Implementation and execution

AI implementation that ends in production

You have decided to put AI to work. The hard part is execution: a system in production, a number agreed before the build, and someone on your side who runs it after the engagement ends.

What implementation means here

Three things are true at the end of every engagement. If one is missing, the work is not done.

A system in production

Running code in your repository or on infrastructure you control, on the path your people or your users already take. A demo is not the deliverable. Neither is a deck.

A baseline before the build

Volume, cycle time, and cost for a workflow, or the eval that defines good enough for a feature. We agree the number before anything is built, so the result is measured against it.

An owner, and an end date

A named person on your side is trained on the system. The engagement is a fixed milestone and is built to end. What happens next is your call, not a renewal.

Where your implementation starts

Implementation takes one of four shapes, and each is an engagement we already run. Pick the sentence closest to yours.

The AI belongs inside a workflow you run

Finance, operations, support, or the back office, on the ERP or CRM you already have. First measurable financial impact is structured for 4–8 weeks, a methodology target, set against your baseline.

Business Process AI

The AI belongs inside the product you sell

A feature on the path your users hit, the eval that decides when it stays on, and engineers on your team who can extend it. First value is structured for 4 weeks. One feature in a live product is a job for a Product AI Engineer.

AI Pivot Squad

The data cannot leave a boundary

Prompts, documents, outputs, and logs stay in the EU, on hardware you control or with an EU operator. The same deployment covers a UAE boundary and US healthcare records.

Sovereign AI

Something is built, and it does not hold

A repo that coding agents wrote faster than anyone reviewed starts with a $10,000 audit: a scored keep, repair, or delete report in 7 days. The repair is a separate quote. A proof of concept that now has users starts with a Harness Engineer.

AI Code Cleanup

Need one person, not a team? Six named seats are on the services page. Rebuilding a product you already run starts with the AI Architect, on a sequence of milestones.

How an implementation runs

Scoping call

30 minutes. You say what you showed up with. We name the entry, or tell you none of ours fits.

Access and a baseline

We need the process, the repo, or the data that sets the baseline. Without that access there is no engagement.

Build on a fixed milestone

The people who design the system write it, and a senior practitioner leads every engagement. Average engineering depth is 15 years. There is no junior bench.

Hand over, then leave

The code lands in your repository with documentation and a trained owner. Thirty days of hypercare after go-live is not a retainer.

The method, phase by phase, is on Our Approach.

What we do not implement

  • A chatbot demo with no workflow or product behind it.
  • A strategy narrative with no production system at the end.
  • An open-ended retainer, or a bench of developers billed by headcount.
  • A new platform before the path your users already take is measured.
  • Any engagement where we cannot see the process, the repo, or the data needed to set a baseline.

Questions about AI implementation

A system running in production, the baseline it is measured against, documentation in your repository, and a named owner on your side who has been trained on it. Then the engagement ends.

It depends on the shape. A first product feature is structured for 4 weeks. First measurable financial impact from a workflow is structured for 4–8 weeks, a target set against your baseline, not a promise. A full product rebuild is a sequence of milestones, and four weeks covers only its first decision and first replaced path.

No. A workflow engagement builds on the ERP, CRM, or system of record you have today. A product engagement works in your codebase. Replacing a system is a rebuild, and a rebuild starts with an AI Architect deciding what stays.

Whichever your task, cost ceiling, and data boundary allow: OpenAI, Anthropic, or open-weights models on hardware you control. The model is a component. The baseline and the eval decide whether it stays.

A 30-minute call, then access to the process, the repo, or the data needed to set a baseline, and one person who will own the system afterwards. Without access there is no engagement.

Each engagement is a fixed milestone, quoted after the scoping call. We publish two prices. The AI Code Cleanup audit is $10,000 for a scored report in 7 days. Priced AI Features costs $10,000 to set up, then $2,000 per AI feature. We do not bill by headcount or on retainer.

Not sure which shape fits?

A 30-minute scoping call is usually enough to tell. If none of our engagements fits, we say so on the call.