People who excel here have something in common: ownership.
We're a small team building custom software and AI-powered workflows for mid-market operations teams. You work directly with clients and co-founders, and you see firsthand whether your work landed. Pods are two to four people, and you'll sit on a few at once. When a build slips, you're the one who tells the client, with your pod behind you.
That suits some people enormously and it suits some people not at all. This page is here so you can tell which one you are before you apply.
One shared canvas, start to live. In AI Jumpstart the client and the strategist put the real problem on the wall, the strategist drafts the prototype with Claude, and the client says yes. In Deployment Studio the project manager sequences the work, the developer builds it, it goes live and the client's team uses it. The same pod stays on it throughout, and you could be any of the three roles.
You've run an operation from the inside. You know the piecemeal systems, gaps in data, and manual steps slowing down processes. You've wanted to fix it properly and never had the mandate. Here, fixing it properly is the job.
You can run a room, synthesize fast, and earn trust with people who've heard a lot of consultants promise a lot of things. Here, strategy runs through building, not decks. You lead discovery, you design the prototype, and through go-live you're the one connecting the client, the developers and the product.
You already use AI to prototype and test ideas before anything gets built. You want more range than one product gives you: new clients, new problems, faster turns. Here, you design the right work first, then the right feature.
You'd rather understand why the invoice approval takes nine days than pick up the next ticket. Here, you're part of shaping what gets built, and when a client asks why we can't just add one more field, you're often the person who explains what it costs. You build knowing you won't be there to maintain it: the client can operate it, and the next developer can extend it.
We hire, develop and promote against four competencies. Every role expresses all four. What changes from role to role is how they show up.
What strong looks like, role by role. Strategist: frame the real problem before reaching for a solution; see where work, time and money leak; know which AI use cases fit the client's workflows. Project manager: turn strategy into clear, buildable, sequenced work; talk trade-offs in time, cost, scope and quality; know where a human still has to check the AI. Developer: write code the next person can extend; explain what one more field costs; ship nothing the AI wrote without reading it. High agency is the same for every role: bring the problem, its cause, the options and your pick.
On the strategy side: understand what people mean beyond what they said in the interview, frame the real problem before reaching for a solution, and land designs that are clean and simple.
On the delivery side: turn that into clear, buildable, sequenced work, hold acceptance criteria, keep the board and reality the same thing, and re-baseline calmly when the ground moves.
On the build side: architecture you can defend, tests that mean something, and code structured so the next person can extend it.
Diagnose how a business runs and where work, time, talent and money leak. Reason about ROI and trade-offs, and talk in time, cost, scope and quality rather than in tasks. Clients trust us because we get their specific problem and have an opinion about it.
How we work: run projects through our Jumpstart and Studio repos, and when you adapt the harness you create a reusable skill for the team. You own the outcome of whatever AI produces, which means no slop and no unread output going to a client.
How we show up: you're the credible person in the room on what AI can actually do, which parts still need a human check, and when the thing they're asking for isn't the best answer.
Own the outcome, not the task. More on this below.
We run one delivery model, in two phases, and everyone works inside it.
AI Jumpstart is the entry point with every client. Discovery, then a working prototype and a set of recommendations, fast. The prototype exists to prove something and act as a decision tool.
Deployment Studio turns that prototype into working software, then leads implementation, training and adoption.
Adoption. The build is done when people use it.
Both phases run through a set of Claude repos we call the harness, supported by playbooks and SOPs. You don't invent your process from scratch, and if a playbook step doesn't fit your project, you change it and tell the team. Improvements from a retro go back into the playbook.
What this means for your week. You're in a pod with a strategist, a project manager and a developer, in client rooms. You'll ship something a client reacts to within weeks, not quarters. And you'll use Claude in every phase.
When you hit a problem, there are five ways to bring it to someone.
“There is a problem.” Then you walk away and someone else deals with it.
“There is a problem, and here are some possible causes.”
“Here is the problem, here are the causes, here are some options.”
“Here is the problem, here is what caused it, here are the options, and here is the one I think we should pick.”
“I found a problem, worked out what caused it, fixed it. Keeping you in the loop.”
Expected on work you ownLevel 4 is our floor. Level 5 is expected on work you own.
It's part of how we develop and evaluate everyone, at every level. Level 5 means acting without asking permission, so we have to be the kind of place where that's safe.
Break down a messy situation and make a sound call when the information is incomplete. Say when you don't know.
Try the thing, find out early that you were wrong, and say so. When the next step is undefined, pick one, write down the assumption it rests on, and start.
You're on the same side as the client. Absorb client anxiety instead of amplifying it into the pod. When you disagree, do it with partnership. When your approach isn't landing, change tack. If an engagement is going off track, turn it around, rather than asking to move off it.
Assume the person who dropped something had a reason, and ask before you conclude. You keep people posted, so nobody has to chase you for status. Nothing reaches a co-founder from a client before the team. When you're frustrated, name what happened and what you need without making it personal.
What happens after feedback, what people learn from working with you, and what you build that nobody asked you for. Feedback you get once shows up in your next piece of work, and you ask for it before it's offered.
You want the scope handed to you. The plan gets written by the people doing the work, inside our playbooks, rather than handed down by a separate project office.
You'd rather be right than land it. A technically perfect system nobody uses doesn't count as done here. If the group isn't adopting, the first question is what the product got wrong, not whether they need another training session.
You want AI to stay a side tool. Reaching for Claude first is how we deliver, in every role.
You're happier when conflict is avoided. We disagree in the open with tact and respect, including with clients and with the co-founders. If your instinct in a tough meeting is to smooth it over and move on, this will take some getting used to.