I co-founded Adroit in April 2021 as Chief Operations Officer. My job was the standard operations playbook: campaign management, sales process and CRM, building every sales support material, running a budget, leading web developers and graphic designers. I developed marketing operations for 14 employees and contractors, and I wrote down processes for every single thing we did so eventually other people could take over the day-to-day work instead of it all being in my head.
Somewhere around year three, things stabilized. The processes I'd built started running on their own. Freedom didn't come first — the work came first.
Then late 2022 arrived and ChatGPT showed up. I adopted it immediately and built it into Adroit's services. By my own estimate, I had largely automated my job by about 2023. I was living off systems I built two years earlier. Around late 2024 my title changed to Chief Innovation Officer, which is the title I prefer now. Co-founder describes my relationship to Adroit accurately, but Chief Innovation Officer is the work I actually perform.
The shift wasn't a single moment. It was a progression: first the processes were written down so people could run them, then the written processes became what agents run.
How coding agents changed the work
The early AI experiments — ChatGPT drafting emails, generating research summaries, writing first-pass sales copy — compressed tasks but didn't change what the role was. I was still directing people and buying tools and validating output. The work stayed operations work.
The turn came when coding agents got good enough to take direction in plain English. Claude Code shipped in late February 2025. It could navigate repos, edit files, run commands and tests — the model plus tools plus a loop plus state, what would later be called a harness. About the same time, on 11 March, OpenAI launched the Responses API and Agents SDK, citing customers struggling with custom orchestration logic, and on 24 March Google Cloud's blog said "agentic orchestration has emerged as a new paradigm for LLM-powered applications to handle more complex tasks." The industry was converging on orchestration as its own architectural layer while I was learning to build inside it.
In March 2025 I started building LEGION, the multi-agent system I now run Adroit on. The vision was simple to say: I wanted to be able to have a discussion with the VP about strategic initiatives while the research, problem solving, asset acquisition and fulfillment were handled by the AI organization without further input from me.
It didn't work. I tried three different approaches across spring and summer 2025. The first version had AI APIs talking to each other on my computer, then became a web app. The second attempt used Docker, FastAPI and a LangChain task runner; agents lost work in wiped containers, edited the UI into black-on-black text, refused tasks, and ignored the handbook I'd written until I forced it into a default prompt. The third attempt was LEGION v2.0 in Lovable; the GitHub connection failed and I parked it.
Those failures were the problems the later harness writing would describe: state that doesn't survive a session, agents ignoring their instructions, agents editing things they shouldn't, the need for a handbook the agent reads by default.
I spent the rest of 2025 learning to work with coding agents properly — how to write effective briefs, when to validate, what gates matter, what the models could and couldn't hold — and the underlying models kept improving. LEGION came together in March 2026. On 10 September 2026, roughly 18 months after I first described it, LEGION was actualized and powerful enough to rebuild Adroit's entire website within a couple of hours.
What the role is now
Chief Innovation Officer describes what I do accurately. I lead technical architecture and AI orchestration at Adroit. The agent writes the code, the tests and the documentation; I direct, decide, buy things and validate.
LEGION is how I build everything now: Content Ops (the long-form engine that drafts from Adroit's approved knowledge base), Sadyr (the conversational AI widget answering visitors on adroitmktg.com from that same knowledge base), a game in development, and a set of personal tools including Cadence, the proofread-by-ear tool I built because I'm dyslexic and my brain is worst at the exact task of proofreading my own writing.
The work is infrastructure and orchestration. I design the harnesses — the context, tools, memory, loop, state and approval gates around the model — so the agents can do reliable work. I write handbooks the agents read by default. I build verification into the system so I can trust what comes out of it: Content Ops runs 839 TypeScript tests and 57 Python tests; Sadyr runs over 1,300 tests; LEGION itself is built on 64 capability packages. The systems run on Adroit's own approved knowledge base, they show their sources, and I validate every output before it ships.
December 2024 I built an AI prompt-engineering curriculum for the Adroit team. The company runs on these systems now, and teaching people to work effectively with agents is part of the role.
AI is positioned against cost and drag, never against people. The work compresses time — blueprint to production in twelve days for Sadyr, 92 of 95 planned pages drafted in two days of batch runs for the website rebuild — but it doesn't replace the team. The 14 employees and contractors I worked with in the operations years were doing different work; the AI handles the tasks I used to do myself after I'd written them down.
Why the title matters
The title change happened around late 2024, after years of AI experimentation and altering our workflows through innovation. Chief Innovation Officer is accurate because the work is building new systems, testing what's possible at the frontier, and integrating those systems into how Adroit operates.
It's the clearest description of what I spend my time on: designing agent harnesses, writing capability packages, validating output, teaching the team to work with the tools, and showing the receipts so other builders can see what actually works and what doesn't.
Operations was writing the company down so it could run without me. Innovation is building the systems that run it.