In my two previous reflections on artificial intelligence, I wrote…
May 2023 – How AI will Revolutionize Industries and Reshape the Workforce
SUMMARY:
- Entire industries will need less than 20% of their current workforce.
- Industries most impacted initially: accountants, lawyers, medical doctors, teachers, customer support, retail.
- Jobs that involve the transfer of information will be replaced completely.
- Least affected jobs: massage therapists, chiropractors, dentists, nurses, leadership, trade skills (plumbers, electricians, welders, ranchers).
Sept 2024 – A Post AI World
SUMMARY:
- AI will unleash a massive surge in global productivity for the next 20 years.
- AI will allow us to solve problems that are either too complex to solve without it or too small to warrant the human effort to solve.
- There will always be work for humans to do, but the nature of that work will morph over time.
- AI is a great problem solver (and efficiency tool), but it is not good at identifying which problems it should solve. The key function for us going forward therefore, is identifying problems worth solving.
2025 UPDATE
Near-term, AI evolution will look something like this:
General AI → AI internal adoption at companies → AI developed for a specific purpose or industry (Agentic AI) → Agentic Workflows → Agentic Systems
AI Internal Adoption at Individual Companies
Non-tech companies are still trying to figure out how to adopt artificial intelligence internally within their own organizations. Most have yet to implement anything, floundering, or barely scratching the surface of the productivity potential of AI. The primary use-case is for information retrieval and content production.1
What needs to be done at individual companies?
Companies have reams of data. But most companies do not structure their data, or if they do, only a small fraction of it, and it often lives within disparate spreadsheets housed across various corporate departments, not systematically collected and assembled into a cohesive relational database.2 And yet, there’s tremendous value in the data and potential to unlock vast intellectual property from the strategic insights AI might provide.
Historically, it has been cost prohibitive to assemble the data and draw meaningful trends, analysis and strategic insights, just for the sheer human hours required staring at spreadsheet pivot tables. But with AI layered on top of internal data, corporates can now see around the corner a little better and anticipate shifts in the marketplace, customer demands, productivity, etc.
The current challenge is that most companies have no idea what AI can do for them and no idea how to even begin implementing a comprehensive AI solution.3
AI Agents: Offloading Repetitive Tasks
AI Agents represent the near-term wave in the progression and adoption of AI. We will soon be inundated with tens of thousands of AI agents built to perform a single task or a series of tasks and to do it quickly. Imagine very specific AI Agents available, much like apps on the iPhone App Store.
AI Agents represent the lowest hanging fruit for AI adoption, the next adjacent possible.
Currently (mid-2025), AI Agents are being built for ultra-specific tasks. Some are job-function focused (finance, project management, engineering, accounting, human resources, software development, sales), some are industry specific (CPAs, doctor’s offices, retail, construction firms, law practices). You’ll see one for reconciling transactions in QuickBooks, another for drafting bespoke non-disclosure agreements, and a third for optimizing staffing schedules in dental clinics. These agents aim to do just one task, and to do it fast, accurately, and repeatedly… and without breaks, sleep, vacation, or set office hours.
Focused AI agents are easy to adopt. Companies just let the agent handle the repetitive work, consistently producing expected outputs based on provided inputs. Just plug these job-specific AI Agents into your existing process.
It’s narrow. It’s useful. And it’s here now.
Agentic Workflows: The Rise of Machine-Led Processes
Once these agents are embedded into your day-to-day operations, a natural question arises: what if one agent’s output could automatically feed into another agent’s input?
That’s when we move from single-purpose agents to Agentic Workflows.
We will then develop tools to string agents together to create agentic processes, where the output of one agent is the input of the next… but the logic won’t just be a linear process, it will follow an entire decision tree through the Agentic Workflow, such that information is processed through a labyrinth of pathways, depending on the nature of the inputs and outputs. In other words, the routing of the data through the Agentic Workflow is contextually adaptive, based on rule sets.
Here, agents are strung together into decision trees and logic branches that represent actual business processes. Think of a deal pipeline in a private equity firm: one agent pulls market data, another builds a financial model, a third flags anomalies, and a fourth creates a summary for partners. No human intervention is needed until something is worth their attention.
These workflows are more than just assembly lines. They’re dynamic. They route tasks based on conditions. They loop back when thresholds are met. They escalate when certainty drops. At this point, we’re not just automating tasks, we’re beginning to automate decisions and processes.
In short, AI Workflows behave more like an army of junior staffers than rigid code.
Agentic Systems
While individuals and companies are creating and implementing Agents and Agentic Workflows, others will be creating protocols to connect your company’s Agentic Workflows to the next company’s Agentic Workflows. Strung together, these collectively represent what I’ll call Agentic Systems.
Today, much of this work, information processing, and routing through a decision matrix is carried out by human operators, intervening in the flow. For processing speed, the human operators are the choke point, although still beneficial (perhaps required) for quality control… currently.
The real shift comes when your firm’s workflows start to plug into workflows at other companies. These Agentic Systems then start to become fully networked. Agentic Systems will provide networks of interoperable workflows spanning companies, departments, and tools. Data and decisions will flow between them like electricity through a power grid.
To make this work, we will need standardized protocols, common interfaces, and trusted identities (a key point). Initially, it might be a little clunky, much like surfing the early internet with the Mosaic browser. But it will streamline with time. It’s inevitable.
Related posts:
Follow Past Midway if you would like an email notification when I post something new.
- At this point, most people recognize AI writing. The prose is often a bit too generic and overly filled with the em dash character (the extra long hyphen). The tale-tell sign of ChatGPT written text is the em dash character with no space before or after.
This is a hyphen: –
This is an en dash: –
This is an em dash: —
ChatGPT loves the em dashes for some reason. In fact, I just asked ChatGPT about it, the first sentence of the reply, and I’m not making this up, “Good question — and you’re right to notice the em dash is a telltale sign of my style.”
- But I thought AI did not need structured data. In theory, it doesn’t. In practice, it does… at least for now… if you want good output to your input prompts.
- I developed a six-step roadmap to implement custom AI solutions at companies, but you’ll need to schedule a call with me to discuss this.
From someone (think OLD) who can barely work her phone, any comment might be laughable. But, I do have one. For one person/company/country to share with another, you have to overcome the human elements of, basically, selfishness, greed, & power. It will happen, but there’s a whole lot of human-ness (is that a word?) to overcome.
Another thought provoking blog to make a mama proud.
AI is certainly a hot topic at the moment. I’ve had many many conversations about it due to my field and the people I usually surround myself with.
I’m curious if you watched any videos leading up to this, specifically from Hank Green, Big Think, or WIRED before writing this article. Your summary section at the beginning is very similar to some of the ideas i’ve seen discussed on those channel, nearly verbatim to some of them. Shows how there are certainly similar thoughts going around in the space with those that are interested in the topic.
I’m curious also how you personally are defining “AI” as “Agentic AI” by your definition has really existed for quite some time in many areas but is now breaching into new industries and roles due to the development and popularization of LLMs and other GenAI models.
Would love to discuss with you some time and pick your brain. It’s certainly relevant to my life (really everyone’s lives) and my industry.