Patterns for exploring "AI" in your organisation
Or what my current recommendations are, as things are moving damn fast
Its the 26th June 2025 in wintery Paekakariki, New Zealand.
The date is important as things in the “AI” space are moving so damn fast.
This evening I am leading a set of discussions on “AI” as part of a roundtable format. This event was planned for 1st May 2025, but a storm blew into windy Wellington and it got postponed until today.
I had an outline on how I was going to lead the conversation a month ago, im still following that outline tonight. Because the discussion is based on patterns and ways of working, not technologies.
If I was going to discuss technologies, in May I would be talking about AI Agents, in June I would be talking about MCP, and in July, well who knows.
The patterns I am going to highlight are:
AI Proof of Concepts (POC’s) are waste
Map your core business processes and then focus on a specific Node or Link
Focus on an Ask AI, Assisted AI or Automated AI pattern
Then build something that might go live (but probably won’t)
Start a 30 day GenAI fluency programme
AI Proof of Concepts (POC’s) are waste
I saw this meme on LinkedIn the other day.
I found this meme on this LinkedIn post https://www.linkedin.com/posts/colinhardie_ai-mlops-digitaltransformation-activity-7343558494018498560-DgE1 I don’t know who (or what) created it, so cant give them the credit sorry. But thank you to Colin for posting it.
It resonated with me as my observation is a lot of organisations are investing in Proof of Concepts (POC’s) as a way of trying to show value for their investment in “AI’.
IMHO its just busy work and typically waste.
Anecdotally I am seeing the majority of the value from those POC’s as being the revenue the large consulting companies are getting from running the POC’s for those organisations.
https://www.perplexity.ai/search/adf8ef03-0092-42fc-85b1-7d72c6db80a0
And of course its pretty easy to find articles that talk about the failure of these POC’s
https://www.perplexity.ai/search/9530b70c-3c5d-4ea8-8986-fe6fec53e3bf
But with any major step changes in technology its easy to find the “80% of all …. fails” nah sayer articles.
So what patterns do I recommend instead of the current goto POC’s.
First, map your Core Business Processes and then focus on a specific Node or Link
First go and watch this video.
”TED talk How to Make Toast”
Its both entertaining and educational.
It teaches you the core concepts of Systems Thinking, how to map out your organisations systems and processes as Nodes and Links.
Then map out a subset of your Core Business Process as these Nodes and Links.
Think of your Core Business Processes as a manufacturing plant. Anything that processes data, system, tool, logical chunk of code, is like a machine in the factory, its a Node. When data or work passes between those things its a Link.
(image generated by my virtual buddy ChatGPT)
Yes if this feels like the ghost of Business Process Mapping from the 80’s come back to haunt us, you are not wrong!
Don’t “boil the ocean” we are not doing a stocktake of all your organisations processes upfront, we are not returning to that 80’s behaviour of massive waste upfront.
Once you have one or a few Core Business Processes mapped end-to-end then put a dot on the 3 Nodes or Links that you think are the most broken.
And then put a dot on the 3 Nodes or Links you would spend money to refactor. \
The ones you would spend money on often end up not being the same Nodes and Links you identified as being the most broken
Second, focus on an Ask AI, Assisted AI or Automated AI pattern
Next think about which ”AI” pattern you want to experiment with for one of the Nodes or Links you identified as broken or worth investing in.
These are the 3 patterns that I use as a reference.
Ask AI
Assisted AI
Automated Ai
Ask AI
Ask AI is where a human asks a question and the machine replies, this is repeated until the human gets what they need, then the human goes and undertakes the data work.
This is the now typical “ChatGPT” pattern. The user starts a conversation, asks a question, gets a reply and asks the next question.
Lets use a familiar example to highlight these 3 patterns:
An example of AskAI is when you ask Siri on your iPhone to find you a cafe nearby, or when you search for cafe in Google Maps, or when you ask ChatGPT which cafes are nearby.
The machine finds the cafes near you and then gives you back that information. You might ask some more questions but eventually you, the human, makes the decision which cafe you will go to.
The key to this pattern is the human asks a question, gets some information, asks another question, gets more information and finally the human does the data work themselves.
The value of the pattern is it reduces the cognition required to do the work.
Assisted AI
Assisted AI, is where the machine helps the human do the data task that is required, without the need for the human to ask.
Following on with our example:
If you decide to go to a cafe that is some distance away, and you decide to drive, you might use Google Maps to help you figure out how to drive there.
Google Maps is tracking all the traffic in the area and at some stage it might suggest you change the route you are taking to get their faster. The machine is observing what you are doing and then making a suggestion on how to do it faster or better. You can accept that recommendation, that assistance, or you can ignore it.
The key to this pattern is the machine is making a recommendation which the human can accept or ignore during the process of doing the data work.
The value of the pattern is it reduces the time and the cognition required to do the work.
Automated AI
Automated AI, is where the machine does the work, the human is no longer involved.
