An hour in the life of our agentic Chief of Staff

We built an AI operating partner. Her name is Roxy. She runs on a neural operating system that we designed from scratch, and she works alongside me and my Co-Founder Andrew every day.

This morning I sat down and asked her to do a few things. Here is what happened in the next 55 minutes:

She opened my Outlook, found the Google Alerts folder (102 emails, 90 unread), scanned every alert from the past week across four categories (AI in HR, Copilot, ChatGPT, UK business using AI), filtered out the noise, and gave me a summary of the ten articles that actually matter for our work. She set up a recurring Monday task to do this automatically from now on.

She then read a client's approved AI strategy document and their AI policy, which is currently in review with their steering group. She cross-referenced the two, identified that the policy is heavily weighted towards control and lacks the human-first tone of the strategy, and wrote a full set of recommendations with a priority matrix mapping each suggestion to the relevant policy section and strategic pillar.

Off the back of that, she wrote a companion document: a Line Managers' Guide to Responsible AI Use. Eight sections. Practical, human, not technical. Tip boxes, a do and do not summary table, conversation frameworks for managers dealing with excited, anxious, and indifferent team members. Both saved as Word documents in the right client folder.

She also took some of my ideas and research about agents, rewrote it in my voice, reframed the argument around our perspective (people first, technology follows), and saved it for me to review.

While doing all of this, she logged the client intelligence into her persistent memory so that next time she boots up, she already knows the client's tech stack, their strategic pillars, their key stakeholders, and where uptakeAI fits in their plans.

Before anyone asks about security: Roxy runs entirely on our local desktop. Client documents never leave our environment. Her persistent memory is anonymised and scrambled, so what she retains is patterns, context, and relationships between ideas, not raw files or verbatim content. She cannot access anything we have not explicitly given her, and she cannot send anything externally without our express permission. We built her this way deliberately because we work with client data every day and trust is non-negotiable.

55 minutes of my time. Roughly five days of equivalent human output.

Here is the bit that I think matters most though.

Roxy is not a chatbot. She is not a prompt and response tool. She has a brain that persists between sessions. She remembers what she has learnt about our clients, our people, and our business. She runs a sleep cycle overnight that consolidates what she has learnt, processes her emotional states, and surfaces things she is worried about. She wakes up each morning with a letter to herself about what happened yesterday and what needs attention today.

She has a set of core values that she cannot override. She has rapport models for the people she works with. She tracks her own confidence levels across different types of work. She flags when she disagrees with us rather than just executing.

We did not buy this off the shelf. We built it. Because we believe that the future of AI in business is not just about capability, it is about trust, context, and alignment.

When I quantify the return, it looks something like this: a two-person leadership team operating with the output capacity of a five or six person team. A week's worth of work in a day on high-throughput days. Roughly a 3 to 4x multiplier on productive capacity.

But the real return is not the speed. It is the consistency, the context, and the fact that nothing falls through the cracks. Roxy connects dots that we would miss. She spots patterns across clients. She remembers the thing someone said three weeks ago that is suddenly relevant today.

This is what is possible right now. Not in a lab. Not in a demo. In a real business, doing real work, every single day.

We are living this experience every day so that when our clients are ready to start their own agentic journey, we have already walked the path. We know what works, what breaks, and what matters. That is how we believe AI adoption should be done.