I do not know what to do about AI agents.
There. I said it.
I run the workshops. I write the newsletter. I am the AI guy. I am supposed to know.
And most days I do. At least about today's version of this thing. But every Monday morning I open my laptop and the goalposts have moved. Something new has shipped. Something I was excited about last week now looks like a toy.
Some weeks I feel ten years ahead. Other weeks I feel three weeks behind.
Both are true at the same time.
If those of us inside this thing are still catching up, you are completely allowed to feel behind.
Now let me share what I have actually figured out, slowly, after a lot of staring.
What an AI agent is, in one sentence
An AI agent is software that works the way an LIC agent works.
Think about what an LIC agent actually does. They do not just give you advice. They fill the form. They submit it at head office. They follow up. They bring back your policy. They act on your behalf, end to end.
An AI agent does the same with software. You give it a goal. It opens the apps. It does the work. It comes back with the result.
That is the whole thing.
Not a chatbot. Not a robot. Not science fiction. Software that acts on your behalf.
If you remember nothing else from this post, remember that one line.
Three things that make it strange
An agent does three things no human employee can.
One. It learns in seconds.
Hand a new accountant your forty-seven-page SOP. Two weeks of reading. Three months before she is useful. Hand the same SOP to an AI agent. Two seconds later it has read all of it. Ready to work.
Two. It uses many apps at the same time.
You open Tally, then Gmail, then WhatsApp. One after another. An agent opens all three at the same moment and works in all three together.
Three. It can split into copies of itself.
Give Manish ten tasks. He does them one after another. Give an agent ten tasks. It splits into ten copies. All ten finish in the time of one.
This is what makes it a new category. Not faster software. Different software.
Why this matters more than it looks today
Imagine you are alive in 1780. James Watt has just improved the steam engine. It pumps water out of coal mines. Useful, but small.
Most people look at it and think, "Interesting machine. Not for me. I have horses."
Nobody at that moment can imagine that within a hundred years, the same technology will move ships across oceans, lay railways across continents, run factories that change every village in the world. Globalisation as we know it begins with that pump.
You cannot install a steam engine on a horse. The steam engine asks for an entirely new kind of vehicle. New tracks. New ports. New cities. The future it builds is invisible from where you are standing in 1780.
We are at that point with AI agents.
Today's agents do small things. Read your inbox. Summarise a document. Pull a number from Tally. Useful, but small.
What we cannot yet see is what happens when these small things scale.
Ten thousand agents working at once. Agents talking to other agents. Whole categories of work that no human did before because no human could afford to. New kinds of businesses that only become possible because thinking work has become almost free.
We are pumping water out of coal mines. The ocean liners and railways are coming.
Most days, even I do not know what shape they will take. But I know the first one is already on your laptop. Inside the ChatGPT or Claude tab you have opened many times. The same window, with tools turned on, is the first agent.
Why this is hard for everyone
I see other people online teaching the basics of AI. My first reaction is to dismiss it as too obvious. Then I remember: what feels obvious to me is completely new to most owners. My imagination has just run ahead because I am inside this every day.
The gap is real. And it is widening.
The pace is also faster than anyone is admitting. What was advanced a year ago is a free button on ChatGPT today. What feels advanced today will be a free button next year.
This is not a problem you solve in one weekend. It is a thing you sit with. A little more every week.
What to do this week
Almost nothing.
- Re-read the one-sentence definition above. Slowly.
- Tell one person, in your own words, what an AI agent is. Watch their reaction. That is your test.
- Make a quiet list of three tasks in your week that feel like junior-employee work. Reading. Comparing. Summarising. Save the list.
- Do not sign up for anything new.
That is enough for now.
What I will share next
The part where this gets useful. What an agent actually does for a small business in your city, in your kind of work. Not in theory. In your office.
I am still working some of this out for myself. You are working through it too. We are on the same line, just at different points.
Feeling behind is not a reason to stop walking.