Again based on our example:
If I was in a specific city in the USA I might not want to drive myself to the cafe so I could order a Waymo. The Waymo is an autonomous vehicle, the machine drives the car with no input from you, you just sit in the back and watch the car drive itself.
The key to this pattern is the machine is doing the work, the human is not involved at any step in the process.
The value of the pattern is it removes the need for the human to spend time doing the work and removes the need for the human to have the experience and the cognition required to do the work.
Another real life example of the 3 patterns
I recorded an AgileData podcast episode with Petr Pascenko at the beginning of 2025:
https://agiledata.io/podcast/agiledata-podcast/reliability-engineering-of-ai-agents-with-petr-pascenko/
In it he talked about using AI Agents to help with the Core Business Processes relating to Mortgage Contracts in a large financial services organisation.
An example of the AI patterns from that podcast (example has been generate by ChatGPT buddy based on the podcast transcript).
Ask AI
A mortgage analyst chats with AI ChatBot:
“Does this mortgage contract include a variable interest clause?”
“Does it mention early repayment penalties?”
“How do I calculate total interest over the first 5 years?”
AI responds with explanations and guidance on finding clauses or running calculations.
The human reads, interprets, and manually reviews the mortgage documents.
Value: Reduces cognitive load on legal nuance and definitions—but the human still parses the document and makes decisions.
Assisted AI
You upload a mortgage document into a smart platform. As you scroll, the AI:
Highlights clauses talking about variable interest or interest rate adjustment frequency.
Suggests adding a “balloon payment” check if certain terms are present.
Prompts: “You flagged rate variability—also check Section 7 for repayment terms?”
You confirm or ignore these prompts while reviewing.
Value: Cuts down time and effort by guiding the reviewer’s attention. You still drive the review, but with clever nudges and context-based insight.
Automated AI
You upload mortgage documents into an automated mortgage‑review system:
The system scans all contracts overnight.
Automatically identifies and flags any variable-rate clauses, early‑repayment penalties, missing statutory disclosures.
Compiles a report summarising issues and sends notifications to relevant stakeholders (lawyers, underwriters).
Optionally triggers next steps—e.g. “Send to legal for review,” or “Reject if statutory disclosure not found.”
No human involvement is required to detect issues or initiate actions.
Value: Saves time, removes human review effort entirely—perfect for high-volume or standardised mortgage checks.
Then build something that might go live (but probably won’t)
Once you have chosen what AI pattern you want to try and apply, and you have identified the Node or Link where that pattern may deliver organisational value, then you want to experiment with building something to see what happens.
The learning is almost as important as the doing
You don’t want to invest in this as a project where you will spend months and a large amount of time and money building something that may not actually solve a problem or add any value.
You also don’t want to spend a large amount of time and money with expensive consultants on “learning the art of the possible”. Who is getting the most learning them or you?
If you do just want to know whats possible then use this great free “AI” use case resource from Deloittes and save yourself some time and money https://www.deloitte.com/us/en/services/consulting/content/gen-ai-use-cases.html
But if what you build as part of the experiment has legs, you also don’t want to throw it away.
So you want to adopt the Minimum Viable Product (MVP) pattern from the Product Domain.
You want to build something small that might actually go live / make it to production in the first iteration, but probably wont.
It will probably be discarded as an idea that showed minimal value, or as an idea that has value and the build needs a second or third iteration to deliver that value.
So MVP over POC all day long.
Start a 30 day GenAI fluency programme
The last thing I would suggest you should do is create a 30 day GenAI fluency programme.
While you might be fluent in the difference between a “ChatBot” and an “AI Agent”, “ChatGPT” vs ”Gemini”, ”a language model” vs ”a reasoning model”, ”AI Agents” vs ”Agentic/Autonomous Agents”, ”RAG vs Hot Shots” etc most people aren’t and they probably don’t need to be.
But they do need some level of fluency in the language and opportunity the new “AI” domain brings.
So there is a useful pattern where you define a 30 day programme of 15 minute tasks an AI novice could do quickly and safely. And then get the people in your organisation to do then.
Things like, take a photo and upload it to ChatGpt and ask it what the photo contains, give Gemini three ingredients and ask it to find a recipe that uses those ingredients, use Perplexity and ask it to analysis your company and tell you how your company makes money and who the top three competitors are, pass a webpage of content to Google NoteBookLM and ask it to generate a 5 minute podcast episode based on it.
There are lots of great and free resources out there where people have published their versions of the 30 day list.
https://www.perplexity.ai/search/d20c8356-3292-4893-9255-5ddbc3f94ee7
Its easy to make the programme fun and engaging.
Simply Magical “AI” Patterns
These are just a few of the patterns I am regularly recommending to organisations and teams these days to help them with their entry and journey into the “AI” domain.
If you have any other patterns that you think are useful and valuable then post them here, or connect with me over on LinkedIn, I would love to understand them and add them to my pattern library.
https://www.linkedin.com/in/shagility/