janyl

A line drawing of a woman seen from behind at the near end of a long path, its two edges converging to a vanishing point on the horizon. A small upright screen showing two chat bubbles floats close beside her right shoulder, her agent. The horizon line runs edge to edge with a large empty space above it and nothing else in frame. Drawn in a uniform stroke with no fill and no shading.

Our method

We sell AI agents, but we don’t think of ourselves as a tech company. The value of our company lies in our method. It’s no secret what it is, but to do it well, you have to do it over and over. That’s what someone is buying when they build an agent with janyl.

The method starts with an audit, the first part of which is free and accessible from this site. We learn some basic things about your business to determine if and how we can help. Not every business will get value from an agent, and we’ll let you know if we think this is the case. As the audit moves on, we make a plan with your team about when to deploy your agent, for whom, and for what business domains.

The hallmark of a janyl deployment is that it starts with a single person on a single business problem. If an agent cannot provide value to an individual, it will never provide value to a team. Some recent examples from our own deployments are a scheduler at a dental practice who rebuilds the day every time a patient cancels, a warehouse supervisor who checks every delivery note against what actually came off the truck, and an analyst at a fund who reassembles the same position summary out of six spreadsheets every quarter.

A janyl deployment grows organically. As the single person reaches for more colleagues or more tasks, janyl’s agent moves naturally into these domains and towards these people, coding only what it needs as it learns. This part of the process is supervised by a Forward Deployment Engineer, who makes sure that the agent is rewarded for positive interactions and outcomes, that it never cuts corners in terms of security, and that the foundational code it writes for itself is solid.

After two to three weeks, the agent is integrated into the workflows of several team members and will be taking on more and more ambitious tasks. At this point, the focus of the FDE changes from helping the agent grow to helping the agent be creative and proactive. She or he focuses on the agent’s personality, how ambitious the agent is in the tasks it takes on by itself, and how the team reacts to it. Is it too verbose? Not talkative enough? Too theoretical? Too actionable? This is fine-tuned over the days that remain, and by the end of the twenty-four day window, the agent is a bona fide team member and the FDE is ready to step away.

Over time, agents need to adapt to new systems, new employees, new models, new business objectives, and bugs in its own harness. Many of these happen automatically as part of the agent’s internal maintenance cycle. But it will, from time to time, need expert help and advice, particularly around questions of model choice and newly disclosed security vulnerabilities. janyl is there for that as well - we have dedicated FDEs that are embedded with no organization and whose only mandate is to support existing deployments.

If you’ve worked in a larger organization, this process may sound radically compressed compared to other AI deployments. It is. It’s also less than 1/4 of the price. But don’t confuse speed for a lack of quality. Our philosophy is that starting small with an agent that grows into its role is not only cheaper and faster, it is also the path to having the most customized and most autonomous agent possible.

We’re proud of our method, and we’re even prouder of the great places it’s taken our clients. If you’re interested in becoming a new client, please don’t hesitate to reach out.