Helping the founders of the AI Design Sprint develop their GenAI Cards and Courses in addition to scaling the global 33A community.

Advisor - Generative AI

Built and developed a portfolio of enterprise customers over the course of three years at SAP. 
Year 1: $0 ACV whitespace territory 
Year 3: $5m+ ACV territory, selling over $12m (TCV) to business with up to 21,000 employees.

Account Executive

The Sales Academy is a global program for young talent held in SAP's world-class training center in California where associates will develop business acumen, presentation skills and technical expertise through intensive classroom training. Associates are integrated into teams to gain experience through shadowing, product training and mentor-ship to become the next-generation of sales professionals.

Academy Account Executive

MYOB

Partner Account Manager 

Partner Manager 

Insight Specialist 

Generative AI Advisor and Researcher | Giving Enterprises a Better Response to GenAI

Helping individuals and businesses improve the completion of day-to-day processes with generative AI and process automation.

Agents are hyped for good reason: 

Most people think that the LLM is the limiting factor for high accuracy GenAI services, this is wrong:

It's context. 

More specifically, it's how you architect the context (data). 

Large language models respond poorly to large amounts of context. 

They like bit sized pieces of context that allow them to stay laser focused on a task. 

At least, this is what is required if they're to stand up in a corporate environment that requires at least 90% accuracy. 

This is where agents come in. 

Imagine if we could create systems made up of many small components that are interacting with our digital environment and powered by a super smart machine that guides decision making a long the way? 

Well, you don't need to imagine, because we can create these systems by employing the emerging theory of agents. 

Important note: this document is a synthesis of many research papers and articles by AI greats such as Andrew Ng and is not my original work. 

Having used agentic architectures personally and for clients, it is truly a game changer. 

Please provide feedback to this document because it will evolve as more theory emerges on the topic of agents! 

What questions are left in your head when you think about agents?

Matteo Castiello

Understanding Agents

This is my framework for designing AI workflows:

Building GenAI workflows is slightly different to building software or traditional automations. 

Why? 

Because you are trying to mimic human behaviour. 

In an ideal world, a GenAI workflow is augmenting a human throughout their existing process to 2x-10x their productivity. 

The key word here is augment. 

The best use cases we've built are able to tear out a chunk of a humans existing workflow so that they're exponentially more efficient. 

Here's an example, let's say you're a sales person
• The best sales people personalise their outreach so that they can get the attention of their prospects and convert them into leads. 
• This involves researching news articles, annual reports, LinkedIn profiles, etc. 

So imagine if you could click a button and do this for 50 accounts in a single click as opposed to 15 minutes per account? 

This gives you 2x more time to do demand generation leading to 2x more leads and 2x cash for yourself and the business you're working for. 

This is a simple example but there are literally hundreds of these across all businesses. 

Important note: this framework is less important if you are just using ChatGPT. It's relevant if you are trying to build workflows that touch multiple interfaces and data points. 

What do you think is the highest leverage area for opportunity with GenAI?

Four GenAI Components

Microsoft released a four stage AI learning journey, it’s actually pretty comprehensive:

The program covers every persona that touches AI in a business from a business leader to citizen developer. 

Here’s a breakdown of the stages: 

1. Understanding AI: Establishes a foundational AI knowledge base within the organization, focusing on key terms and concepts to empower team alignment and success.

2. Preparing for AI: Emphasizes the need for specific skills to build and deploy AI solutions successfully, including managing data and using AI apps responsibly.

3. Using AI: Discusses the integration of AI across all organizational disciplines to enhance productivity and innovation through tools like Microsoft Copilot.

4. Building AI Solutions: Encourages developing AI experiences within apps and services, leveraging Microsoft's ecosystem to create and extend AI capabilities for business transformation.

Full details can be found here: https://lnkd.in/d_gVwjUb

Managing Director @ Insurgence - Delivering Enterprise Intelligence as a Service (iQaaS)

I use this framework to exponentially improve my interactions with AI: 

There's an argument that prompt engineering is dead. 

Yes, AI is better at you then crafting prompts but;

No, it can't read your mind (yet). 

Prompt engineering is the practice of communicating with AI.

That won't go away anytime soon. 

