Why I am building this way
In my article about starting again with an AI workforce, I described the connection to my years building autonomous robots. The work taught me to think about perception, memory, feedback, boundaries and human control as parts of one system. That experience shapes how I approach business agents today.
A founder still sets direction and remains accountable. Specialised agents can support research, communication, administration and complete workflows. The next organisation has to prove itself through its work. I use my own projects as a place to build, learn and improve the operating method before asking another company to imagine its own version.
Show a completed piece of work
A good demonstration follows a recognisable business problem. The agent receives an input, uses the relevant information, creates something useful and shows what happened. People should be able to see the result and understand the decisions that remain theirs. That makes a demo concrete enough to start a commercial conversation.
The next question is more demanding: can this work again tomorrow, with a different input and without somebody remembering to write the perfect prompt? That is where my focus has moved. An Agentic Enterprise needs a coordinated system of work. Agents are part of that system, alongside triggers, shared knowledge, tools, checks and the people accountable for the outcome.
Connect the loop
Consider a proposed software maintenance loop. User feedback becomes a queue of bugs and ideas. Agents can organise the reports, prepare a reproducible case, propose a change and run checks. A responsible person decides whether a feature belongs in the product and whether a release is ready. Approved changes can then feed distribution and communication, with new feedback returning to the queue.
Every step needs a clear completion condition and a next owner. A test failure should produce a useful recovery task. Missing information should create a question. An approval should stay pending until the responsible person decides. Continuous work comes from these explicit transitions, rather than from leaving a chat window open.
24/7 is an operating design
Around-the-clock capacity is useful when there is worthwhile work and the system has authority to perform it. Recurring triggers, durable queues and recovery behaviour keep work moving. Budgets, permissions and escalation rules set the boundaries. A dashboard should distinguish active work, waiting, blocked tasks and verified results.
The Agentic Twin makes that organisation visible. You can see projects and roles and move into the work behind them. It is a way to understand and direct an operating system. The proof remains in completed tasks and their evidence. I want a leader to leave a demonstration understanding which work moved forward and where a human decision was needed.
Pick the first loop for its business value
Choose work that repeats, has usable inputs and produces an outcome somebody cares about. Establish the current effort and delay before adding automation. Start within a bounded environment, learn from exceptions and widen the scope deliberately. Model choice, hardware and integrations should support that workflow; commercial and local open-source models can each have a place depending on the task and constraints.
My role spans that whole journey: strategic choices, infrastructure advice, hands-on implementation and coaching people to work with agents. The exciting moment opens the door. A dependable process gives the business a reason to keep it open. If you want to explore this, bring one recurring process to a conversation. We can make the next demo about work your company recognises.