t need a core competency in building generative AI applications for standard backoffice automation.

If your goal with generative AI is mainly cost reduction and internal productivity enhancement in non-strategic areas, it’s a different story. There’s nothing wrong with seeking a bottom-line benefit, and plenty of opportunities to do so, but you probably won’t need to become experts in building generative AI applications to achieve this. The areas of back-office automation and staff productivity already have numerous vendors offering AI solutions for this, at least in the areas that are common to all types of businesses.

Harsh Truths

I’m optimistic about the value of AI in both business and consumer use cases. But there’s a lot of hype and hybris, so it’s not all strawberries and cream. Here’s some negative thoughts to ponder:

The AI will be wrong at times. Expect it, and learn to live with it. Be very careful about putting AI in places that cannot be overruled or bypassed by humans.
A large percentage of AI projects are being built with no real goals, just to be seen to be doing something (for the C-suite or the investors).
LLMs are really just not that smart. They don’t generalize well. Advanced multi-step reasoning techniques like “Chain-of-Thought” are in some ways more like workarounds than real intelligence.
If you’ve built a customer support chatbot, you’re not that special (everyone has). Worse, your RAG-based customer support chatbot is outdated already (and needs more money spent). Customer expectations are increasing, and there are newer advanced RAG methods.
Don’t fall down the rabbit hole and assume AI can do everything; it can’t. Just because a few test queries/prompts produce good results does not mean all your use cases will produce good results.
Data handling, cleanup, and preparation remains one of the top hidden costs of getting your AI projects sorted.
Don’t make the mistake that the next big leap forward is only a few weeks/months away. It’s likely far longer than that. Don’t bet your business on the expectation of something coming soon.
Everything your team does today will be outdated within 12 months. Or maybe sooner.
Building a core competency and an “AI platform” in your business, as we’ve advocated above, is really, really hard!

We don’t have a solution for you about all these issues, but you should at least consider them fully. Hopefully, there are some answers in this book.

Strategic Steps

You’ve probably already done some of these things, since the AI buzz has been going for a while. Nevertheless, here’s some thoughts on what to do at the very top level:

Create a special AI budget allocation
Create an “AI group”
Establish internal AI evangelists
Hire a Chief AI Officer (CAIO)

Here are some of the things that your top-level AI team need to consider strategically:

Opportunities to focus on the top or bottom of the P&L (i.e., customer engagement versus internal productivity).
Identify existing competitive advantages to extend with AI.
Consider buy versus build with intentionality in terms of strategy.
Look beyond buy-vs-build to strategic partnerships, investments, or M&A possibilities.
Think about who will control the changing customer touchpoints in an AI-first world.
Ponder the puzzle of where all your business data is located.
Consider consolidating and reducing the number of places where the data is.
Understand vendor roadmaps for AI so you don’t duplicate efforts already under way.
Extend security-related and privacy governance to include AI.

And here are some even more specific project steps, some of which are further covered in later chapters:

Seek consultant advice — not just tactical, but also strategic!
Inventory all AI IP assets and gaps to fill with future additions.
Identify staff capabilities and needs for re-training or hiring.
Review existing data in terms of breadth and depth.
Plan for future data capture including tech issues and user legal agreements.
Rush out and fill your shopping cart with a thousand GPUs.

The last one was a joke (or was it?), because I don’t want you to forget Step 2 of the Aussie Method: everybody needs a good laugh now and then.

References

Nayan Paul, Aug 27, 2024, Gen AI & LLM Adoption through a Common Platform instead of enabling ‘be-spoke’ use cases, https://medium.com/@nayan.j.paul/gen-ai-building-adoption-through-a-common-platform-instead-of-enabling-be-spoke-use-cases-b0cbc2e185a8
Sangeet Paul Choudary, Nov 7, 2023, Why your super-app strategy will (most likely) fail: Building competitive advantage in the attention economy, https://platforms.substack.com/p/why-your-super-app-strategy-will
Janelle Teng, June 21, 2024, State of the Cloud 2024, The Legacy Cloud is dead — long live AI Cloud! https://nextbigteng.substack.com/p/state-of-the-cloud-2024
Nayan Paul, Aug 27, 2024, Gen AI & LLM Adoption through a Common Platform instead of enabling ‘be-spoke’ use cases, https://medium.com/@nayan.j.paul/gen-ai-building-adoption-through-a-common-platform-instead-of-enabling-be-spoke-use-cases-b0cbc2e185a8
Simeon Emanuilov, Apr 4, 2024 LLM agent operating system (AIOS) and the future of LLM-powered agents, https://medium.com/@simeon.emanuilov/llm-agent-operating-system-aios-and-the-future-of-llm-powered-agents-3d08b4e91c34 https://unfoldai.com/aios-llm-powered-agents/
Eddie Forson, Apr 29, 2024, Why I’m building my own AI Agent library, https://medium.com/@Ed_Forson/why-im-building-my-own-ai-agent-library-e20ec9aa3647
A. R. Ali, K. Kumar, M. A. Siddiqui and M. Zahid, 2024, An Open-source Cross-Industry and Cloud-agnostic Generative AI Platform, 2024 International Joint Conference on Neural Networks (IJCNN), Yokohama, Japan, 2024, pp. 1-10, doi: 10.1109/IJCNN60899.2024.10650688, https://ieeexplore.ieee.org/abstract/document/10650688
Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, Joseph E. Gonzalez, 12 Feb 2024 (v2), MemGPT: Towards LLMs as Operating Systems, https://arxiv.org/abs/2310.08560 https://memgpt.ai/
Michael A. Cusumano, Vivek F. Farias, and Rama Ramakrishnan, Sep 2024, Generative AI as a New Platform for Applications Development, https://mit-genai.pubpub.org/pub/r8xcl5ol/release/1
Nicholas Grous, Andrew Kim, June 04, 2024, Generative AI: A New Consumer Operating System, https://www.ark-invest.com/articles/analyst-research/generative-ai-a-new-consumer-operating-system