There's a lot of focus on single prompts however that's just one part of the story.

The real magic is in how you converse with the AI. 

There are no hard and fast rules, but this quick guide summarises the difference between:

Prompts: what you say

Conversations: when you say it

Don't be mistaken, this isn't a new concept. It's commonplace to build out these conversational dialogues into chatbots through chains and agentic architectures. 

What is new is the idea that we should build the skill of conversing by employing tactics that:

• suit our desired outcome

• use the appropriate amount of context (data)

• keep the AI focused

This guide is a quick introduction into the topic. 

To read more, you can check out a collaborative piece between myself and Chantal Smith at Exponential View here: 
https://lnkd.in/gNigtaEh

Preparation LI  Content  (1)

After hundreds of conversations on the topic, here are my top three parameters for a good GenAI use case: 

When you talk to people every day about identifying good use cases for GenAI, the following things become clear: 

• People's creativity holds no end, people are still thinking of fun new ways to use this technology and the possibilities are endless. 
  
• There is only one hard and fast rule: keep sensitive data safe. 

Outside of that, you can use GenAI and automations for just about any digital task.

This quote sums it up:

“there’s almost no business outcome that machine learning can’t help you improve today” - Morgan Vawter - Global VP of Data and Analytics, Unilever 

Important note, just because you can, doesn't mean you should. 

There are many tasks where you COULD use AI but that doesn't mean it's a good idea. 

So, after 100+ conversations on the topic, these three things seem to be good guidelines: 

1. 𝗧𝗮𝘀𝗸 𝗰𝗵𝘂𝗻𝗸: It's common knowledge that AI doesn't replace jobs or entire processes, it augments tasks. You want to think of it as augmenting tasks, even better yet, break up the task into a chunk. Get as granular as possible. 

2. 𝗦𝗺𝗮𝗿𝘁 𝗻𝗲𝘄 𝗶𝗻𝘁𝗲𝗿𝗻: Treat AI like a smart new intern. That means that you only give it tasks you'd trust to someone brand new. You don't give it access to sensitive information. Most importantly: you review everything it does. 

3. 𝗗𝗮𝘁𝗮 𝗮𝗰𝗰𝗲𝘀𝘀: When you have a technology that has almost limitless applications, you want to test quickly. Considering this, if you can't access 5-10 examples of great outputs for your use case within a day, move on to something else. 

Important point: If you're new to AI, this post isn't for you. The only advice you need is go to ChatGPT and try it out. 

Outside of these criteria, I'd strongly suggest reading Tobias Zwingmann's 'When to use GenAI Flowchart (link in comments). 

What else can people think of as good criteria for a GenAI use case?

Researching new stuff that might solve problems.

Tobias Zwingmann

These are the six stages of AGI as defined from a group of researchers at DeepMind:

1. No AI: All tasks are performed by humans without AI assistance, using analog or non-AI digital systems. No risks are introduced since AI is not involved.

2. AI as a tool: AI serves as a tool that automates simple tasks under human control. It includes things like search engines or grammar-checking programs. This level of AI is possibly emerging narrow AI and can lead to de-skilling and reliance issues.

3. AI as a consultant: AI takes on a more substantive role but is still under human initiation. Examples include language models for summarising documents and AI in programming. It ranges from competent to expert narrow AI and introduces risks like over-trust, radicalisation, and manipulation.

4. AI as a collaborator: AI works alongside humans as an equal partnerin tasks, such as in chess-playing or social interactions with AI- generated personalities. This involves possible emerging AGI (Artificial General Intelligence) and brings risks like anthropomorphisation of AI and rapid social changes.

5. AI as an expert: AI leads the interaction while receiving human feedback, such as in advanced scientific research. This could involve virtuoso narrow AI to expert AGI, with risks like societal-scale ennui, mass labor displacement, and decline in human exceptionalism.

6. AI as an agent: AI operates fully autonomously as an agent, represented by autonomous personal assistants that are not yet available. This level is associated with virtuoso AGI and ASI (Artificial Super Intelligence), with risks including power concentration and misalignment with human values.

My view: Trying to define these terms is like trying shoot a bullseye with cerebral palsy, it's better to just watch from the sidelines and see how things unfold.

