This summer I built Industrial Safety IQ mostly on my own, with AI tools open all day. Here's how I estimated the hours, how that compares with a normal team, and where the tools helped and where they didn't.
~480
Engineering hours with AI
~6,000
Traditional team hours (mid)
10 wks
Calendar vs ~6 months
People ask me how long a platform like Industrial Safety IQ takes to build. This one took ten weeks on the calendar. I did the engineering myself and used AI as a working partner every day.
I didn't keep a timesheet, so the hour figure is an estimate. It comes from lining up the git history, what the product does, and a conservative guess at how a normal team would staff it.
How I counted
I built the production system between July 8 and September 15, 2026. The repository has 288 commits, all mine, and roughly 197,000 lines of TypeScript and TSX. The live product runs on federated GraphQL, AWS Lambdas, Postgres, and Firebase authentication, with Stripe for billing. On top of that sit Lockout/Tagout authoring, training, and computer vision, plus a web application and a thin mobile shell.
My hours come from that calendar and the commit cadence. The busiest weeks had 50 to 63 commits, which is well past a normal 40-hour week for one person. Ten weeks at 40 to 60 hours a week works out to 400 to 600 hours. I use the midpoint, 480, or about 48 hours a week.
The AI work left a record as well: dozens of main chat sessions and more than two hundred agent transcripts. The assistants wrote a lot of first drafts. I decided what actually shipped.
Source: tracked TypeScript and TSX in the monorepo. SQL, YAML, GraphQL schema, and shell scripts are extra.
What a traditional team would look like
Built the usual way, a platform this size needs a real team. I'd staff it with a tech lead, two backend engineers, and two web engineers, plus part-time help on mobile, DevOps, and QA. Call it six to eight people for five to seven months.
The hour math is simple. Five people for five months is 4,000 hours, and that's too tight once billing, vision, and federation are in scope. Six people for six months comes to about 5,760. Add dedicated QA, design, and a project manager and you're near 9,000 hours over seven months.
I use about 6,000 hours as the midpoint. Next to my estimate, that's roughly 12 times the person-hours, and about six months on the calendar instead of ten weeks.
What AI actually changed
The biggest savings came from boilerplate and coordination. Scaffolding subgraphs, packaging Lambdas, building module screens, and writing GraphQL resolvers used to mean days of copy-paste. A feature that would normally be a ticket, a pull request, and a wait for review usually got done the same day. I'd describe the change, read the diff, and then try it in the app.
Production problems got shorter too. When the auth subgraph started returning 500s, and later when a schema library mismatch took GraphQL down, I found the cause and had a fix deployed in one sitting. On a bigger team that kind of outage usually pulls in whoever is on call plus a frontend and a backend engineer, and it can eat half a day.
There were hardly any standups or review queues, and very little time lost to environments drifting apart. I owned the whole repository, so there was nobody to sync with.
What it didn't replace
Line count overstates how much of this is new work. UI panels and resolvers repeat a lot. A traditional team would also spread tests more evenly. There are fifty-four test files here, which is light for a product this big.
AI doesn't know what a lockout procedure has to say. It doesn't hold the AWS account, the GraphQL graph, Stripe, or Firebase. I handled all of that myself, and it's counted in my hours. Code review and production hardening are thinner than a six-person team would manage. That's why a subgraph library upgrade and a training control both still needed me to step in.
Industrial software still needs someone accountable reading what the tools produce, and I don't want to suggest otherwise.
If you're planning a build like this
If you're weighing a build like this, start by asking whether you have a senior engineer who can direct the tools, own production, and stay close to the domain. Without that person, AI mostly helps you ship fragile software sooner. Taking those builds the rest of the way to production is one of the services we offer at Carolina Software Engineering.
I built Industrial Safety IQ this way on purpose. The case study covers the product itself: authoring, review, posted floor docs, training, and audit. If you'd like a similar engagement, tell us what you're trying to ship.
