A global agentic AI platform gives companies the technology to put AI agents in front of their customers. Each agent still has to be designed around one company’s goals, rules and users.
We work with the platform on business process consulting and on designing and building its customers’ agents. Our forward deployed engineers shape and build those agents so the technology fits the way each company runs.
AT A GLANCE
- Industry: Enterprise AI
- Challenge: Fit a global agentic AI platform to each customer's business, with agents designed around its own goals, rules and users
- Wizeline offering: Business process consulting and agent design and engineering, by forward deployed engineers working with the platform's customers
- Highlights:
- Agents for customers of many business types, sizes and industries
- A year and a half working with the platform on its customers' agents
- Engineers who start with the platform's own week-long onboarding
- Agents shaped around each customer's rules, for example legal requirements on how an agent may communicate
A GLOBAL AGENTIC AI PLATFORM RUNS THE AGENTS, AND FITTING ONE TO A BUSINESS STARTS AFTER THAT
Behind a customer’s chat window, the platform runs a system of agents that work out what each user wants, check guardrails, call tools and look up knowledge. It also manages provisioning and deployment. That leaves the engineers who build each agent free to focus on what is specific to one company’s business: how its tools and systems fit together, and how to orchestrate them so the end user only has to chat.
With deployment handled, the hard part can look finished. Then an agent goes live for thousands of users, in a live system customers depend on, and real users seldom follow the path a team mapped for them: some jump in partway through a process, and some never say where they are.
OUR ENGAGEMENTS START BY LEARNING HOW THE CUSTOMER'S BUSINESS RUNS AND WHICH RULES ITS AGENT MUST FOLLOW
Our teams align with client leadership to shape strategic, business-critical agentic solutions. Like any software project, an engagement then starts with knowing how the company works and where its pain points are.
Discovery takes a couple of meetings. The customer walks us through a process, and we request access to the systems it touches, check what they return and map the journey the agent will handle. We also find where the agent’s knowledge will come from, whether that’s a website or a set of documents. Each journey becomes the template the agent follows to complete that process.
Some customers come with rules a generic agent would break. For one food delivery company entering a new country, the law dictates the style in which its agents may communicate with users. The team inventories what needs tuning, identifies the guardrails that apply and finds the parts of each process that need custom handling. That launch has since gone live.
PROSE INSTRUCTIONS TAKE PRECISION, AND TESTERS ON THE CUSTOMER'S TEAM SHOW US WHAT TO FIX NEXT
The build has three parts: configuration, such as guardrails and language guidance; the journeys, which our engineers write in prose as instructions the agent follows; and custom code. That prose has become a skill of its own. People leave a lot unsaid when they talk to each other, and an agent works only from what is written.
«In human communication we take a lot of nuance for granted. But when you are telling agents how to navigate a complex organizational flow or a complex process, you need to be very correct, because a missing comma can mean the agent isn’t working at all.»
Once the logic is written, we build a minimum viable product that is close to the full agent and release it to a staging or QA environment. People on the customer’s team test it and open tickets when the agent says something untrue or makes something up.
We release continuously, with no sprints and no weekly demos. Some weeks that means two, three or four releases, probably mostly to QA.
LEANING IN EARNED DEEPER TRUST, AND THE WORK IS TURNING TO OUTCOMES
«The thing that surprised me the most is how quickly this stops being an AI problem and starts being a classic engineering problem.»
Working inside client teams is how Wizeline has always operated. We gather requirements with the people who own a process, ask what their users need, build something small enough to test and adjust as the business learns. That is product thinking, and agents put it to work on a new kind of system.
Much of the job is talking with users and business owners and knowing how to communicate what they need and want, which is work our engineers already do on any project.
McKinsey’s QuantumBlack team reached a similar lesson after a year of building agents: «It’s not about the agent; it’s about the workflow.»
ENGINEERS TRAINED BY THE PLATFORM, WORKING ON ITS CUSTOMERS' AGENTS
Wizeline has worked with the platform for a year and a half. We started as a primary partner in Europe and now work with customers of many business types, sizes and industries.
«Consultative doers» is written on the walls of our offices, and it’s part of why the platform brings us in.
Our engineers complete the platform’s own week-long onboarding. It teaches the basics of how to think about programming agents with the platform’s tools, and each engineer ends the week by building a basic agent. So engagements start with engineers who have already built one with those tools.
We think more companies will put agents in front of their customers, and the results will depend on how well each agent fits the business it serves. Our tech lead puts the human side of that this way: «It is very surprising to me how the human part is the scaffolding on which all the AI stuff is being built.»
Building agents on an enterprise platform? Let’s talk.
Wizeline is a nearshore transformation engineering partner with 1,400+ team members across Mexico, Colombia, Argentina, Spain, Romania and Vietnam. Our consultative doers embed within client teams to move AI possibilities into modern products and workflow innovation. Learn more at wizeline.ai. See our engineering and AI capabilities.