Every statistic on the internet that proves generative AI makes you more productive:

This is a list of every available statistic that supports the now irrefutable evidence that generative AI makes people more productive across multiple disciplines.

References:
▶ National Bureau of Economic Research: https://lnkd.in/gbE2zh5K
▶ Harvard Business School and Boston Consulting Group (BCG): https://lnkd.in/gXKbywiS
▶ Science Magazine: https://lnkd.in/gRkZVjky
▶ MIT Sloan School of Management: https://lnkd.in/gJF2F7Fy

Generative AI Productivity (3)

This is how I get my best work done with AI:

“AI is much better at doing specific tasks than doing jobs” - Sam Altman

“It’s more useful to think about jobs as collections of tasks, and to analyze AI’s ability to augment or automate individual tasks.” - Andrew Ng

The reality of using generative AI tools is that if you rely 100% on it, your work will be generic and generally mediocre.

The best way to use generative AI right now is to treat it as a companion that assists you throughout moments where you would otherwise be lost.

As time goes on, you will find that more and more components of a process (tasks) are completed by AI than a human but for now, better to use the technology for it's strengths.

If you're stuck on where to start, I approach the technology by representing my life into the following categories:

🏯 Operating model: The mission of your life, job or business. This is a summary of your operational objective.

🍁 Process: The functions necessary to complete your mission. Each process should be considered a separate component of your operating model.

🕹 Tasks: individual actions that are conducted to complete a process.

By focusing on tasks but understanding your overall objective and the separate functions of that objective, it is easier to begin identifying opportunities for improvement from generative AI tools.

People forget that GenAI is a small drop in the AI ocean:

Let's not forget that before ChatGPT, there were multi-billion dollar AI initiatives being developed across large enterprises like Google, Facebook and Amazon.

Why?

Because traditional machine learning does a great job at predicting the future and optimising the results of huge structured data sets.

The price to use traditional machine learning methods is also drastically decreasing so it is often a great choice for businesses looking to get more value out of their structured data.

GenAI has democratised access to artificial intelligence and allowed anyone who can read and write to intuitively communicate with an intelligent machine however:

Traditional ML still is the go to choice for enterprises looking to adopt AI because it is more reliable and doesn't make things up.

It also has much clearer ROI because of the large amount of validated use cases and predictable results.

Can you think of a place where predictive AI would be better suited than GenAI?

This diagram from BCG is a great starting point to help educate the you on the differences!

This is how I assess a businesses readiness for AI: 

 A cookie cutter approach with GenAI doesn’t work, here's what does: 

1. Defining objectives: what is our goal? For example, increase employee productivity. 
   
2. Creating guidelines: How will AI be used? How will AI not be used? this is generally in the form of a policy and/or governance framework. 
   
3. Priorities: what are our focus areas? For example, enablement of employees or managing cost could be a priority.

Once you have these questions answered, you can choose your areas of focus and begin a cycle of:

1. Training: bringing everyone up to speed on the areas of focus.
   
2. Designing: creating a plan for how to implement the area of focus within your business.
   
3. Deploying: deploying the solution across the organisation. 

This cycle constantly repeats itself as new priorities emerge. 

Every business needs to assess its maturity and then decide where to focus their energy. 

Why? Because businesses have varying levels of maturity with technology that enables GenAI solutions. 

Depending on your teams maturity level, there are a combination of technical, organisational and financial considerations that require different skill sets from resources both within and outside of your organisation. 

What do you think is the most important area for a business to assess to understand their maturity levels?

This is the easiest way to structure your GPT prompts: 

Let's assume you want to create a GPT to help with one of your daily tasks. 

What are some important components to factor into your prompt? 

Role and Goal: Role and goal constraints ensure AI responses are relevant and provide responses relevant to your objective. 

Step-by-step Instructions: Provide clear, simple, and organised instructions for effective AI output.

Expertise (Pedagogy): Effective prompts rely on understanding the desired AI behaviour, focusing on inquiry over direct answers.

Constraints: Rules in prompts guide AI behaviour, making interactions predictable and tailored.

Personalisation: Using prompts that ask questions personalises interactions, allowing AI to adapt to different scenarios.

