Tools that make AI transformative by default
The processes of work and the tools that support them are what limit the transformation.
We are in a process transformation unlike any in history. It is reshaping our work on the order of the move from farming to post-industrial life, but it will do so in 40 years rather than 100. Yes, 40 might seem slow. Change is happening so fast. That speed is just a sign of how big the change we are in really is.
The coming changes don't depend on model improvement. They depend on reworking how we work: who does what, how work moves between us, and how we judge whether it was done well. That is what we mean by process. The processes of work and the tools that support them are what limit the transformation. While it will take most of us 40 years to see this transformation through, some will transform faster. Craftmill aims to build the tools that let you be one of them.
Today, when AI transforms how a team works, it is because someone pushed hard for it or because they stumbled into it. Effort or accident. Nobody picks the default. The shape of the tool does. So we build tools whose shape makes the new way of working the place you land without trying. Loopway is the first. There will be more.
Here are the principles behind our tools
We build tools from the following 3 principles. They are connected, and together they are why Craftmill tools default you into transformation.
Bots > Agents
A bot is an agent with a name and a job to be done. The name doesn't have to be human. A role like dev-bot is a name. The job is a standing responsibility. When we call this a prompt, we undersell what it is. It's a responsibility, not the start of a conversation.
Prompts lead to a generic agent session that feels like an extension of us. We talk to it like a private DM. That has two costs. You can't delegate to something that feels like yourself, so you sit there and supervise every turn. And with no job there is no standard. The bot can't be good or bad at anything because it was never asked to do anything in particular.
Naming the bot and giving it a job flips this. The bot becomes an addition to our team rather than a multiplication of our personal work. Our tools don't let you skip this step. Every bot has a name and a job before it does anything.
Bots belong to the team
Sessions are the active memory of a bot. Today most sessions are tied to a single human, so our bots work in silos. Yes, they can read shared context, but the silo keeps them stuck at speeding up tasks rather than taking on teammate-like roles.
Part of the why here comes straight from the first point. When a session feels like a private DM, private things end up in it. People say what they are worried about. Sharing that with a teammate feels weird, so nobody does. Each person builds their own bots, and the company ends up with more bots than it needs and none that belong to anyone but the person who made them. Most underperform.
A bot that is a teammate belongs to the team. Its memory is shared because it is the bot's memory, not yours. Our tools build bots that way from the start. Anyone on the team can see what a bot did, talk to it, and change it.
Bot performance is not human performance
As bots do work we have historically done, we feel their performance is our performance. Part of the why here is us not naming bots, so it is unclear what is being evaluated. Part of the why here is that session silos make evaluation personal. Reviewing a bot's effectiveness feels like reviewing a human's, so we give it the same care: gently, in private, once or twice a year. That is right for people. Sadly, for a bot it means it never improves.
A bot is a system. It should be reviewed like one, on the outcomes of its job, with numbers rather than feelings. And the review should be a team ritual, like a retro, so nobody feels personally judged. Our tools make reviewing bots something the team does together, and something that is built in rather than something someone has to decide to do.
Name it. Give it a job. Let the team own it. Review it like a system. Do that by default and AI stops being a faster version of each of us and becomes part of the team.