For AI startups selling into the enterprise
Most applications describe the work.
Mine is running.
It sorts each message, scores how good a fit the work is, and drafts a reply — then stops, because it is not allowed to send anything. A person decides. That restriction is the interesting part, not a limitation. It is the same discipline I would bring to your customers: prove it works on real cases first, publish the failures, and let it earn responsibility one step at a time rather than all at once.
Why this role exists
Your model is fine. The last mile isn't.
You have customers who bought the thing and still aren't getting value from it, and it is almost never the model's fault. The value is on the far side of somebody's twenty-year-old process — the one with forty senders, nine formats, and a person called Marta who knows all the exceptions.
Closing that distance needs someone who can sit in the room with your customer, work out how the job is really done, and then go and build it. The second half is where most candidates fall over. Plenty can run the workshop. Fewer can ship the thing that comes out of it and then prove it works.
The offer
What I'd do in your first 90 days.
Written as a proposal rather than a promise — the specifics would change once I'd met the customer. The shape wouldn't.
Sit with the customer, ship one thing
I go to where your customer actually works and follow one process end to end — the real one, exceptions and all. By day 30 something is working, and the account team has numbers they can quote.
A working agent covering that workflow, exposing every step it takes, plus a workflow map and baseline metrics.
Make it survive reality, then measure it
The demo becomes something that doesn't fall over when the input is malformed or a service goes down. Then we start scoring it honestly against real cases, and it goes live alongside the customer's team without touching anything.
Structured outputs, schema validation, checkpointing, retries, a named failure taxonomy. Then a golden dataset and rubrics drawn from real cases, a cost and latency budget, and shadow mode on live traffic.
Earn the first real responsibility, then repeat
The evidence from running alongside the team buys it the right to actually help. The customer's own engineers get everything they need to run it without me — then we start the next one, faster.
Shadow-mode agreement buys the move to assist. Runbook, monitoring and rollback triggers hand over to the customer. A second workflow starts on the existing scaffolding.
Fit
Why me specifically.
Full-stack across web and iOS. I've shipped production TypeScript and production Swift, which means I can follow a workflow into whatever system it actually lives in rather than stopping at the API boundary.
FalconryLab is a real iOS product, built by embedding in a niche domain and learning how practitioners actually work — the same motion this role runs, just self-directed. The agent above is the other half: shipped, measured, and still running.
The differentiator in this role isn't the code, it's being able to defend the system to an engineer and to a non-technical executive in the same afternoon. I've been writing for as long as I've been building.
What I build with
Background
Track record.
I started building for the web at thirteen, making custom MySpace profiles in hand-written HTML, and had moved on to whole websites by fourteen. That turned into a career in IT and software development, most of it in iOS, across a mix of employment and my own projects.
The through-line is that I ship and then keep running things. Not prototypes handed off — software with people depending on it.
A native iOS app for falconers — weights, feeding calculations, training logs, hunts. Built by embedding in a niche domain and learning how practitioners actually work, which is the same motion this role runs, just self-directed. Currently in beta.
A private outing club on Bear Lake in Michigan, with generations of member families. I built the site and I still run it — real users, real expectations, and nobody to escalate to when something breaks.
Code is on GitHub, and the agent documented on the proof page is running on this site right now.
The practical details
| Based | Michigan, US Eastern time |
|---|---|
| Available | Two to four weeks' notice |
| Location | Remote by preference, and I'll travel to customers as the work needs it. I would relocate for the right role. |
| Compensation | Negotiable, and I'll flex on cash for meaningful equity — particularly at the stage where one person can still change how the whole company deploys. |