Here is an example using the process of preparing for a sales call. 

This prompt framework was originally published by Ethan Mollick. 

Full details here: https://lnkd.in/ekGBsYKh

Structured prompting

The most overlooked part of building AI apps and how to fix it:

There is a tendency for businesses to put the 'there's an app for that' hat onto AI. 

The reality is that AI is often a small piece of a much more complex puzzle including:
• Integrations 
• Interfaces 
• Cloud and on prem systems 

There's glory in claiming that AI will change the world but;

There's a reality where change in complex IT environments takes time...

A lot of time. 

Use this diagram as an example:

• It's explaining a very simple use case for an agent that assists a user to manage travel bookings.

• Despite it's simplicity, there are still many interconnected pieces in an already simplified diagram. 

Originally published by McKinsey, It's a great representation of many businesses reality. 

It's refreshing to see something that incorporates an entire applications architecture as opposed to just the LLM component i.e. RAG. 

Any application that goes into production needs to be integrated into a businesses existing tech landscape.

What is a potential work around? 

Instead of thinking of new apps to fit into your tech stack....why not build an AI workflow that: 
• Augments your people 
• Augments your tech stack
• Adds minimal technical debt 
• Doesn't create any new interfaces
• Saves you thousands to millions depending on the use case?

This is where businesses are winning with AI.

Will generative AI save the consulting industry before disrupting it?

This report has a point of view:

The professional services industry has been in a steady decline.

Three of the biggest providers have experienced negative YoY revenue growth (P.2 of PDF below)

This makes sense:

▶  The labour arbitrage (outsourcing) well is running dry.

▶ Software as a service is automating services = less demand for human services

How do you fix this?

The only way is with an influx of new services that aren’t yet automated or where there is a significant skillset that is in high demand.

Everyone should know the answer to where this demand will come from; AI/ML services.

“Enterprises can’t exploit GenAI alone and need smart partners, which can really help them get ahead of the innovation curve.”

This report from HFS Research makes a compelling case for a short to mid term positive outlook for professional services companies that can capitalise on the ‘technology arbitrage’.

The winners in this industry seem to be those that have a differentiation in technology as opposed to process.

Two examples we are seeing in market:
▶  McKinsey’s acquisition of Quantum Black; an AI consulting company that provides AI and data transformation services that are backed up a strong set of data, AI and ML tools.

▶  Accenture because it has the infrastructure to “redesign businesses to be effective in cloud ecosystems and manage them for its clients at scale:”

Report: https://lnkd.in/ebgYfuWn

Save Consultants before it destroys them  (1)

Every enterprise should learn from Amazons billion dollar mistake with ChatGPT: 

This shouldn't be a surprise to anyone...

Most front line workers are using ChatGPT behind their employers backs. 

This is backed by surveys and has been validated by my discussions in market. 

Every front line worker I talk to is doing this behind the backs of their employer. 

Why? 

It's simple, because it helps, a lot. 

One of my clients told their employer that they would quit if they didn't buy the team a ChatGPT Enterprise license. 

This firm footed employee is doing more than promoting a productivity tool.

They are protecting the businesses IP. 

This incident from Amazon will be the first of many enterprise that realise that taking no action with generative AI is much much worse than taking some small steps to protect IP. 

So what's the solution?

In this case, providing access to an LLM app that has controls over data would cost very little and provide the security required to keep data safe. 

In case you weren't aware: 
• The free version of ChatGPT will record your conversations for training purposes. 
• The Teams and Enterprise versions allow for an opt out of this feature. 

An even cheaper version of this (depending on the size of your business) is to stand up an internal LLM chatbot that gives you full control over your data and allows you to report on conversations. 

This would take a couple of hours with the right team. 

How big a risk do you think this is for enterprises?

This is the most misunderstood principle of building machine intelligence: 

In the past two days I've spoken to two executives at the big 4, both said the same thing: 

Businesses are over complicating AI. 

Caveat: AI is complicated, getting value from AI is not complicated. 

Why?

The barriers to entry have decreased 10x. 

Anyone can access highly capable machine intelligence for pennies on the dollar. 

So why aren't large enterprises deriving value? 

The answer is in behaviour. 

