Engineering and AI delivery terms

Staff augmentation

Adding engineers to a team you already run, under your manager and your process, billed by time. The engineer works your tickets and sits in your standup.

The comparison a buyer usually has in mind is an outsourced project, where a vendor owns the scope and hands back a deliverable. The difference is who the engineer reports to and who owns the scope.

Nearshore

Engineers in a country close enough that their working day overlaps yours. For a US buyer that usually means Latin America.

The point is overlap. A question asked at 10am gets answered before you leave, and a review and a handoff can happen within one business day.

Bench

Engineers a vendor employs who are not currently billing a client.

A big bench means a fast start. It also means somebody is paying for idle people, and that cost is in the rate you are quoted.

Time and materials

Billing by hours worked at an agreed rate, with no fixed total.

Fixed scope moves the delivery risk to the vendor and the price up. Time and materials keeps the price lower and leaves the scope risk with you. Which one you want depends on how well the work is defined.

Ramp

The time between an engineer starting and an engineer being useful.

It is a real cost that rarely appears in a quote, and it depends more on your documentation and how fast you can grant access than on the engineer.

System of record

The application a business treats as the truth for a domain. The ERP for inventory, the EHR for patient data, the reservation platform for bookings.

An agent that cannot write to the system of record cannot change how work gets done. It can only produce a recommendation that a person then types in by hand.

Pilot purgatory

An AI project that works in a demo and never ships.

The usual cause is structural. The demo runs on sample data, never touches the system of record, and was reviewed by the team that built it, so nobody with release authority will sign it.

Independent review

A check on a build done by someone who did not build it.

Self-certification is what stalls a pilot at the release gate. The person who signs is accepting the risk, and they know what their own team’s review is worth.

Human in the loop

A person who reviews an AI output before it takes effect.

It is the control a compliance team asks for first. It only holds if that person has the time to look and the standing to say no. A reviewer with three other jobs is a rubber stamp with a name on it.

Shadow AI

AI tools employees use without the company approving them. Usually a browser tab with company data pasted into it.

It shows up in an audit before it shows up in a budget, and by then the exposure is already months old.

Model drift

A model that performed well at launch and performs worse over time, because the data it sees has changed.

Nothing breaks and no alarm goes off. The outputs just get less right, and someone notices from a downstream complaint.

Legacy modernization

Changing a system the business depends on without stopping it.

The new code is usually the easy part. The undocumented behavior that other systems have quietly come to rely on is what breaks.

Technical debt

Work deferred to ship sooner, and the interest paid on it later in slower changes and more defects.

It is a real balance on a real ledger, and most teams cannot say what theirs is. The first useful step is naming the three places it hurts most.

Exception handling

What a process does when the case does not fit.

The software you bought covers the common path. The exception is where the spreadsheet appears and where a person becomes the integration.

Statement of work

The document that says what a vendor will deliver, by when, and what counts as done.

If the definition of done is vague, the argument happens at the end, when both sides have already spent the money.