The first wave of agent adoption and creation inside Arinco was fast and loose, but this was by design.
The technology was evolving at unprecedented rate, and our crewies were rapidly building agents for themselves, their teams and customer engagements. Someone would build something useful, share it with colleagues, and others would take the idea further. There was no grand operating model behind it. We were learning what the technology could do and where it was genuinely useful, and we wanted the freedom to adapt as quickly as the technology was changing.
This effectively made us our own Customer Zero. In Microsoft language, Customer Zero means applying new technology and ways of working inside your own organisation first, learning from that experience, then taking those lessons to customers. For us, that has meant working through the same shift many of our customers are now making as they become Frontier Firms: changing how work gets done across people, processes and agents.
Eventually, informal sharing could no longer carry the volume of experimentation. We needed a way to capture what our crewies were creating, make good ideas visible, improve them and apply the right level of governance as they moved into wider use. At the same time, the technology and approaches were starting to mature. We were seeing clearer patterns emerge around how agents could be built and used effectively.
That is when we built Arinco AgentHQ. A few lessons have become clear since then.
Give experimentation somewhere to grow
Arinco has always had a strong culture of internal knowledge sharing. As agent technology became more accessible, that naturally extended into agents and skills.
By early 2026, we were seeing new agents, ideas and skills in our technology sharing sessions almost every week.
Arinco AgentHQ became the place to bring that activity together. It now includes more than 50 agents and skills spanning internal operations and customer delivery, including cloud and data migration, assessments, security operations, commercial review, change management and solution scoping. We encourage crewies to contribute early-stage and pilot agents as well as more mature capabilities because the visibility helps others see what is being built, learn from it and extend it.
Our model is broadly capture, nurture, govern and scale.
An agent might begin as something one person builds for a specific task. If it proves useful and repeatable, we can improve it, test it more thoroughly and apply stronger controls as its use expands.
Agents progressing towards “certified” status need to meet higher expectations around quality, security, deployment, observability and governance before they can be used more broadly across Arinco or in customer delivery. Some agents remain useful pilots. Some develop into trusted delivery assets.
Crew Mate shows how individual know-how can become shared capability
Crew Mate is one of the clearest examples of that progression. It began with one of our consultants, Lukas Schauffele, building something for himself to reduce the administration involved in running consulting engagements. It has since developed into an agent-assisted project delivery capability that we are rolling out more broadly across Arinco.
Crew Mate combines a structured project workspace with an agent that can work with the same project information. It can help consultants maintain project records, manage risks and actions, update plans, prepare status reports and reconcile delivery activity. It can even do your timesheets, which I know for many consultants, myself included, is life changing!
As its use has expanded, we have added more guardrails, integrations and controls. A capability designed for one person requires a different level of rigour once it starts supporting a wider consulting team. It is also starting to embed our Done Right™ delivery approach into the way project work is managed.
That is the pattern we want AgentHQ to support: a crewie identifies a useful problem, builds something that works, and the strongest ideas have a pathway into broader use.
Agents are emerging across the business
One of the biggest surprises has been the breadth of agent use. I initially expected most of the activity to come from developers and our more code-heavy practices. Instead, agents have emerged across technical and non-technical teams.
Our Intelligent Workplace teams have built agents at a similar pace to more development-focused teams. Other examples have supported strategy and assessment work, test-case development, data engineering and even application packaging!
Seeing what others have built through AgentHQ has helped crewies recognise more opportunities in their own work. They are increasingly looking at repeatable parts of their work and considering where an agent could support them. Sometimes that means reusing something that already exists. Sometimes it sparks a completely new idea.
That change in behaviour is an important part of becoming a Frontier Firm.
Measure agents against real work
The clearest evidence comes from what agents are doing in practice.
Our Agentic Security Operations Analyst has reduced security investigation times by more than 90%, taking workflows that previously required up to 40 minutes to under two minutes. False-positive alert noise has also been reduced by up to 87% in some scenarios.
Arinco AgentFlow has been created for customer delivery to support GitHub pipeline migrations, a traditionally labour-intensive activity. In one example engagement, it reduced the delivery effort required by 50%.
Another agent designed to improve the context and performance of other agents has been used across three customer engagement, showing improved agent accuracy by 68%.
Crew Mate addresses a different part of the business again, supporting the administration and governance surrounding consulting delivery.
Across these examples, agents take on repeatable work while our crewies remain responsible for direction, context, judgement and outcomes. That is how we see agentic delivery developing.
Increase governance as agents become more consequential
We have also learned that governance needs to grow with the role an agent plays. A crewie experimenting with an agent in a controlled internal environment has different requirements from an agent supporting a production workflow or accessing customer systems. As an agent gains greater access, supports more users or becomes part of a more important workflow, our expectations increase.
We look at its purpose, quality, authentication, permissions, security, deployment, observability and where human accountability sits. That gives crewies room to experiment while creating a clear pathway towards production-ready capability.
It also means we can stop developing an agent when the value is not there. Every experiment does not need to become part of the operating model.
Teach the mental model early
One of the clearest lessons from our own adoption has been the importance of education. Crewies need to understand what an agent is, the kinds of work it can support, how it differs from a prompt or traditional automation, and what tools are available. Once that mental model starts to click, they become much better at identifying useful applications in their own work.
That learning has also informed Frontier Ready, our internal AI skilling and adoption program. Across the first seven sessions, it reached more than 80% of our workforce, with average self-reported confidence increasing by around 23% across the six sessions where it was measured.
Frontier Ready helps build the understanding. AgentHQ gives crewies somewhere to apply, share and develop what they learn.
Expect a learning curve
I often describe the early stage of adoption as a negative productivity cycle.
People see the potential quickly. Then they start encountering inconsistent outputs, quality issues and situations where fixing the result can take longer than doing the work without an AI assistant. Getting through that stage requires people to learn the tools, change established habits and develop a different way of approaching work. If everyone is left to work through it themselves, that period can drag on long enough for people to lose confidence in the technology.
Structured training, practical examples, peer learning and safe environments for experimentation can shorten that cycle and help people reach the point where productivity starts to improve more consistently. We have been through that ourselves, which has changed the way we think about adoption with customers.
Success is when this becomes normal
Arinco AgentHQ now includes more than 50 agents and skills. I’m less interested in whether that becomes 100, 500 or 1,000. The stronger measure is whether useful agents become part of the normal way we work.
Crew Mate is moving in that direction. The Agentic Security Operations Analyst is another example. Over time, I expect a relatively small number of agents to become deeply embedded into important internal and customer workflows.
That takes us back to where we started.
The first wave of agent adoption inside Arinco was unstructured because our crewies were experimenting, learning and solving problems close to the work. Arinco AgentHQ gave that experimentation somewhere to grow, while giving us a way to add structure and governance as the strongest ideas matured. Being our own Customer Zero has given us a clearer view of what works, where more structure helps and what is not worth taking further.
We are still learning. The technology will keep changing, and so will the way our crewies use it. I think staying close to that experimentation is one of the most useful things we can do.
If you’re working through the same questions in your organisation, we’d love to chat about what is and isn’t working for you.
