▲ /about
daniel kinney-spears
anduril · aws alum · operator-shaped
— what i'm doing here —
loworbit is a workshop in public. each prototype starts as a one-page prd, gets built end-to-end by an ai agent, ships, and then i write a verdict. the prompts are visible. the code is visible. the failures are visible. the point isn't the prototypes — it's the loop, and what gets kept, cut, and scrapped between the deploy and the next prd.
the career shape this maps to is solutions-architect-meets- operator: somebody who can spec, ship, judge, and write up what they learned in the same week, working alongside agents that do most of the typing. that's the bet. this site is the artifact of betting on it.
— prior work, in order of weight —
ai capture + proposals toolset @ aws
10,000+ docs processed. est. $2–4m in direct labor savings.
led the design and rollout of the internal toolset capture and proposal teams used to draft, qualify, and refine pursuit work. the throughput number is real and was tracked monthly; the labor savings number is conservative because it counts displaced first-draft hours only, not the secondary effects on win rate.
content management overhaul @ aws
supports $10b/yr in business. ~90% reduction in maintenance load.
rebuilt the content stack the field uses for proposal-grade source material. the bulk of the work was structural — turning a sprawl of overlapping docs into a small set of canonical, versioned sources — so downstream toolchains had a stable thing to point at.
global genai training @ aws
200+ people trained across sales, capture, and proposal teams.
designed and delivered the curriculum that taught the field organization how to actually use the genai toolchain in the work, not just be aware of it. the test was whether trainees changed how they wrote pursuits. they did.
genai challenge @ aws
60+ micro-apps produced. best of them shipped as a curated toolset.
organized and judged a company-wide build sprint. the curation pass was the actually interesting part — separating the apps that solved a real recurring problem from the ones that were demos of a model trick. roughly half the same skill the loworbit verdict step is reaching for.
design + layout team @ aws
launched a new team for custom, customer-focused proposal design. handed off, still running.
started a function that didn't exist, hired into it, set the standards, and stepped out. the hand-off held — the team kept producing without me, which is the only test that matters for that kind of work.
— what i'm not —
not a research scientist. not someone who'll write you an eval paper. the work i do well is at the seam between what a model can do and what a real organization will actually adopt — speccing the integration, judging the output, training the people, killing the parts that don't earn their keep. loworbit is that seam, made into a public artifact instead of an internal one.