How we built this service
AI did not merely help build this website. It became an engineering and planning partner across Road2Work—from auction research and bid ceilings to rental pricing, fleet workflows, customer records, maintenance decisions, and deployable software. Morgan remains the person making the decisions and accepting the consequences.
01
The software follows the service—not the other way around. Morgan specified customer protections, application requirements, driving-record authorization, fleet statuses, rental pricing, Pay It Forward, maintenance handling, accidents, towing, turnover, extensions, notifications, and signatures.
Those decisions were broken into small, reviewable stages. Each stage could be challenged, corrected, tested, and approved before deployment.
02
AI assisted with planning workflows, drafting policy language, identifying edge cases, translating approved requirements into React and TypeScript, writing tests, checking builds, and preparing deployments.
Morgan remained the human in the loop: setting policy, rejecting bad assumptions, choosing language, and approving releases. AI can move quickly; it can also be confidently wrong. Validation is not optional.
03
At auction, AI helps organize listings, compare candidate vehicles, calculate buyer fees and out-the-door acquisition cost, and identify the questions that matter before bidding. High mileage and ugly paint may be acceptable; drivetrain, cooling-system, timing-belt, and underside condition can kill the deal.
The same analysis develops a stop-bid ceiling for each lot. That ceiling can account for acquisition cost, auction fees, transport, immediate repairs, operating assumptions, and the price Road2Work can fairly charge. It is decision support, not automatic bidding. Morgan inspects, judges risk, and decides whether to raise the paddle.
04
AI helped model the relationship between acquisition cost, insurance, registration, maintenance, and Pay It Forward. The result is not a mystery number and it is not a pile of bullshit fees: each vehicle has an explicit weekly cost structure and a rate that can be tested against the mission.
Pricing scenarios also show what happens when the purchase price changes, how long fleet reinvestment takes, and when the model can support another vehicle. Morgan chooses the policy and the final numbers.
05
A winning bid is only the beginning. The operating model tracks acquisition, inspection, repairs, readiness, public availability, customer assignment, agreements, billing, maintenance, accidents, towing, extensions, return, condition photos, turnover, and the decision to place the vehicle back in service.
AI helps structure checklists, surface missing information, reason through failure modes, draft customer and staff communications, and turn events into trackable work. It does not certify a repair, inspect a vehicle, approve a driver, or replace accountable human review.
06
The current interface connects structured applications, explicit MVR authorization, customer and vehicle records, staff pricing, customer-specific rate adjustments, and nine operational queues covering the rental lifecycle.
The verified stack uses React 19 and Next.js 16, compiled through Vinext and Vite with TypeScript. The deployed site runs as a Cloudflare Worker through ChatGPT Sites. Builds, lint checks, and Node-based tests run before releases are pushed.
07
The deployed prototype is functional, responsive, and device-persistent. Its records currently live in browser storage. A shared production database is not connected. Protected document storage is not connected. Production customer and staff authentication, payment processing, legal document retention, auction-data ingestion, and external service integrations still require implementation and security review.
AI-assisted planning is already part of the work. Automated authority is not. Installing a library or describing a workflow is not the same as shipping a secure production capability, and we do not pretend otherwise.
08
Every approved change is implemented against the existing source, built, linted, and tested before deployment. The same discipline applies outside the code: verify the listing, inspect the vehicle, respect the bid stop, document the repair, and do not confuse a plausible answer with a proven one.
AI makes iteration faster and expands what a small organization can analyze. It does not make security, mechanical safety, legal compliance, financial discipline, or operational judgment somebody else’s problem.
The result