Every agent gets a state: Draft, Testing, Live. NextGen AI
NDS Logistics live at about $60K closed. 3 agent states in production. Cycle time 2 to 3 days across 8 sprints.
An agent is Draft, Testing, or Live, and the customer's admin decides which.
NextGen AI is an enterprise AI platform. Companies use it to run agents that do a business process for them. I joined on January 15, 2026 as the first product hire, and the brief was one line: we need the 2.0 version of the platform, make it more usable.
I spent January on analysis. I used Claude Code to read the codebase and Cursor to start the design system. In March I spent a week in Florida with the lead developer, working out how we would build together. In April I went to Albania to sit with the developers and co-lead the team with him.
Version 1.0 was a form. A customer filled in what they wanted the agent to do, the way you would fill in a Google Form, and then developers wired it up by hand in the backend. It was the same kind of work I did at my first UX job.
The customer got their change fast. The developers paid for it. Every new customer project pulled every developer off whatever they were building, and what they were building got forgotten. Two to three years of tickets had piled up that way, and nobody could say what should be built next.
The company was small and chaotic, and the idea of the product lived in everyone's head at once. Roles were not defined, so the same decision got made three times by three people. The industrial client's project was running the whole time and could not stop.
I was also on the sales calls. Whatever a customer asked for had to land in the roadmap the same week, or it went back into the developers' heads and out again.
The first thing I did was draw where the platform was going, not what it should look like. NextGen Studio: a workflow builder where anyone creates an agent by having a conversation with it. The agent asks you for what it needs, you answer, and the connectors are the only thing you wire by hand.
That first cut was high level and full of AI jargon. It showed the developers the direction, and that was its job.
The average person needs to be able to create an agent with the least intervention possible. The agent should ask them for the information it needs, and let them give it. Building an agent should not feel like patching cables on a modular synth. It should feel like the conversation they would have had with us.
To get the team there, I split it into two pods. Platform developers own the architecture, backend and frontend. Agent runtime developers build the agents our customers run and the APIs that reach outside the platform. We moved to two-week sprints, brought in an SDLC consultant, and cleaned up Jira so a ticket meant one thing.
The design system shipped first and the developers reference it now when they make changes. Agents carry a state: Draft, Testing, or Live, shown as a pill on the agent card.
In Draft you can talk to the agent but it reaches nothing outside. In Testing it can hit connectors, phone numbers, and APIs. The organization admin moves it between states, not a developer.
When NDS Logistics asked for a driver-hiring agent, we met on a Tuesday and they said yes to the designs on Thursday. That agent hires and manages their contracted drivers, from hiring to firing, and keeps their records and background checks current. It closed at about $60K.
Demos for companies that did not want agentic processes. Every one of those demos pulled developers onto work we did not need. I said we focus on one product, the industrial one, because those customers had walked us to the finish line of what they wanted and most other sales calls were unsure.
Letting developers be the only people who could put an agent live. Before the states existed, there was no testing, because the people flipping the switch were the people who built it.
Closed deals have slowed, and so has demo work. I am sure about not chasing clients who pull developers off the product, and that shows in the pipeline: about $60K closed and about $100K in pipeline today, down from the number the page used to claim. The industrial client I told the company to focus on took the product in-house.
Cycle time has held at two to three days across eight sprints and average velocity rose from 54 points to 68. Unplanned work is still a third of every sprint. The conversational Studio is still an idea in a room, not a screen. The biggest thing live today is the billing page, and Stripe is being built so customers can pay without a billing cycle.
The design system shipped first, and it is on its third version.
You describe the agent, and Prime builds it. In build since October 2026.
New customers signed up, landed on an empty dashboard, and never found agents. This is the Studio idea becoming screens, in three stages: today, the onboarding flow, then the simplified platform.
Working branches with sample data. No customer names. October 2026.
What shipped, and what is in build.
Shipped: the WCAG AA design system. Developers reference it when they make changes.
Shipped: the three agent states, Draft, Testing, and Live. The NDS Logistics driver-hiring agent runs on them.
Release 1.7 shipped in July 2026. A release every two weeks since. NDS is on the current version.
In build: Stripe. A second NDS agent is waiting on their budget sign-off.
NDS Logistics is named with permission. Other client workflows and revenue figures are sanitized.