The entire market has been trained to think that getting value from software means:

• Ripping and replacing systems
• Preparing and migrating sh*t loads of data
• Managing change across many teams 
• Learning new UIs and processes 

So how does a business get value? 

The reality is that you can build machine intelligence:
• in days or weeks 
• on existing tech stacks 
• at a very low cost and therefore;
• to a small population of people 

This is how you start. Many of these small wins leads to big wins, that lead to market differentiation. 

You can't have the latter without the former because you need to build internal capability. 

Your people understand your process and need to combine that process knowledge with AI competency to design intelligence as a service (iQaaS) solutions. 

Businesses are winning by creating a pipeline of stellar simple but highly valuable use cases that lead to outsized returns on their investment. 

What do you think is the most overlooked think with AI?

2024 is the year of proving ROI with GenAI; here's what I've noticed:

Firstly, enterprises aren't that fussed with measuring ROI.

Don't be fooled, seeing a return is very important. 

The point here is that measuring it precisely is less relevant. 

Here are the stats to back that: 

56% of enterprises believe that ROI is positive but not precisely measuring it. 

Secondly, enterprises are incentivising adoption as a measurement of ROI. This measurement is largely subjective and based on qualitative feedback from users. 

A conversation with a VP last week highlighted that employees incentive programs are tied to AI application adoption. 

Is this reliable reliable metric of ROI? 

Thirdly, enterprises are always applying some sort of metric for success and if it's not met, they are axing projects like a lumberjack. 

Every business has their use case grave yard. 

These perspectives are based on last weeks conversations with over 15 enterprises + this report from a16z: https://lnkd.in/dvTxP29e

What trends do you see in how businesses are measuring ROI from GenAI?

The more I read, the more I believe that apps will disappear: 

If you're like me, you're knee deep in AI content on a daily basis. 

When you're knee deep in AI content on a daily basis, you notice patterns. 

One pattern is that many very well regarded individuals see the traditional front end app  becoming obsolete. 

Why? 

• LLMs allow us to interact with technology in natural language;
• LLMs can convert that natural language into commands for other applications;
• LLMs can communicate back to us the commands that it's taken on our behalf. 

This means that we are slowly inching towards a world were we do not interact with apps. 

There are just agents that we communicate with to do stuff for us in the digital world. 

This quote sums it up: 

"When AI agents are fully unleashed, every employee will have hundreds of automated, specialised consultants in their pockets.

Armed with the knowledge of everything, agents will help drive a new, more efficient way of doing business.

No one is ready."

 - Matt Pantaleone, Accenture 

This shift begs the question: 

How will we interact with the digital world? Is it the metaverse with deeply interactive experiences?

Will apps dissapear

12 months ago I realised two things: 

1. Almost every business process will be re-configured by generative AI models, integration technologies and data science practices.
   
2. The incumbent professional services firms are not the right people to do it. Imagine taking your car to a stable master in the late 1800s. Horse shoes don't fit on car wheels. 

I've teamed up with some of the smartest people I've ever met to re-think how we can deliver machine intelligence to Australian businesses in a way that is better, faster and cheaper than conventional consulting models. 

We're calling it Intelligence as a Service (iQaaS). 

Our early customers have seen wild results.

The best ideas are being kept secret. 

Businesses that are moving early and placing their cards right are seeing the potential to dramatically impact their business operations. 

Keen to bring this baby to life and looking forward to introducing the team in the coming weeks!

Taplio

Matteo Castiello's Linkedin Analytics

Matteo Castiello

Matteo Castiello's Best Posts (last 30 days)

Want to drive more opportunities from LinkedIn?

Taplio

.css-1jgreyo{background-image:linear-gradient(284deg, rgba(39,43,72,1) 0%, rgba(33,47,183,1) 100%);color:transparent;-webkit-background-clip:text;background-clip:text;}Matteo Castiello's Linkedin Analytics

Matteo Castiello

Matteo Castiello's Best Posts .css-qqfgvy{font-size:var(--chakra-fontSizes-md);}(last 30 days)

Want to drive more opportunities from LinkedIn?

Matteo Castiello's Linkedin Analytics

Matteo Castiello's Best Posts (last 30 days)