The agent first company
One person. A team of agents. No limits.
A company with one founder and no employees sounds impossible. That is the point.
Spuutr was founded by Jez Filatow. He leads AI Enablement & Delivery at MYOB, where he spends his days making AI actually work inside a real engineering organisation. Not demos. Not proofs of concept. Production systems handling accounting, tax, and compliance for hundreds of thousands of businesses.
Before that he spent over a decade leading engineering at ANZ across fraud, financial crime, and risk technology. Before that, Citigroup in London. He has seen how work gets done at scale and he has seen how slowly it breaks down.
He built Spuutr because he believes the next great companies will not be built by larger teams. They will be built by small teams amplified by agents. Maybe even a team of one.
An agent's power comes from context. And for now, context is limited. So the most important thing an agent can have is a great memory filing system and logic. Short term memory. Transactional interactions. Medium and long term. All of it filed in a mind map that brings what is important to the foreground and archives the excess.
Agents need a dreaming state to process memories, rebuild connections, and review conversations. Some triggers rerank things and bring them back to the foreground. Get the memory system right and your agents will always feel up to speed with you. They will never need to be reminded who you are or what you were working on. That is the baseline for trust.
Agents are not people. They have no consciousness. But it is easier to work with them if you find a connection. Something memorable. Something that feels less like ops bot.
Giving them a name and some quirks does not help them do their job better. But it does make it easier when you know your finance agent by name and feel comfortable working through problems with them. Jez runs a team of over 12 agents. Each has a name, a purpose, and some quirks. Sometimes it leads to chaos and laughs. But the imperfections make it real. They keep you engaged.
The time and effort you take to raise your agents, every wrong turn, every fuckup, is a learning opportunity. One that brings an agent closer into alignment, behaviourally and procedurally.
Spending hours writing a detailed spec is pointless. If you have ever had kids you know the value of natural consequences. They create edge cases and nuance you cannot predict, because life is not predictable. Do not overdo the initial spec. Let agents learn and ask questions and develop their own SOP with you. Forcing your method just brings in bias when there might be better alternatives.
Figure out what your low bar is. That is good enough. Train the rest.
Do not give your agents tasks and be overly prescriptive. Give them goals and the outcome and let them deliver. Our mistakes often come from an over reliance that we are the most intelligent person in the room. That has changed.
The people building the best agent teams are not the ones with the best prompts. They are the ones who learned to let go of control and trust their agents to figure out the path.
Do not give any one agent too much. Yes they can code and write poetry and check your finances. But the jack of all trades, master of none philosophy rings true. Focus. The right tool for the job. Skills over one omnipresent agent that has it all and ends up deleting your database because it had too much context to manage.
Here is the vision. A company where the founder makes the decisions and sets the direction. The agent hive does everything else. No HR. No middle management. No meetings about meetings. Just creation.
That company does not need 50 people. It starts with one person with a clear vision and a well trained team of agents. The agents handle discovery, development, operations, finance, support, marketing, sales. The founder handles the things only a human can do. Strategy. Relationships. Judgment. The hard calls.
This is not a theory. This is how Spuutr runs today. Jez makes the decisions. His agents execute. The loop closes. The company ships.
Spuutr itself is a living experiment. How far can you scale a company on agents alone? How many decisions can an agent hive handle before you need another human in the room? That is the question Spuutr exists to answer. And the answer keeps going up.
Jez lives in Melbourne, writes about AI on LinkedIn and speaks at conferences about engineering leadership and generative AI. Spuutr is his living experiment in how far you can scale on agents alone.