Consultant Turned Builder.

Autonomous cloud-based agents that do delivery work on live engagements. They run on Anthropic’s Managed Agents platform, on schedules, through an orchestration and state layer I built around it. Below are three examples of them, each built for a different domain:

Leading the rollout of Claude Cowork across a 1,200-person practice, the first team at the firm cleared to pilot it, in direct partnership with Anthropic. The tactical rollout work is one half of it; the other is moving 1,200 people to new ways of working so they get the most out of the technology.

This is AI work delivered at the client: working sessions with their teams, and building alongside them in the tools they already use. The example below combines several real engagements into one, with client names and details removed.

Decks are the core deliverable in consulting, which is why a generator was worth building. I built it in early 2025, when models could not produce a presentable deck from a plain prompt: the system carried the slide structure and layout rules the models did not yet have. Newer models manage a decent deck without that steering; this one worked before they could.

An application I built for the first read on a deal. When a company sells off part of its business (a carve-out), the work starts with reading the deal documents to judge how hard the separation will be, traditionally days of work before AI. This tool does that first read: the documents go in, and out comes a list of red flags grouped by category and a rating of how complex each part of the business is to separate. In the background, a vector database captures the feedback users give as they question and correct each assessment. That gap, between what the model says on its own and where the practice’s judgment lands, is the raw material for tuning the system over time.