๐๐ฟ๐ฎ๐ป๐ธ ๐๐ต๐ฒ ๐๐ถ๐ป๐ฎ๐ป๐ฐ๐ฒ ๐๐ด๐ฒ๐ป๐. ๐๐ฎ๐๐ต๐ ๐๐ต๐ฒ ๐๐ฟ๐ฒ๐ฎ๐๐ถ๐๐ฒ ๐๐ด๐ฒ๐ป๐. ๐ฆ๐ฎ๐บ ๐๐ต๐ฒ ๐ฆ๐ฎ๐น๐ฒ๐ ๐๐ป๐ฎ๐ฏ๐น๐ฒ๐บ๐ฒ๐ป๐ ๐๐ด๐ฒ๐ป๐. ๐ฅ๐ผ๐ฟ๐ ๐๐ต๐ฒ ๐ฅ๐ฒ๐๐ฒ๐ฎ๐ฟ๐ฐ๐ต ๐๐ด๐ฒ๐ป๐.ย
The world of agentic workforces are storming in to sight, but what exactly are they, and what are they not?
Hereโs a primer thatโll hopefully help you to break this all down and figure out how to make the most of them. After all as the saying goes, AI wonโt replace you, but humans backed by a team of AI agents probably will.
Iโve been building specialized agents for a while now, in order to help me plan serendipity in to my holidays, help me with my financial goals or help me to address chronic health issues with preventative behavioural changes instead of medicine prescriptions which is the only thing human doctors were giving me.
I also write a lot, and nowadays I write the seed of a weird idea in to a โwriting agentโ which is trained on my writing style and idea extrapolation, and riffs with me. I then use itโs output in different formats to provide myself with better ideas, and then I write the final draft up myself.
Up until this week, building an agent that does anything more than talk to you has required you to manually create workflows, connect APIs and use a string of different tools to tackle each part of the work flow. Errors and clumsiness has made the widespread creation of agents pretty difficult unless you either have a lot of money or technical expertise to hand. In my view, thatโs all turned on its head this week.
I think this is a seminal moment for agents - weโre seeing a number of incredible tools come to the fore which enable you to build visual workflows and orchestrate teams of agents that co-operate together to get things done.
What are agents? Specialists or generalists?
Agents are different to Gen AI tools, which generate content using intelligence. Agents use AI to get specific tasks done within the bounds of goals or objectives that have been set for it by a human or another agent. Does that sound kind of like an employee? Bingo.
Up until now Gen AI has been the ultimate generalist, trawling the entire internet to answer your question. Agents are specialists. Think of AI agents as hyper-specialized freelancers: they're task-driven, autonomous digital workers capable of performing specific actions independentlyโlike booking meetings, analyzing contracts, or running entire marketing campaigns. Unlike generalist AI (such as ChatGPT), which generates broad content based on massive datasets, AI agents are action-oriented + designed to execute precise tasks or sequences of tasks with ideally not much oversight. I for example can already see multiple agents I build, thinking through a problem together and giving each other feedback, but each doing their own separate tasks. Itโs a wild thing to watch!
So the difference between Gen AI and AI agents lies in breadth VS depth.
Generative AI = generates and synthesizes content, providing broad solutions across varied topics. Think of gen AI as the guy who just โgets it doneโ. The number two. The fixer guy who sorts the ambiguous problem out, but goes a bit rogue and doesnโt always listen to you.
AI agents = go deeper and narrower, completing specific workflows with specific guardrails and training, with precision, reasoning and autonomy. Think of it as the employee you interview, qualify for the role, onboard and train. Itโs your job to keep them on track and performing.
So whatโs next?
Will the future workplace will change as a result of this? Youโd better believe it will. Pre internet people were mainly full-time generalists (business people, farmers, shop owners etc) and then once the internet came about we moved towards being full time specialists (tech sales people, data scientists) who use technology and are highly competent at one specific task.
The next era shows up as a fluid and likely fractional workforce comprising both human specialists and AI agents, or potentially human generalists orchestrating specialist agents. Each agent can be optimized to be great in a specific job role, and made better with constant feedback loops. Very much like training a new hire to master one critical role exceptionally well. Sort of like you'd deploy a brilliant salesperson specifically trained to close enterprise deals, youโd similarly activate an AI agent finely tuned to optimize digital ad spend, do market research for you or dynamically adjust pricing models.
For now, agents can either be your assistant doing the jobs you hate, or your analyst helping you to get to your end point faster. Business leaders have been hiring people to do the jobs they hate for years, so this is nothing new.
But in the future weโll have teams of agents. Every human team will have an AI agent (or multiple) in big companies. In start ups, one person might run a business with 50 AI agents doing most of the work. And, youโll probably have an agent acting as your digital twin helping you with every task you do.
In this fractional future, the pain and risk of traditional hiring is gone โno recruitment delays, onboarding overhead, or realizing the person you hired isnโt right. Instead, you'll โonboardโ specialized digital expertise on-demand, dynamically scaling up or down depending on immediate business needs. Youโll train agents, but you wonโt have to tell them twice if you tell them properly once. Youโll be free of the emotional baggage that comes with large teams and focus on the optimal efficiency of the tasks at hand. Youโll focus your time on outputs, and itโll focus its time on problem solving.
The major risks I see in this area are similar to what weโve experienced with v1 of the internet - kind of like most of us have forgotten how to hand write or navigate themselves around a city without Google Maps, I see a world where lots of humans forget how to do basic things. I love thinking about the future, but Iโm also a bit old school; I handwrite all my tasks each day and I always turn off my phone and wander around when I arrive in a new city. My personal attitude is that itโs important to embrace new technologies, but I wouldnโt ask my employees to do something which I donโt even understand. So retaining a basic working understanding of the things youโre asking Agents to do is - for me - a critical part of the first wave of successful agent builders.
Why is it exploding now? Didnโt this already happen?
Up until very recently, LLMs have acted more like chatbots that you canโt really give boundaries to, nor keep them on a single trail of thought. Youโll have noticed if you use a big LLM that once it starts going off track itโs very difficult to get it back on track (usually you have to literally start a new session), and it also struggles to interact with the world around it properly or remember/stick to a pre-defined set of instructions.
Whatโs changed? Amidst a bunch of model improvements, the main one is - in technical terms - that AI is now incorporating MCP (Model Context Protocols). MCP tackles these issues and has a single instance of an AI LLM acting in a more consistent way, with stored context from your instructions or prior convos, feedback loops that make more sense and also the ability to interact with the internet around it instead of just scraping it for knowledge.
Because of the above, Agents are not just a one way dialogue from internet>AI like your favourite chatbot. Theyโre now able to follow rules, loop back on themselves, and have a two way conversation with the internet. These changes are pretty recent, and the difference is pretty big.
Humans + agents > humans or agents
We will (or should) all spend the next two years building personal agentic workflows that amplify our unique skill sets and delegate the manual and tedious tasks that take up your valuable time. It is pretty likely that the new world of agentic work forces wonโt require as many humans doing the same things as theyโre doing now, which means we as a species will be either finding things that AI cannot do, or orchestrating AI to make us 10x more productive.

