Start Date: ASAP but open to discussion
Type: Part-Time
Location: Remote (US-based, NYC hybrid)
About Usde Anda Capital is an independent sponsor and family office with over 140 deals and over 3 billion USD in investor value creation. We are based in Los Angeles, California and New York City, New York. Our main office is located in New York City. We acquire, operate and consolidate businesses in property management and real estate services. We work across several verticals, including property management, payroll and payments, decarbonization, manufacturing, and lending and financing. About de Anda Labsde Anda Labs is the our internal development function. Labs is organized into two functions: The first function is internal, and the framing we use for it is that Labs exists to create leverage for the rest of the firm, which means its output is measured by how much faster and how much more reliably the analysts, the deal teams, and the principal can do their work. The second function is client-facing, and it is the work of building agents and tools as products for the companies we advise and the companies we own, where the user is outside our walls and the tool has to survive without the builder sitting next to it. The Farm Program Structure: refer to https://recruitment.deanda.capital/ for further detail Through traditional finance paths, most professionals reach partner in 10–12+ years. Through the Farm Program, you can become a Junior Venture Partner in about two to three. We will accelerate your career by eight to ten years in only three to four years. We provide young professionals with the opportunity for 7-figures in fees and equity based on performance, rather than adhering to arbitrary timelines. We empower young professionals with responsibility, experience across multiple verticals, and direct mentorship. Phase 1: 1-6 months as an internship, and you will be doing real, important work from day 1. Phase 2: 6-9 months as a Junior Engineer, paid hourly.
The position:
The role sits inside de Anda Labs and builds AI agents that take over the repeatable parts of how the firm works, and, in time, agents that we deliver to clients as products.
The intern works alongside two other developer interns, who currently cover web and CRM connectors on one side and workflow automation on the other.
This is a build role rather than a research role, and the expectation is working tools shipped on a weekly cadence.
Everything built here is used by people who did not write it and who will not read the code, so the written specification and the handover are part of the deliverable rather than an afterthought.
Early work is internal, and client-facing work follows once the quality of the internal work has earned it.
How the work begins
Before anything is built, the team runs a first-principles test on a process, and this is the part of the job that a candidate should understand before applying because it governs everything downstream. We take a standard operating procedure that someone at the firm performs, whether that is assembling a diligence checklist, reconciling a rent roll, drafting a first-pass memo, or updating a board, and we break it into the individual judgments and the individual mechanical steps it is actually made of. We then ask, step by step, which of those steps require a person and which of them only appear to.
The output of that session is a short list of steps that an agent can own, and only then does the building start. The reason we work this way is specific rather than a matter of preference. Our experience with prior intern cohorts has been that people who build with AI tools can produce working code faster than they can define the problem the code is meant to solve, and development that starts without a clear definition of the problem and the end product produces tools that no one adopts. The first-principles session is how we avoid that, and the Development Intern is expected to participate in it as a full contributor rather than to wait for a specification to arrive.
Responsibilities:
Participate in first-principles sessions with the team, in which an existing standard operating procedure is decomposed into its individual steps and the steps that an agent can own are identified and agreed on. Write the product requirements document for each agent before building it, covering the problem, the user, the inputs and outputs, the failure modes, and the definition of done. A short and honest document is worth more to us than a long and aspirational one. Build the agent, using Claude Code, Codex, and whichever of the mainstream agent and vibe-coding tools fits the job, and connect it to the systems the firm already runs on, which currently include Excel, Google Workspace, Framer, Pipedream, and the firm’s CRM.Test the agent against real work rather than against invented cases, sitting with the analyst or the associate whose process it is meant to absorb and watching where it breaks. Write the standard operating procedure for the agent once it works, so that someone who has never met the builder can run it, and revise that document when the agent changes. Maintain the task queue in the Dev Build Master spreadsheet, keeping status, scope, and blockers current, because the queue is how Labs reports to the rest of the firm. Hand off finished work, which means a short demonstration to the team that will use the agent and a written record of what it does and does not do. On client engagements the same handover is made to the client, and the documentation has to stand on its own because we will not always be in the room when a question comes up.
Required qualifications:
We care considerably more about what a candidate has actually built than about where they are enrolled, but the following are the things we do ask for. Current enrollment at a four-year institution in computer science, software engineering, data science, or a related technical field, or graduation from such a program within roughly the past year. Working coding ability in at least one general-purpose language, with Python or JavaScript being the most immediately useful here, at a level where the candidate can read unfamiliar code and modify it rather than only write their own from scratch. Demonstrated proficiency with Claude Code, Codex, or comparable agentic coding environments, and with the mainstream vibe-coding tools more broadly. We will ask what the candidate has shipped with them and what they learned about where those tools fail. Experience building agents rather than only building applications, meaning tools that take an instruction, plan, call other systems, and return a result, along with some considered view on how to keep such a tool from behaving unpredictably. Experience writing product development standard operating procedures and product requirements documents. We ask for a writing sample on this point, and the sample matters as much as the code sample does. The ability to write clearly in full sentences, because roughly half of what this role produces is prose that other people at the firm have to act on.
Preferred qualifications:
None of the following is required, and a candidate should apply without them.
Familiarity with workflow automation platforms such as Pipedream, Zapier, or n8n, and with connecting APIs that were not designed to work together.
Some exposure to finance, private equity, real estate, or professional services, sufficient to understand why a rent roll or a quality of earnings review is structured the way it is.
Front-end work in React, Next.js, or Framer, since a number of the agents we build need an interface that a non-technical person will actually open.
Experience running or contributing to evaluations of an AI system, which is to say some discipline around measuring whether a change made the tool better rather than assuming that it did.
