What you're hiring for.

The forward-deployed loop is the job I've done for years under a different name: embed with a team, learn their pipeline, build the thing that replaces the slow part, teach them to run it. I spent 8 years embedded in live newsrooms, where the workflow was the product. My broll-pipeline shows the whole loop: a 7-stage pipeline on the ElevenLabs stack, public on GitHub, with a costed 69-second live demo covering narration, score, sound, generated shots, and a Spanish dub at $3.32 all-in. The prototype isn't a deck. It ships with receipts.

Production agents fail on the parts nobody specced, so I build the spec into the system. My tax-verification agent is citation-gated: no claim without a source, and it caught a ~$19K error. Monolith enforces deterministic constraint gates and returns INSUFFICIENT_DATA instead of guessing. Career-ops runs about 50 scheduled agents under multi-model orchestration. The skill underneath is a producer's: turning messy human requirements into boundaries a machine can actually hold.

Adoption isn't a mandate problem. It's a legibility and trust problem, and that's comms work. At Google xGE I ran communications for 1,000+ Principal, Distinguished, and Fellow engineers, after 6 years reaching 75,000+ employees at CorpEng. Then I built the tools myself: two production AI agents shipped solo. One was a comms-triage agent I handed off with a full operator runbook. The other was an executive digital-twin writing system that cut VP review back-and-forth by about 70%, an internal figure I measured myself.

A live rundown is a program plan with zero slack: fixed air time, moving inputs, a team that needs to know exactly what happens next. I ran that discipline through sixteen years of newsroom and Google deadlines. Now I apply it to machines. I operate a production fleet of about 50 scheduled agents on launchd, each with a budget, a schedule, and a dead-man heartbeat that reports when a job goes quiet. The fleet is the proof: autonomous work made observable and accountable, running every day.

AI-native companies need communicators who actually understand what the models do. Your audience can tell when they don't. I wrote executive communications for Google's most senior engineers, then built the system that scaled it: a digital-twin writing tool that cut VP review back-and-forth by about 70%. Before Google, I ran a team at AJ+ during its 500M-weekly-view era. The habit that runs through all of it: every claim gets a source. This site practices what I'm pitching. The Impact page is the proof.

Developer education is explanation under constraint: limited attention, a skeptical audience, zero hand-waving. Sixteen years doing this, in newsrooms and at Google, making complex systems land on camera. I produced Google's public Life at Google recruitment and interview-prep videos. Before that, I engineered the explainer format that hit 50M views. I also write an operator runbook for everything I ship. A tool nobody else can run is just a demo.

Editorial at an AI company is a systems job now, and I build the systems, not just the edits. Voice-os encodes a writing voice as six scored axes plus a banned-phrase list, so quality holds without me in the loop. I've applied editorial judgment under real pressure, including litigation-sensitive filters run live. And I develop people: I coached AJ+ producers into on-camera careers that went on to win a Webby and a Daytime Emmy. The proof is voice-os: taste, made measurable.