# AI engineering — the discipline under the four services AI engineering is designing, building, testing, and running AI software that does real work in your business every day. At Helixfield that work is the Field Survey, then a helper install or a shared-facts system, then teaching, then monthly care. We connect it to the data and systems you already pay for, check its answers against real examples before anyone relies on it, and roll it out in stages with a person who can stop it. We lead with the newest tools that fit the job — OpenAI, Grok, Claude, Gemini, Cursor, or what ships next. Loop & Graph is how we organize the work today — not a promise for forever. ## What it covers ### AI systems that run every day Software that does a job your team already does, on a schedule or on a trigger, in your systems. Built to be restarted, watched, and handed over — not a prototype that lives on one laptop. ### Quality checks before anyone relies on it We write down what a good answer looks like, then score the system against ten to twenty real examples from your work. We keep scoring after launch so drift shows up as a number, not a complaint. ### Your own records, when the job needs them Some jobs need the system to look things up — contracts, price lists, past tickets, policies. We connect it to the right source, keep it read-only until trusted, and make sure it says it does not know instead of guessing. ### Several steps in the right order A job with several steps — read, look up, draft, check, file — becomes a pipeline with a clear order. Each step can be tested on its own, and a person steps in at the point that matters. ### Safe rollout and day-to-day running Read access first. Then drafts a person approves. Then, if earned, the system acts on its own for low-risk steps. A log, a cost line, and an off switch are part of the build. ### Built into what you already pay for Microsoft 365, Salesforce, NetSuite, your ERP, your ticketing system. We build in, not beside. Permissions are written down. AI seats and API bills stay in your name. ## How a build runs 1. **Name the job and the measure.** One sentence for the work that repeats. One number that tells you it worked. If either is missing, we start with a Field Survey. 2. **Look at the data and the systems.** What the system must read, what it may write, and who owns each. We ask for read access only at this stage. 3. **Design on paper first.** The steps, the hand-offs, and where a person checks. We pick the AI tools from the job — the newest that fit, not last year’s stack. 4. **Build in stages, score each one.** Read-only first. Each stage is run against real examples from your work and scored before the next stage starts. Invented numbers fail the test. 5. **Roll out with a switch.** A named owner, a log you can read, a monthly cost line, and a way to turn it off. Nothing leaves the company until a person says send. 6. **Keep it current.** Models and tools change. Monthly care re-scores the system when they do, and moves it when a better tool ships. ## What you leave with - Working software in your systems, with the code, settings, and permissions in your name. - A written set of quality checks and the scores at launch. - A runbook: how to restart it, how to read the log, how to switch it off. - A monthly cost line from the AI vendors, in plain numbers. - Thirty days of fixes after launch, then monthly care if you want it. ## Pricing Engineering is scoped in the Field Survey ($9,500) and priced as one of the published setups: $24,000 to $48,000. When a build is bigger than the rate card, we write a fixed quote before we start. AI seats and API usage stay on your bill. If a buyer insists on hourly, the fallback is $285 per hour. ## Questions ### What does AI engineering mean at Helixfield? Designing, building, testing, and running AI software that does a real job in your business every day. That work is the Field Survey, then a helper install or a shared-facts system, then teaching, then monthly care. It is a plain-English name for careful software work, not a fifth buy-path card. ### How is AI engineering different from setting up a helper? A helper setup configures a current AI platform on one job with the facts it must follow. More engineering is the right buy when the system must look things up in your own records, run several steps in a fixed order, prove its quality before go-live, or connect to systems with no ready-made connector. Same four-step path either way. ### What is an eval, in plain English? A quality check. We take ten to twenty real examples from your work, write down what a good result looks like for each, and score the system against them. Invented numbers fail. We run the check before launch and keep running it after. ### Do you train custom AI models? No. We do not train models from scratch. We build systems around the best current models from OpenAI, xAI, Anthropic, Google, and their peers, and we move when a stronger one ships. ### How much does AI engineering cost? Scoped in a $9,500 Field Survey and priced as one of the published setups: $24,000, $36,000, or $48,000. When a build is bigger than the rate card, we write a fixed quote before we start. If a buyer insists on hourly, the fallback is $285 per hour.