How Does Custom AI Development Work?

That is the process, in full, as we run it. There is no eighth step, no hidden phase for "requirements gathering" that stretches into quarter two. Here is what happens at each stage and where the model differs from a fixed-price agency engagement.

The seven steps in order

1. Intake. A call or a message describing the problem. No qualifying gauntlet, no multi-week vendor selection process on our end.

2. Discovery session. One working session covering the business, the constraints, and the actual requirements: what the system needs to do, what data it touches, and what "done" looks like for the first release.

3. Free prototype. We build a working, clickable prototype before any money changes hands. You click through it and either approve it or send it back with changes. You are committed to nothing at this stage. If you walk away, you keep the prototype.

4. Pod deployment. A pod, sized to the project, starts working inside your own codebase and repository. Per our pods page, a Builder Pod is a lead plus a two-engineer bench; a Growth Pod adds a third engineer and a second concurrent build track. The pod typically starts within five business days of approval, with first shipped work landing in week one or two.

5. Daily standups. In your existing channel: Slack, Teams, email, whatever you already use. No new tool to adopt.

6. Weekly shipping. Code and demos ship every week, and you reset priorities weekly based on what you see. This is the mechanism that replaces a fixed scope document: the roadmap adjusts as fast as the product does.

7. Handover. Full transfer of the repository, migrations, deploy pipeline, and documentation into your own accounts. Nothing in the finished system depends on continued access to us.

How billing and cancellation actually work

This is the biggest structural difference from a traditional agency or a fixed-bid consultancy. There is no per-hour billing and no statement of work for ongoing work. Capacity is billed monthly at a fixed rate instead of a fixed-price contract with change orders every time scope moves. Per pricing, that is $5,000/month for a Builder Pod (one build track), $10,000/month for a Growth Pod (two build tracks), or custom pricing for an Enterprise Pod (three or more tracks, 3-8 engineers).

Every tier is month-to-month with a 30-day cancellation notice sent by email. A paused month is not billed, and the pod seat is held. There is no change-order process because weekly reprioritization already absorbs scope shifts. If your roadmap changes on a Tuesday, it shows up in the following week's ship, not in a renegotiated statement of work.

This matters most for regulated or data-heavy builds, where requirements genuinely do move as compliance review surfaces new constraints. A fixed-price contract punishes that discovery. Monthly capacity billing absorbs it.

Who is actually on the pod, and what gates the code

A pod is a lead plus senior engineers and QA, sized 2 to 5 people depending on the tier. Every engineer works inside your own repository from day one; nothing sits in a shared internal codebase we control.

AI-assisted code goes through the same gate as any other code we write: a pull request in your repository, reviewed by the named engineer who owns it, typed contracts, and tests running in CI. Schema changes go through reviewed migrations, not ad hoc scripts run against a production database. This is not a special "AI code" lane with lighter review; it is the same review every commit gets.

Two case studies show the pattern under real constraints. On a compounding-pharmacy platform, each phase was checked against numbered requirements by an adversarial verifier before it merged, and the build shipped eleven epics behind that gate with 490+ unit tests. On a dental EHR project, every requirement in the brief was tracked against a traceability matrix so each line mapped to shipped, tested code. That discipline is the default whenever a build touches regulated data or clinical workflow, not a one-off for those two clients.

What this looks like for a regulated build

For healthcare, fintech, or public-sector work, three things need to be true before code touches production data, and all three are structural, not promised in a sales deck.

First, a SOC 2 Type II report is available under NDA on request, covering the security controls the engineering process runs under. Second, we sign Business Associate Agreements on request and operate HIPAA-aligned controls; there is no such thing as "HIPAA certified" because HIPAA is a regulation with no certification to hold. Third, deployment happens inside your own cloud account or VPC from week one, not a shared multi-tenant environment we control.

Two shipped builds carry this posture directly: a compounding-pharmacy portal with a seven-year immutable audit log and patient-facing screens passing WCAG 2.1 AA, and a public-sector spend auditor built to run fully offline, with zero external API calls, so sensitive financial data never leaves the premises. Full detail on the security posture, including what the SOC 2 report covers, is on the security page.

Who owns what when the pod leaves

You own all code, data, models, and IP, with no license-back. If the engagement ends, whether by choice, by cancellation, or because the build is finished, nothing in the running system depends on continued access to us. Handover includes the repository, migrations, deploy pipeline, and documentation, matching the ownership terms described on the how it works page.

The short version

Custom AI development, run this way, is seven fixed steps: intake, discovery, a free prototype, a pod deployed inside your own repository within about five business days, daily standups, weekly shipping, and full handover. Billing is monthly capacity, not a fixed-price contract with change orders, and cancellation takes 30 days' notice with no penalty. Small builds take one to three months, medium builds three to twelve, and every commit, AI-assisted or not, goes through the same PR review, typed contracts, and CI tests before it merges into code you own outright.