Building a conversational AI agent has always meant more than designing good prompts. It means orchestration: process guides, node transitions, data fields, all wired by hand to handle a specific customer intent. That work is precise, necessary, and slow. It’s also the first place the interface shift we’ve been talking about is showing up inside our own engineering team.
Orchestration ate the day
Every new capability an agent needed, a new API, a new use case, a new data field, meant an engineer manually building the orchestration behind it. The model’s reasoning wasn’t the bottleneck. The translation of that reasoning into a working, tested conversation flow was. Adding something as contained as flight check-in could take the better part of a day, not because the problem was hard, but because building it was mechanical.
Giving the system context, not just instructions
We started giving agentic coding tools direct access to the same material our engineers use: agent definition files, the documentation behind how we build agents, source data, and the capabilities already in production. With that context, the coding agent can write the necessary orchestration itself and run it against our existing simulation suite before anything ships. The engineer’s job shifts from building the flow to specifying the outcome and verifying the result.
A day of work now in twenty minutes
One of our AI engineers, John Andersen, used this approach to add check-in functionality to an airline agent. What used to take the better part of a day took about 20 minutes, evals included, with nothing in the existing agent broken. The work didn’t get easier. It moved up a level, from hand-building orchestration to directing and checking it.
Building the future we’re selling
This matters beyond our own velocity. If we expect a future in which enterprises run their operations through AI rather than around it, we have to build that way ourselves first. The way we’re changing how we build is the same shift we’re building toward for our customers: a future where teams operate through AI.