In itโs simplest form, you have an AI agent trained on you, your personality and how you like to receive information. You can then ask it questions and hone it & discover from there where you want it to be better. This is different to current LLMs which answer your questions one by one in a conversation (although Open AIโs Chat GPT paid subscription has a lot of the features of a PA agent).

In this scenario, you have different โagentsโ that are particularly good at specific things you want, loaded with the information that will help it understand how to best help you (writing agent is loaded with what you like to write about and where specifically you need help, what your writing style is etc).

In this more complex scenario, youโve orchestrated a layer of agents to conduct value adding tasks like Sales, Marketing and Admin/Management. Each of these orchestration agents then โmanageโ sub agents which are highlly specialized in task handling. FYI the reason that youโd rather have sub agents than get one agent to do it all is that kind of like humans, AI agents get confused, get stuck in loops and are harder to train efficiently if youโre asking them to do tons of unrelated stuff with vague guidelines.
If thereโs two things you should take away from this post, itโs that:
Agents are coming, like it or not. If you use a laptop to do your work and avoid the topic of AI entirely, youโll likely be replaced pretty quickly unless you're the best in the world at your job or have skills AI can't yet replicate. But if you're reading this, you're right on time.
Agents are like employees that you onboard, give a job and course-correct them when theyโre having issues.
In the next series of posts, I will share some simple workflows, my favourite tools/guides on training agents, and some tips that everyone can use in order to get their heads in the Agent game. Iโll provide the first steps Iโd recommend everyone take to become literate, and Iโll also make myself available for anyone in my network that wants to discuss this topic or get any advice on how to get started. Stay tuned!

