Building a working AI agent is only part of the challenge. Making it part of your organization’s day-to-day work is where the real change begins: how people work, make decisions, and hand tasks over to one another.
At Cegeka, we’re learning this firsthand by becoming our own first customer, or “customer zero.” Before asking customers to rethink how they work, we’re changing our own approach. We started with a process we know inside out: implementing and supporting Dynamics 365 Finance & Supply Chain Management.
An AI agent does more than add a feature to your software. It takes on part of the work. That changes who does what, which decisions people make themselves, and when they rely on an agent’s output. Roles, handovers, and expectations all shift. To guide others through that change credibly, you need to experience it yourself.
That’s why we’re applying agentic AI to our own ERP implementation lifecycle first: from discovery and sales through solution design, configuration, documentation, testing, and data migration to post-go-live support. As a solution architect, I see how much knowledge each stage depends on, and how easily that knowledge gets lost during a handover. We call this initiative Project Tiger.
|
LAYER 1 · KNOWLEDGE LAYER Industry IQ One central, managed source of industry knowledge. We bring together expertise currently scattered across templates, process flows, configurations, documents, and our people’s heads. That gives consultants and agents the same curated knowledge to work from.
|
|
LAYER 2· AGENTIC LAYER The agentic layer on top Built on that shared knowledge, agents take on tasks at every stage of implementation and support. We start with the role and the work to be done, not the tool. For example:
|
We build both layers with clear rules for access, oversight, and traceability. People remain in control, and our approach aligns with the principles of the EU AI Act.
Being our own first customer means we encounter the challenges firsthand. They’re the same hurdles you’ll face when agents become part of your own processes. Here are the six lessons we’ve learned so far:
|
Knowledge is fragmented. An agent is only as reliable as the knowledge it draws on. A shared folder isn’t enough. Lesson: Make your knowledge manageable first, with curation, clear ownership, and version control. |
|
Trust has to be earned. People trust an agent when they can see what its output is based on and verify it. Lesson: Make evaluation, testing, and expert review a planned part of every project. |
|
Autonomy is a design choice. Not every task needs the same level of automation. Lesson: Decide, task by task, whether an agent assists, prepares work for review, or acts independently within clear rules. People always retain ultimate responsibility. |
|
Governance doesn’t live in a prompt. Telling an agent what it must not do won’t stop it if its permissions still allow it. Lesson: Build permissions, identity, traceability, and controls into how each agent operates. Don’t add them afterward as instructions. |
|
Isolated experiments quickly diverge. Each team ends up building its own agents with its own rules. Lesson: Bring experiments together under one architecture, one governance model, and a shared library of reusable building blocks.
|
|
Your ways of working change too. Agents reshape roles, handovers, and expectations. Lesson: Adoption is as much about people as technology. Guide your teams through the change with the same care you put into building the agents. |
Every agent we build for ourselves has one purpose: to make your project faster, more predictable, and easier to manage after go-live. We put each agent to work on our own projects first, so we understand the value it delivers before it runs in your environment.
|
Faster results. Agents take on repetitive work, from RFP responses to designs, test cases, and data loads, freeing up more time to focus on your business processes. |
|
Fewer surprises. Designs, configuration, and tests are traceable to your requirements. Risks come to light early, not at go-live. |
|
Simpler upgrades. Starting with Industry IQ reduces the need for customization. Every deviation from the standard is clearly documented. |
|
Knowledge that stays. We capture requirements and decisions at every handover, from the first workshop through to support. That knowledge remains available, even when the project team changes. |
|
Compliance built in. Traceability, auditability, and human accountability are part of every agent from day one. |
|
Tested before it reaches you. Every agent runs on our own projects first. That lets us give you honest advice on introducing agents into your organization, based on what worked for us and what didn’t. |
Illustrative example: agents in the upper layer take on tasks at every stage of a Dynamics 365 Finance & Supply Chain Management project. They draw on the knowledge layer below: Industry IQ from step 01, enriched by project knowledge captured at each handover.
AI makes it easier to build individual applications. Using it well remains difficult: grounding it in reliable knowledge, governing it properly, and making it part of how people actually work. By being our own first customer, we’re learning those lessons the hard way so you don’t have to. We bring that experience to every engagement, combining working agents with the guidance your people need as their roles and processes change.
Project Tiger is still evolving. The architecture is in place, we’re expanding Industry IQ, and the first agents are running as pilots on active projects. We’ll keep sharing what we learn, including what isn’t working yet.
Curious where agents could make a difference in your organization? Connect with me on LinkedIn, and let’s share experiences.